ai-agent-book 精选快照（<2MB 代码与文档，来自 github.com/bojieli/ai-agent-book）
Build latest book artifacts / build (push) Canceled after 0s
dependency resolution / resolve (3.11) (push) Canceled after 0s
dependency resolution / resolve (3.13) (push) Canceled after 0s
deploy-pages / build (push) Canceled after 0s
deploy-pages / deploy (push) Canceled after 0s
i18n consistency check / check (push) Canceled after 0s
provider adoption tests / test (chapter2/context-compression) (push) Canceled after 0s
provider adoption tests / test (chapter2/prompt-injection) (push) Canceled after 0s
provider adoption tests / test (chapter2/system-hint) (push) Canceled after 0s
provider adoption tests / test (chapter3/log-sanitization) (push) Canceled after 0s
web-search-agent tests / test (push) Canceled after 0s
web-search-agent tests / agentbook (push) Canceled after 0s
Build latest book artifacts / build (push) Canceled after 0s
dependency resolution / resolve (3.11) (push) Canceled after 0s
dependency resolution / resolve (3.13) (push) Canceled after 0s
deploy-pages / build (push) Canceled after 0s
deploy-pages / deploy (push) Canceled after 0s
i18n consistency check / check (push) Canceled after 0s
provider adoption tests / test (chapter2/context-compression) (push) Canceled after 0s
provider adoption tests / test (chapter2/prompt-injection) (push) Canceled after 0s
provider adoption tests / test (chapter2/system-hint) (push) Canceled after 0s
provider adoption tests / test (chapter3/log-sanitization) (push) Canceled after 0s
web-search-agent tests / test (push) Canceled after 0s
web-search-agent tests / agentbook (push) Canceled after 0s
This commit is contained in:
@@ -0,0 +1,122 @@
|
||||
# Execution Tools MCP Server Dockerfile
|
||||
# Supports multi-language code execution with scientific computing packages
|
||||
# Uses latest stable versions of all tools (as of 2025)
|
||||
|
||||
FROM ubuntu:24.04
|
||||
|
||||
ENV DEBIAN_FRONTEND=noninteractive
|
||||
ENV PYTHONUNBUFFERED=1
|
||||
ENV TZ=UTC
|
||||
|
||||
# Install system dependencies
|
||||
RUN apt-get update && apt-get install -y \
|
||||
curl \
|
||||
wget \
|
||||
git \
|
||||
build-essential \
|
||||
software-properties-common \
|
||||
ca-certificates \
|
||||
gnupg \
|
||||
lsb-release \
|
||||
&& rm -rf /var/lib/apt/lists/*
|
||||
|
||||
# Install Python 3.13 (latest stable)
|
||||
RUN add-apt-repository ppa:deadsnakes/ppa && \
|
||||
apt-get update && \
|
||||
apt-get install -y \
|
||||
python3.13 \
|
||||
python3.13-dev \
|
||||
python3.13-venv \
|
||||
python3-pip \
|
||||
&& update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.13 1 && \
|
||||
update-alternatives --install /usr/bin/python python /usr/bin/python3.13 1 && \
|
||||
rm -rf /var/lib/apt/lists/*
|
||||
|
||||
# Install Node.js 22.x LTS for JavaScript/TypeScript support
|
||||
RUN curl -fsSL https://deb.nodesource.com/setup_22.x | bash - && \
|
||||
apt-get install -y nodejs && \
|
||||
npm install -g npm@latest && \
|
||||
npm install -g typescript@latest tsx@latest ts-node@latest
|
||||
|
||||
# Install common Node.js packages globally
|
||||
RUN npm install -g \
|
||||
lodash \
|
||||
axios \
|
||||
chalk \
|
||||
commander \
|
||||
express \
|
||||
@types/node \
|
||||
@types/express
|
||||
|
||||
# Install Go 1.22 (latest stable)
|
||||
RUN wget https://go.dev/dl/go1.22.10.linux-amd64.tar.gz && \
|
||||
tar -C /usr/local -xzf go1.22.10.linux-amd64.tar.gz && \
|
||||
rm go1.22.10.linux-amd64.tar.gz
|
||||
ENV PATH="/usr/local/go/bin:/root/go/bin:${PATH}"
|
||||
ENV GOPATH="/root/go"
|
||||
|
||||
# Pre-download common Go modules (will be cached in module cache)
|
||||
RUN go install github.com/go-delve/delve/cmd/dlv@latest
|
||||
|
||||
# Install Java 21 LTS (latest LTS)
|
||||
RUN apt-get update && \
|
||||
apt-get install -y openjdk-21-jdk && \
|
||||
rm -rf /var/lib/apt/lists/*
|
||||
ENV JAVA_HOME=/usr/lib/jvm/java-21-openjdk-amd64
|
||||
|
||||
# Install C++ compiler (GCC 13) with common libraries
|
||||
RUN apt-get update && apt-get install -y \
|
||||
g++-13 \
|
||||
gcc-13 \
|
||||
libboost-all-dev \
|
||||
libssl-dev \
|
||||
libcrypto++-dev \
|
||||
cmake \
|
||||
&& update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-13 100 \
|
||||
&& update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-13 100 \
|
||||
&& rm -rf /var/lib/apt/lists/*
|
||||
|
||||
# Install Rust (latest stable)
|
||||
RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y --default-toolchain stable
|
||||
ENV PATH="/root/.cargo/bin:${PATH}"
|
||||
|
||||
# Install PHP 8.3 with common extensions
|
||||
RUN apt-get update && apt-get install -y \
|
||||
php8.3-cli \
|
||||
php8.3-curl \
|
||||
php8.3-mbstring \
|
||||
php8.3-xml \
|
||||
php8.3-zip \
|
||||
&& rm -rf /var/lib/apt/lists/*
|
||||
|
||||
# Install Bash (already in base image, but ensure latest)
|
||||
RUN apt-get update && apt-get install -y bash && \
|
||||
rm -rf /var/lib/apt/lists/*
|
||||
|
||||
# Set up working directory
|
||||
WORKDIR /app
|
||||
|
||||
# Copy requirements and install Python packages
|
||||
COPY requirements.txt .
|
||||
RUN python3 -m pip install --upgrade pip setuptools wheel && \
|
||||
python3 -m pip install -r requirements.txt
|
||||
|
||||
# Copy application files
|
||||
COPY . .
|
||||
|
||||
# Expose default MCP stdio interface (no port needed for stdio)
|
||||
# The server communicates via stdin/stdout
|
||||
|
||||
# Set environment variables
|
||||
ENV WORKSPACE_DIR=/workspace
|
||||
RUN mkdir -p /workspace && chmod 777 /workspace
|
||||
|
||||
# Create non-root user for security
|
||||
RUN useradd -m -u 1000 mcpuser && \
|
||||
chown -R mcpuser:mcpuser /app /workspace
|
||||
|
||||
# Switch to non-root user
|
||||
USER mcpuser
|
||||
|
||||
# Run the MCP server
|
||||
CMD ["python3", "server.py"]
|
||||
@@ -0,0 +1,203 @@
|
||||
# Experiment 4.3: Execution Tools MCP Server
|
||||
|
||||
## Objective
|
||||
|
||||
Implement a comprehensive MCP server that provides execution tools with built-in safety mechanisms, demonstrating real-world best practices for AI agent tool execution.
|
||||
|
||||
## Experiment Overview
|
||||
|
||||
This experiment explores three critical aspects of execution tools:
|
||||
|
||||
1. **Safety Mechanisms**: LLM-based approval for dangerous operations
|
||||
2. **Result Processing**: Automatic summarization of complex outputs
|
||||
3. **Verification**: Automatic validation of tool execution results
|
||||
|
||||
## Architecture
|
||||
|
||||
### Safety Layer
|
||||
|
||||
The safety layer implements a multi-level protection system:
|
||||
|
||||
**LLM-Based Approval**: Before executing irreversible operations (file overwrite, system commands, external API calls), the system consults a secondary LLM to evaluate the risk. The approval process analyzes the operation for potential data loss, security risks, and resource consumption concerns. This mirrors real-world approval workflows where critical operations require managerial sign-off or risk control review.
|
||||
|
||||
**Result Summarization**: When execution tools (code interpreter or virtual terminal) produce output exceeding 10,000 characters, the system automatically invokes an LLM to distill the essential information. Outputs under this threshold are returned as-is to preserve full detail for smaller results. This summarization focuses on key results, errors, warnings, and actionable insights, enabling the primary agent to process information more efficiently without being overwhelmed by raw data.
|
||||
|
||||
**Automatic Verification**: Operations that produce verifiable outputs undergo automated validation. Code files are checked for syntax errors, terminal commands are evaluated for successful execution, and API responses are validated against expected schemas. Verification results feed back into the agent's context, allowing it to self-correct without manual intervention.
|
||||
|
||||
### Tool Implementation
|
||||
|
||||
#### File System Tools
|
||||
|
||||
The file system tools provide safe, verified file operations. The write operation supports automatic syntax checking for code files in Python, JavaScript, and TypeScript, preventing the creation of invalid source files. The edit operation generates diff previews before applying changes, allowing the agent to understand the impact of modifications. Both operations enforce workspace boundaries, preventing accidental file access outside designated directories.
|
||||
|
||||
#### Generic Execution Tools
|
||||
|
||||
The code interpreter executes Python code in a controlled environment with namespace restrictions. It captures both standard output and error streams, detects dangerous function calls like system commands or eval statements, and provides detailed error analysis when execution fails. The virtual terminal executes shell commands with configurable timeouts, monitors for destructive operations, and automatically summarizes verbose output to highlight relevant information.
|
||||
|
||||
#### External Integration Tools
|
||||
|
||||
The Google Calendar integration adds events with validation of datetime formats and logical consistency checks. The GitHub integration creates pull requests with branch verification and approval workflows. Both tools demonstrate patterns for safely interacting with external systems while maintaining visibility and control.
|
||||
|
||||
## Setup
|
||||
|
||||
### Prerequisites
|
||||
|
||||
```bash
|
||||
# Create virtual environment
|
||||
python -m venv venv
|
||||
source venv/bin/activate # On Windows: venv\Scripts\activate
|
||||
|
||||
# Install dependencies
|
||||
pip install -r requirements.txt
|
||||
```
|
||||
|
||||
### Configuration
|
||||
|
||||
1. Copy environment template:
|
||||
```bash
|
||||
cp env.example .env
|
||||
```
|
||||
|
||||
2. Configure your LLM provider:
|
||||
|
||||
```bash
|
||||
PROVIDER=kimi
|
||||
KIMI_API_KEY=your-key
|
||||
```
|
||||
|
||||
**Supported Providers:**
|
||||
- **SiliconFlow**: `SILICONFLOW_API_KEY` - Uses Qwen/Qwen3-235B-A22B-Thinking-2507
|
||||
- **DashScope / Bailian (Qwen)**: `DASHSCOPE_API_KEY` - Uses `qwen3.7-plus`; select with `PROVIDER=dashscope` (or `qwen`/`bailian`)
|
||||
- **Doubao**: `DOUBAO_API_KEY` - Uses doubao-seed-1-6-thinking-250715
|
||||
- **Kimi/Moonshot**: `KIMI_API_KEY` - Uses kimi-k3 (default)
|
||||
- **OpenRouter**: `OPENROUTER_API_KEY` - Uses google/gemini-3.5-flash
|
||||
|
||||
3. (Optional) Configure external services:
|
||||
```bash
|
||||
# Google Calendar
|
||||
GOOGLE_CALENDAR_CREDENTIALS_FILE=credentials.json
|
||||
|
||||
# GitHub
|
||||
GITHUB_TOKEN=your-github-token
|
||||
```
|
||||
|
||||
### Safety Settings
|
||||
|
||||
```bash
|
||||
# Enable/disable safety features
|
||||
REQUIRE_APPROVAL_FOR_DANGEROUS_OPS=true
|
||||
AUTO_SUMMARIZE_COMPLEX_OUTPUT=true
|
||||
AUTO_VERIFY_CODE=true
|
||||
MAX_OUTPUT_LENGTH=1000
|
||||
```
|
||||
|
||||
## Running the Experiment
|
||||
|
||||
### Quick Start
|
||||
|
||||
```bash
|
||||
python quickstart.py
|
||||
```
|
||||
|
||||
This demonstrates all major features with minimal setup.
|
||||
|
||||
### Individual Tool Tests
|
||||
|
||||
```bash
|
||||
# Test file operations
|
||||
python test_file_tools.py
|
||||
|
||||
# Test code execution
|
||||
python test_execution_tools.py
|
||||
|
||||
# Test external integrations (requires credentials)
|
||||
python test_external_tools.py
|
||||
```
|
||||
|
||||
### Comprehensive Examples
|
||||
|
||||
```bash
|
||||
python examples.py
|
||||
```
|
||||
|
||||
### Running as MCP Server
|
||||
|
||||
```bash
|
||||
python server.py
|
||||
```
|
||||
|
||||
The server will start in stdio mode, ready to accept MCP protocol connections.
|
||||
|
||||
## Experiment Results
|
||||
|
||||
### Safety Mechanism Evaluation
|
||||
|
||||
Test the approval system by attempting dangerous operations:
|
||||
|
||||
1. File overwrite of important files
|
||||
2. Terminal commands with destructive patterns
|
||||
3. Code execution with system calls
|
||||
|
||||
Observe how the LLM evaluates risk and makes approval decisions.
|
||||
|
||||
### Summarization Effectiveness
|
||||
|
||||
Generate complex outputs and measure summarization quality:
|
||||
|
||||
1. Execute commands that produce verbose output (>10,000 characters)
|
||||
2. Run code that generates extensive logs
|
||||
3. Verify that outputs under 10,000 characters are returned unchanged
|
||||
4. Compare original vs. summarized information density for large outputs
|
||||
|
||||
### Verification Accuracy
|
||||
|
||||
Test automatic verification across different scenarios:
|
||||
|
||||
1. Valid code with correct syntax
|
||||
2. Code with syntax errors
|
||||
3. Code with runtime errors
|
||||
4. Terminal commands that succeed/fail
|
||||
|
||||
## Key Observations
|
||||
|
||||
### Safety Trade-offs
|
||||
|
||||
The approval mechanism introduces latency as each dangerous operation requires an additional LLM call. However, this overhead prevents catastrophic failures and provides audit trails for critical actions. The system can be tuned by adjusting `REQUIRE_APPROVAL_FOR_DANGEROUS_OPS` based on trust level and use case requirements.
|
||||
|
||||
### Summarization Benefits
|
||||
|
||||
Automatic summarization significantly reduces token consumption when dealing with verbose tool outputs exceeding 10,000 characters. The LLM effectively extracts actionable information while preserving critical details. For terminal errors spanning hundreds of lines, summarization typically captures the root cause in a concise format. Outputs under the threshold are returned as-is, ensuring no information loss for moderately-sized results.
|
||||
|
||||
### Verification Limitations
|
||||
|
||||
While syntax verification catches many issues before execution, it cannot predict runtime failures or logical errors. The system works best when combined with error analysis that provides suggestions for fixing failed operations. For Python, compile-time syntax checking is highly accurate; for other languages, LLM-based validation serves as a reasonable approximation.
|
||||
|
||||
## Discussion Questions
|
||||
|
||||
1. How does LLM-based approval compare to rule-based safety checks?
|
||||
2. What are the trade-offs between automation and human oversight?
|
||||
3. How can verification be extended to more complex validation scenarios?
|
||||
4. What metrics should be used to evaluate summarization quality?
|
||||
5. How should the system handle edge cases where approval is needed but the LLM is unavailable?
|
||||
|
||||
## Extensions
|
||||
|
||||
### Suggested Improvements
|
||||
|
||||
1. **Caching**: Cache approval decisions for identical operations
|
||||
2. **Rollback**: Implement undo functionality for file operations
|
||||
3. **Sandboxing**: Use containers for true code isolation
|
||||
4. **Multi-step Planning**: Break complex operations into verified steps
|
||||
5. **Learning**: Train models on historical approval patterns
|
||||
|
||||
### Additional Tools
|
||||
|
||||
Consider implementing:
|
||||
- Database query tools with schema validation
|
||||
- API calling tools with rate limiting
|
||||
- File backup/restore functionality
|
||||
- Distributed execution across multiple machines
|
||||
|
||||
## Conclusion
|
||||
|
||||
This experiment demonstrates that production-ready execution tools require multiple layers of safety, verification, and result processing. The combination of LLM-based approval, automatic summarization, and verification creates a robust system suitable for real-world autonomous agent deployments. The architecture patterns shown here can be adapted to virtually any tool category where safety and reliability are paramount.
|
||||
@@ -0,0 +1,488 @@
|
||||
# Execution Tools MCP Server / 执行工具 MCP 服务器
|
||||
|
||||
> Companion code for *AI Agents in Depth*, Chapter 4 — **Experiment 4-3 ★★**. MCP execution tools with LLM approval, auto-verification, and long-output truncation/persist.
|
||||
> 配套《深入理解 AI Agent》第 4 章 **实验 4-3 ★★**。带 LLM 事前审批、自动校验、长输出截断与持久化的执行工具 MCP 服务器。
|
||||
|
||||
← [Chapter 4 index / 返回第 4 章目录](../README.md)
|
||||
|
||||
## Code map
|
||||
|
||||
- **Run first:** `python cli.py demo` (offline end-to-end path).
|
||||
- **Start here:** `cli.py::cmd_demo` constructs `ExecutionTools`; `execution_tools.py::ExecutionTools` is the shared execution surface.
|
||||
- **Core behavior:** `file_tools.py::FileTools`, `terminal_controller.py::TerminalController` and `multilang_executor.py::LanguageExecutor` implement validation, execution and output handling.
|
||||
- **State / protocol:** `experiment_protocol.json`, workspace boundaries, approval flags and structured tool-result fields.
|
||||
- **Verifier:** `test_execution_tools.py`, `test_file_tools.py`, `test_terminal_controller.py` and `run_experiment_4_3.py` acceptance gates.
|
||||
- **Experiment variable:** approval, syntax verification, long-output summarization/truncation and sandbox settings.
|
||||
- **Skip on first pass:** MCP transport, calendar/GitHub integrations and provider-specific LLM adapters.
|
||||
|
||||
---
|
||||
|
||||
## English
|
||||
|
||||
An MCP (Model Context Protocol) server that provides comprehensive execution tools with built-in safety mechanisms for AI agents.
|
||||
|
||||
This project corresponds to Experiment 4-3 in the book’s “Execution Tools” section. It focuses on layered safety (input validation, permission control, LLM pre-approval), automatic syntax verification and feedback loops, and truncation plus persistence of long outputs. Recommended start: `python cli.py demo`.
|
||||
|
||||
### Features
|
||||
|
||||
#### Safety Mechanisms
|
||||
|
||||
1. **LLM-Based Approval**: Irreversible operations require approval from a secondary LLM before execution
|
||||
2. **Result Summarization**: Execution tool outputs larger than 10,000 characters are automatically summarized by an LLM for easier processing
|
||||
3. **Automatic Verification**: Operations that can be verified (e.g., syntax checking) are automatically validated
|
||||
|
||||
#### Tool Categories
|
||||
|
||||
##### File System Tools
|
||||
- **file_write**: Write content to files with automatic syntax verification
|
||||
- **file_edit**: Edit existing files with diff preview and verification
|
||||
|
||||
##### Generic Execution Tools
|
||||
- **code_interpreter**: Execute Python code in a sandboxed environment with result analysis
|
||||
- **virtual_terminal**: Execute shell commands with error summarization
|
||||
|
||||
##### External System Integration Tools
|
||||
- **google_calendar_add**: Add events to Google Calendar
|
||||
- **github_create_pr**: Create GitHub Pull Requests with validation
|
||||
|
||||
### Installation
|
||||
|
||||
```bash
|
||||
# From the repository root: use the shared Chapter 4 environment
|
||||
uv sync --locked --python 3.12 --extra ch4
|
||||
|
||||
# Activate it before changing directories:
|
||||
# macOS/Linux:
|
||||
source .venv/bin/activate
|
||||
# Windows PowerShell: .venv\Scripts\Activate.ps1
|
||||
# Windows cmd: .venv\Scripts\activate.bat
|
||||
|
||||
# pip fallback when uv is not installed:
|
||||
# python -m pip install -e ".[ch4]"
|
||||
|
||||
cd chapter4/execution-tools
|
||||
|
||||
# Exact legacy parity path, including optional scientific/ML spreadsheet packages:
|
||||
# python -m pip install -r requirements.txt
|
||||
```
|
||||
|
||||
### Configuration
|
||||
|
||||
1. Copy `env.example` to `.env`:
|
||||
```bash
|
||||
cp env.example .env
|
||||
```
|
||||
|
||||
2. Configure your environment variables:
|
||||
```
|
||||
# LLM Configuration (for safety checks and summarization)
|
||||
PROVIDER=kimi
|
||||
|
||||
# API Keys (set the one for your provider)
|
||||
KIMI_API_KEY=your_kimi_key
|
||||
# DashScope / Bailian (Qwen)
|
||||
# PROVIDER=dashscope # qwen and bailian are accepted aliases
|
||||
# DASHSCOPE_API_KEY=your_dashscope_key
|
||||
# SILICONFLOW_API_KEY=your_siliconflow_key
|
||||
# DOUBAO_API_KEY=your_doubao_key
|
||||
# OPENROUTER_API_KEY=your_openrouter_key
|
||||
|
||||
# Model (optional, defaults to provider's default)
|
||||
# MODEL=kimi-k3
|
||||
|
||||
# Model parameters
|
||||
TEMPERATURE=0.7
|
||||
MAX_TOKENS=4096
|
||||
|
||||
# External Services (optional)
|
||||
GOOGLE_CALENDAR_CREDENTIALS_FILE=credentials.json
|
||||
GITHUB_TOKEN=your_github_token
|
||||
|
||||
# Safety Settings
|
||||
REQUIRE_APPROVAL_FOR_DANGEROUS_OPS=true
|
||||
AUTO_SUMMARIZE_COMPLEX_OUTPUT=true
|
||||
AUTO_VERIFY_CODE=true
|
||||
```
|
||||
|
||||
**Supported Providers:**
|
||||
- `siliconflow`: Qwen/Qwen3-235B-A22B-Thinking-2507
|
||||
- `dashscope` / `qwen` / `bailian`: qwen3.7-plus (Alibaba Cloud Model Studio)
|
||||
- `doubao`: doubao-seed-1-6-thinking-250715
|
||||
- `kimi`/`moonshot`: kimi-k3
|
||||
- `openrouter`: google/gemini-3.5-flash (or openai/gpt-5.6-luna, anthropic/claude-sonnet-4.6)
|
||||
|
||||
> **Universal OpenRouter fallback**: when the configured `PROVIDER`'s key is
|
||||
> missing but `OPENROUTER_API_KEY` is set, the LLM steps (approval,
|
||||
> summarization, error/syntax analysis) transparently switch to `openrouter`
|
||||
> via `Config.effective_provider()`. Set `MODEL` to a `provider/model` id for
|
||||
> OpenRouter, e.g. `MODEL=openai/gpt-5.6-luna`.
|
||||
|
||||
### Usage
|
||||
|
||||
#### CLI entry (`cli.py`)
|
||||
|
||||
`cli.py` is the unified command-line entry for listing tools, calling each execution tool, and running end-to-end demos. It reuses the same tool implementations as the MCP server, so behavior matches.
|
||||
|
||||
```bash
|
||||
# Overview and all subcommands
|
||||
python cli.py --help
|
||||
|
||||
# List all execution tools
|
||||
python cli.py list
|
||||
|
||||
# End-to-end offline demo (recommended first; no API key)
|
||||
python cli.py demo
|
||||
|
||||
# Call a tool individually
|
||||
python cli.py code --language python --code "print(2 ** 10)"
|
||||
python cli.py shell "python3 --version"
|
||||
python cli.py write --path notes.txt --content "hello" --overwrite
|
||||
python cli.py edit --path notes.txt --search hello --replace world
|
||||
```
|
||||
|
||||
Global flags (before the subcommand):
|
||||
|
||||
| Flag | Effect |
|
||||
|------|------|
|
||||
| `--provider` | Override LLM provider (`PROVIDER`) |
|
||||
| `--workspace` | Override workspace directory (file ops restricted here) |
|
||||
| `--no-approval` | Disable LLM pre-approval for dangerous ops |
|
||||
| `--no-verify` | Disable auto syntax check for write/code |
|
||||
| `--no-summarize` | Disable LLM summarization of long output (still truncates and persists) |
|
||||
|
||||
**Offline operation**: `list`, `demo`, and `code`/`shell`/`write`/`edit` with approval/summarize/non-Python verify off need no API key. API key is needed for: LLM pre-approval, LLM summarization of long output, non-Python syntax checks. `calendar` and `pr` also need their external credentials.
|
||||
|
||||
> **Warning — `--no-approval`**: this flag bypasses the LLM pre-approval check for dangerous operations. Use it only in controlled local demos (e.g. a throwaway workspace). Never combine it with real workspaces or destructive commands.
|
||||
>
|
||||
> **Long-output truncation and persistence**: when `code_interpreter` / `virtual_terminal` output exceeds the threshold (default 200 lines or 10000 characters), the tool keeps only the first and last 50 lines in context, writes the full output to a temp file, and returns the path in `stdout_file` / `stderr_file`. This path does **not** depend on an LLM and works offline.
|
||||
|
||||
#### Running the MCP Server
|
||||
|
||||
```bash
|
||||
python server.py
|
||||
```
|
||||
|
||||
#### Using with MCP Client
|
||||
|
||||
```python
|
||||
import asyncio
|
||||
|
||||
from mcp import ClientSession, StdioServerParameters
|
||||
from mcp.client.stdio import stdio_client
|
||||
|
||||
async def use_tools():
|
||||
server_params = StdioServerParameters(
|
||||
command="python",
|
||||
args=["server.py"],
|
||||
)
|
||||
|
||||
async with stdio_client(server_params) as (read, write):
|
||||
async with ClientSession(read, write) as session:
|
||||
await session.initialize()
|
||||
|
||||
# Use file write tool
|
||||
result = await session.call_tool("file_write", {
|
||||
"path": "test.py",
|
||||
"content": "print('Hello, World!')"
|
||||
})
|
||||
|
||||
# Use code interpreter
|
||||
result = await session.call_tool("code_interpreter", {
|
||||
"code": "import math\nprint(math.sqrt(16))"
|
||||
})
|
||||
|
||||
# Use virtual terminal
|
||||
result = await session.call_tool("virtual_terminal", {
|
||||
"command": "ls -la"
|
||||
})
|
||||
|
||||
|
||||
asyncio.run(use_tools())
|
||||
```
|
||||
|
||||
#### Testing Individual Tools
|
||||
|
||||
```bash
|
||||
# Test file operations
|
||||
python test_file_tools.py
|
||||
|
||||
# Test execution tools
|
||||
python test_execution_tools.py
|
||||
|
||||
# Test external integrations
|
||||
python test_external_tools.py
|
||||
```
|
||||
|
||||
### Architecture
|
||||
|
||||
The server implements a layered architecture:
|
||||
|
||||
1. **Safety Layer**: Intercepts dangerous operations and validates them
|
||||
2. **Tool Layer**: Implements individual tool logic
|
||||
3. **Verification Layer**: Validates outputs and provides feedback
|
||||
4. **Integration Layer**: Connects to external services
|
||||
|
||||
### Real desktop and Android environments
|
||||
|
||||
The exact Experiment 4-3 runner includes two action probes instead of treating
|
||||
installed packages as execution evidence:
|
||||
|
||||
- `virtual_desktop_execute` starts a bounded Xvfb display and headful Chromium,
|
||||
enters an HTTPS URL through `xdotool` keyboard events, verifies the resulting
|
||||
window title, and hashes a real framebuffer screenshot captured by FFmpeg.
|
||||
- `virtual_mobile_execute` connects to a running AndroidWorld Docker emulator,
|
||||
opens Android Wi-Fi Settings through ADB, verifies the focused activity,
|
||||
captures and hashes its pixels, then returns to the launcher with a real
|
||||
input event.
|
||||
|
||||
The AndroidWorld image is external and is not vendored. With a populated image
|
||||
available locally, start an API-33 emulator with KVM and run the campaign:
|
||||
|
||||
```bash
|
||||
docker run -d --name exp4-3-android --privileged --device /dev/kvm \
|
||||
-p 127.0.0.1:5000:5000 android_world_patched:populated3
|
||||
|
||||
python run_experiment_4_3.py \
|
||||
--android-container exp4-3-android \
|
||||
--github-head-branch <pushed-experiment-branch> \
|
||||
--github-base-branch <base-branch>
|
||||
```
|
||||
|
||||
The host desktop path requires `Xvfb`, `xdotool`, FFmpeg, and Chromium; the
|
||||
spreadsheet screenshot gate additionally requires LibreOffice Calc. GitHub PR
|
||||
creation queries for an existing head/base PR before mutation, so a campaign
|
||||
retry verifies and reuses the first PR instead of creating a duplicate.
|
||||
External Calendar, GitHub, and email mutations remain credential-gated and are
|
||||
reported as blocked if their real providers are unavailable.
|
||||
|
||||
### Examples
|
||||
|
||||
See `examples.py` for comprehensive usage examples.
|
||||
|
||||
---
|
||||
|
||||
## 中文
|
||||
|
||||
为 AI Agent 提供带内置安全机制的综合执行工具 MCP（Model Context Protocol）服务器。
|
||||
|
||||
本项目对应书中第 4 章「执行工具」一节的实验 4-3，聚焦执行工具的安全机制：
|
||||
分层安全防护（输入验证、权限控制、LLM 事前审批）、自动语法验证与反馈闭环、
|
||||
以及长输出的截断与持久化。推荐从 `python cli.py demo` 开始。
|
||||
|
||||
### 功能
|
||||
|
||||
#### 安全机制
|
||||
|
||||
1. **基于 LLM 的审批**：不可逆操作在执行前需经二级 LLM 审批
|
||||
2. **结果总结**：执行工具输出超过 10,000 字符时由 LLM 自动总结，便于处理
|
||||
3. **自动校验**：可校验的操作（如语法检查）自动验证
|
||||
|
||||
#### 工具分类
|
||||
|
||||
##### 文件系统工具
|
||||
- **file_write**：写入文件，自动语法校验
|
||||
- **file_edit**：编辑已有文件，带 diff 预览与校验
|
||||
|
||||
##### 通用执行工具
|
||||
- **code_interpreter**：沙箱中执行 Python，带结果分析
|
||||
- **virtual_terminal**：执行 shell 命令，带错误总结
|
||||
|
||||
##### 外部系统集成工具
|
||||
- **google_calendar_add**：向 Google Calendar 添加事件
|
||||
- **github_create_pr**：创建 GitHub Pull Request（带校验）
|
||||
|
||||
### 安装
|
||||
|
||||
```bash
|
||||
# 在仓库根目录使用统一的第 4 章环境
|
||||
uv sync --locked --python 3.12 --extra ch4
|
||||
|
||||
# 切换目录前先激活环境：
|
||||
# macOS/Linux：
|
||||
source .venv/bin/activate
|
||||
# Windows PowerShell：.venv\Scripts\Activate.ps1
|
||||
# Windows cmd：.venv\Scripts\activate.bat
|
||||
|
||||
# 未安装 uv 时可用 pip 兜底：
|
||||
# python -m pip install -e ".[ch4]"
|
||||
|
||||
cd chapter4/execution-tools
|
||||
|
||||
# 精确复现旧版单项目环境，含可选科学计算/机器学习/表格处理依赖：
|
||||
# python -m pip install -r requirements.txt
|
||||
```
|
||||
|
||||
### 配置
|
||||
|
||||
1. 复制 `env.example` 为 `.env`：
|
||||
```bash
|
||||
cp env.example .env
|
||||
```
|
||||
|
||||
2. 配置环境变量：
|
||||
```
|
||||
# LLM Configuration (for safety checks and summarization)
|
||||
PROVIDER=kimi
|
||||
|
||||
# API Keys (set the one for your provider)
|
||||
KIMI_API_KEY=your_kimi_key
|
||||
# DashScope / Bailian (Qwen)
|
||||
# PROVIDER=dashscope # qwen and bailian are accepted aliases
|
||||
# DASHSCOPE_API_KEY=your_dashscope_key
|
||||
# SILICONFLOW_API_KEY=your_siliconflow_key
|
||||
# DOUBAO_API_KEY=your_doubao_key
|
||||
# OPENROUTER_API_KEY=your_openrouter_key
|
||||
|
||||
# Model (optional, defaults to provider's default)
|
||||
# MODEL=kimi-k3
|
||||
|
||||
# Model parameters
|
||||
TEMPERATURE=0.7
|
||||
MAX_TOKENS=4096
|
||||
|
||||
# External Services (optional)
|
||||
GOOGLE_CALENDAR_CREDENTIALS_FILE=credentials.json
|
||||
GITHUB_TOKEN=your_github_token
|
||||
|
||||
# Safety Settings
|
||||
REQUIRE_APPROVAL_FOR_DANGEROUS_OPS=true
|
||||
AUTO_SUMMARIZE_COMPLEX_OUTPUT=true
|
||||
AUTO_VERIFY_CODE=true
|
||||
```
|
||||
|
||||
**支持的 Provider：**
|
||||
- `siliconflow`：Qwen/Qwen3-235B-A22B-Thinking-2507
|
||||
- `dashscope` / `qwen` / `bailian`：qwen3.7-plus（阿里云百炼 / Model Studio）
|
||||
- `doubao`：doubao-seed-1-6-thinking-250715
|
||||
- `kimi`/`moonshot`：kimi-k3
|
||||
- `openrouter`：google/gemini-3.5-flash（或 openai/gpt-5.6-luna、anthropic/claude-sonnet-4.6）
|
||||
|
||||
> **OpenRouter 通用兜底**：当配置的 `PROVIDER` 对应 Key 缺失，但设置了
|
||||
> `OPENROUTER_API_KEY` 时，LLM 步骤（审批、总结、错误/语法分析）经
|
||||
> `Config.effective_provider()` 透明切换到 `openrouter`。
|
||||
> 为 OpenRouter 设置 `MODEL` 为 `provider/model` 形式，例如
|
||||
> `MODEL=openai/gpt-5.6-luna`。
|
||||
|
||||
### 使用
|
||||
|
||||
#### 命令行入口（`cli.py`）
|
||||
|
||||
`cli.py` 是统一的命令行入口，用于列出、单独调用每个执行工具，并运行端到端演示。
|
||||
它复用与 MCP 服务器相同的工具实现，因此行为完全一致。
|
||||
|
||||
```bash
|
||||
# 查看总帮助与所有子命令
|
||||
python cli.py --help
|
||||
|
||||
# 列出所有执行工具
|
||||
python cli.py list
|
||||
|
||||
# 端到端离线演示（推荐先看这个；无需 API key 即可运行）
|
||||
python cli.py demo
|
||||
|
||||
# 单独调用某个工具
|
||||
python cli.py code --language python --code "print(2 ** 10)"
|
||||
python cli.py shell "python3 --version"
|
||||
python cli.py write --path notes.txt --content "hello" --overwrite
|
||||
python cli.py edit --path notes.txt --search hello --replace world
|
||||
```
|
||||
|
||||
全局开关（放在子命令之前）：
|
||||
|
||||
| 开关 | 作用 |
|
||||
|------|------|
|
||||
| `--provider` | 覆盖 LLM 提供商（`PROVIDER`） |
|
||||
| `--workspace` | 覆盖工作目录（文件操作被限制在此目录内） |
|
||||
| `--no-approval` | 关闭危险操作的 LLM 事前审批 |
|
||||
| `--no-verify` | 关闭写文件/代码的自动语法校验 |
|
||||
| `--no-summarize` | 关闭长输出的 LLM 总结（仍会截断并持久化） |
|
||||
|
||||
**离线运行**：`list`、`demo` 以及关闭了审批/总结/非 Python 校验的
|
||||
`code`/`shell`/`write`/`edit` 均无需 API key。需要 API key 的场景为：LLM 事前审批、
|
||||
长输出的 LLM 总结、非 Python 语法校验。`calendar` 与 `pr` 还额外需要相应外部凭据。
|
||||
|
||||
> **警告 —— `--no-approval`**：该开关会绕过危险操作的 LLM 事前审批，仅适用于受控的本地演示（如一次性临时工作区）。切勿在真实工作区中使用，也不要与破坏性命令搭配使用。
|
||||
>
|
||||
> **长输出的截断与持久化**：当 `code_interpreter` / `virtual_terminal` 的输出
|
||||
> 超过阈值（默认 200 行或 10000 字符）时，工具只在上下文中保留头尾各 50 行，
|
||||
> 完整输出落盘到临时文件，并在返回值的 `stdout_file` / `stderr_file` 字段给出路径。
|
||||
> 该机制不依赖 LLM，可离线工作。
|
||||
|
||||
#### 运行 MCP 服务器
|
||||
|
||||
```bash
|
||||
python server.py
|
||||
```
|
||||
|
||||
#### 配合 MCP 客户端
|
||||
|
||||
```python
|
||||
import asyncio
|
||||
|
||||
from mcp import ClientSession, StdioServerParameters
|
||||
from mcp.client.stdio import stdio_client
|
||||
|
||||
async def use_tools():
|
||||
server_params = StdioServerParameters(
|
||||
command="python",
|
||||
args=["server.py"],
|
||||
)
|
||||
|
||||
async with stdio_client(server_params) as (read, write):
|
||||
async with ClientSession(read, write) as session:
|
||||
await session.initialize()
|
||||
|
||||
# Use file write tool
|
||||
result = await session.call_tool("file_write", {
|
||||
"path": "test.py",
|
||||
"content": "print('Hello, World!')"
|
||||
})
|
||||
|
||||
# Use code interpreter
|
||||
result = await session.call_tool("code_interpreter", {
|
||||
"code": "import math\nprint(math.sqrt(16))"
|
||||
})
|
||||
|
||||
# Use virtual terminal
|
||||
result = await session.call_tool("virtual_terminal", {
|
||||
"command": "ls -la"
|
||||
})
|
||||
|
||||
|
||||
asyncio.run(use_tools())
|
||||
```
|
||||
|
||||
#### 测试单个工具
|
||||
|
||||
```bash
|
||||
# Test file operations
|
||||
python test_file_tools.py
|
||||
|
||||
# Test execution tools
|
||||
python test_execution_tools.py
|
||||
|
||||
# Test external integrations
|
||||
python test_external_tools.py
|
||||
```
|
||||
|
||||
### 架构
|
||||
|
||||
服务器采用分层架构：
|
||||
|
||||
1. **安全层**：拦截危险操作并校验
|
||||
2. **工具层**：实现各工具逻辑
|
||||
3. **校验层**：验证输出并反馈
|
||||
4. **集成层**：对接外部服务
|
||||
|
||||
### 示例
|
||||
|
||||
更完整的用法见 `examples.py`。另见 [`EXPERIMENT.md`](EXPERIMENT.md) 中的实验说明。
|
||||
|
||||
---
|
||||
|
||||
## Notes / 说明
|
||||
|
||||
- Start with `python cli.py demo` (no API key).
|
||||
- 建议从 `python cli.py demo` 开始（无需 API Key）。
|
||||
- Long-output truncation/persistence works offline without LLM.
|
||||
- 长输出截断与持久化不依赖 LLM，可离线。
|
||||
@@ -0,0 +1,410 @@
|
||||
#!/usr/bin/env python3
|
||||
"""执行工具统一命令行入口（实验 4-3：执行工具 MCP 服务器）。
|
||||
|
||||
本文件提供一个 argparse 命令行界面，用于列出、单独调用每个执行工具，并运行
|
||||
一个端到端的离线演示。它复用 server.py 背后的同一批工具实现，因此命令行的
|
||||
行为与 MCP 服务器完全一致。
|
||||
|
||||
工具清单（与 server.py 一致）：
|
||||
file_write 写文件（写入前自动做语法/linter 校验）
|
||||
file_edit 按“搜索-替换”编辑文件（带 diff 预览与校验）
|
||||
code_interpreter 多语言沙盒代码执行（危险操作审批、长输出截断持久化）
|
||||
virtual_terminal Shell 命令执行（危险命令检测、长输出截断持久化）
|
||||
google_calendar_add 创建 Google 日历事件（需要凭据）
|
||||
github_create_pr 创建 GitHub Pull Request（需要 token）
|
||||
|
||||
安全机制（与书中“执行工具”一节对应）：
|
||||
- LLM 事前审批：不可逆/危险操作在执行前交由独立 LLM 审查
|
||||
- 自动验证：Python 语法通过 compile() 本地校验，其他语言由 LLM 兜底
|
||||
- 长输出截断与持久化：超过阈值时仅保留头尾若干行，完整输出落盘到临时文件
|
||||
|
||||
用法示例：
|
||||
python cli.py list
|
||||
python cli.py demo
|
||||
python cli.py code --language python --code "print(2 ** 10)"
|
||||
python cli.py shell "python3 --version"
|
||||
python cli.py write --path notes.txt --content "hello" --overwrite
|
||||
python cli.py --no-approval --no-summarize shell "ls -la"
|
||||
|
||||
不需要 API key 的命令：list、demo（离线路径）、以及关闭了审批/总结/非 Python
|
||||
校验的 code/shell/write/edit。需要 API key 的场景：LLM 审批、长输出 LLM 总结、
|
||||
非 Python 语法校验。calendar 与 pr 还额外需要相应的外部凭据。
|
||||
"""
|
||||
|
||||
import argparse
|
||||
import asyncio
|
||||
import json
|
||||
import os
|
||||
import sys
|
||||
import tempfile
|
||||
import textwrap
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# 工具元数据（供 `list` 子命令展示）
|
||||
# ---------------------------------------------------------------------------
|
||||
TOOL_CATALOG = [
|
||||
("file_write", "文件系统", "写文件，写入前自动做语法/linter 校验"),
|
||||
("file_edit", "文件系统", "按搜索-替换编辑文件，附带 diff 预览与校验"),
|
||||
("code_interpreter", "通用执行", "多语言沙盒代码执行（Python/JS/Go/Java/C++/Rust/PHP/Bash）"),
|
||||
("virtual_terminal", "通用执行", "Shell 命令执行，含危险命令检测与长输出截断"),
|
||||
("google_calendar_add", "外部系统", "创建 Google 日历事件（需要 credentials.json）"),
|
||||
("github_create_pr", "外部系统", "创建 GitHub Pull Request（需要 GITHUB_TOKEN）"),
|
||||
]
|
||||
|
||||
|
||||
def _apply_global_env(args: argparse.Namespace) -> None:
|
||||
"""把全局开关写入环境变量，供 config.py 在导入时读取。
|
||||
|
||||
config.Config 在模块导入时读取环境变量，因此所有涉及配置的模块都必须在此
|
||||
函数执行之后才导入（本文件中的工具模块均为函数内延迟导入）。
|
||||
"""
|
||||
if args.provider:
|
||||
os.environ["PROVIDER"] = args.provider
|
||||
if args.workspace:
|
||||
os.environ["WORKSPACE_DIR"] = os.path.abspath(args.workspace)
|
||||
if args.no_approval:
|
||||
os.environ["REQUIRE_APPROVAL_FOR_DANGEROUS_OPS"] = "false"
|
||||
if args.no_verify:
|
||||
os.environ["AUTO_VERIFY_CODE"] = "false"
|
||||
if args.no_summarize:
|
||||
os.environ["AUTO_SUMMARIZE_COMPLEX_OUTPUT"] = "false"
|
||||
|
||||
|
||||
def _build_tools():
|
||||
"""构造共享的工具实例（延迟导入，确保环境变量已就绪）。"""
|
||||
from llm_helper import LLMHelper
|
||||
from file_tools import FileTools
|
||||
from execution_tools import ExecutionTools
|
||||
from external_tools import ExternalTools
|
||||
|
||||
llm_helper = LLMHelper() # 客户端惰性创建，离线时不需要 API key
|
||||
return {
|
||||
"llm": llm_helper,
|
||||
"file": FileTools(llm_helper),
|
||||
"exec": ExecutionTools(llm_helper),
|
||||
"external": ExternalTools(llm_helper),
|
||||
}
|
||||
|
||||
|
||||
def _print_result(result: dict) -> None:
|
||||
"""统一以 JSON 打印工具返回结果。"""
|
||||
print(json.dumps(result, indent=2, ensure_ascii=False))
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# 子命令实现
|
||||
# ---------------------------------------------------------------------------
|
||||
def cmd_list(args: argparse.Namespace) -> int:
|
||||
print("可用执行工具：\n")
|
||||
print(f" {'工具名':<20} {'类别':<8} 说明")
|
||||
print(f" {'-' * 20} {'-' * 8} {'-' * 40}")
|
||||
for name, category, desc in TOOL_CATALOG:
|
||||
print(f" {name:<20} {category:<8} {desc}")
|
||||
print("\n用 `python cli.py <子命令> --help` 查看每个工具的参数。")
|
||||
print("用 `python cli.py demo` 运行端到端离线演示。")
|
||||
return 0
|
||||
|
||||
|
||||
def cmd_code(args: argparse.Namespace) -> int:
|
||||
code = args.code
|
||||
if args.file:
|
||||
with open(args.file, "r", encoding="utf-8") as f:
|
||||
code = f.read()
|
||||
if not code:
|
||||
print("错误：请通过 --code 或 --file 提供要执行的代码。", file=sys.stderr)
|
||||
return 2
|
||||
|
||||
tools = _build_tools()
|
||||
result = asyncio.run(tools["exec"].code_interpreter(
|
||||
code=code,
|
||||
language=args.language,
|
||||
timeout=args.timeout,
|
||||
stdin=args.stdin,
|
||||
))
|
||||
_print_result(result)
|
||||
return 0 if result.get("success") else 1
|
||||
|
||||
|
||||
def cmd_shell(args: argparse.Namespace) -> int:
|
||||
tools = _build_tools()
|
||||
result = asyncio.run(tools["exec"].virtual_terminal(
|
||||
command=args.command,
|
||||
timeout=args.timeout,
|
||||
))
|
||||
_print_result(result)
|
||||
return 0 if result.get("success") else 1
|
||||
|
||||
|
||||
def cmd_write(args: argparse.Namespace) -> int:
|
||||
content = args.content
|
||||
if args.content_file:
|
||||
with open(args.content_file, "r", encoding="utf-8") as f:
|
||||
content = f.read()
|
||||
if content is None:
|
||||
print("错误：请通过 --content 或 --content-file 提供文件内容。", file=sys.stderr)
|
||||
return 2
|
||||
|
||||
tools = _build_tools()
|
||||
result = asyncio.run(tools["file"].write_file(
|
||||
path=args.path,
|
||||
content=content,
|
||||
overwrite=args.overwrite,
|
||||
))
|
||||
_print_result(result)
|
||||
return 0 if result.get("success") else 1
|
||||
|
||||
|
||||
def cmd_edit(args: argparse.Namespace) -> int:
|
||||
tools = _build_tools()
|
||||
result = asyncio.run(tools["file"].edit_file(
|
||||
path=args.path,
|
||||
search=args.search,
|
||||
replace=args.replace,
|
||||
))
|
||||
_print_result(result)
|
||||
return 0 if result.get("success") else 1
|
||||
|
||||
|
||||
def cmd_calendar(args: argparse.Namespace) -> int:
|
||||
tools = _build_tools()
|
||||
result = asyncio.run(tools["external"].google_calendar_add(
|
||||
summary=args.summary,
|
||||
start_time=args.start,
|
||||
end_time=args.end,
|
||||
description=args.description,
|
||||
location=args.location,
|
||||
))
|
||||
_print_result(result)
|
||||
return 0 if result.get("success") else 1
|
||||
|
||||
|
||||
def cmd_pr(args: argparse.Namespace) -> int:
|
||||
tools = _build_tools()
|
||||
result = asyncio.run(tools["external"].github_create_pr(
|
||||
repo_name=args.repo,
|
||||
title=args.title,
|
||||
body=args.body,
|
||||
head_branch=args.head,
|
||||
base_branch=args.base,
|
||||
))
|
||||
_print_result(result)
|
||||
return 0 if result.get("success") else 1
|
||||
|
||||
|
||||
def cmd_demo(args: argparse.Namespace) -> int:
|
||||
"""端到端离线演示：模拟一个 Agent 用执行工具完成一个真实小任务。
|
||||
|
||||
场景：Agent 需要写一个词频统计脚本、生成样本数据、运行统计、再用 shell
|
||||
校验结果。演示同时覆盖四个安全机制：linter 校验、危险命令 fail-safe 审批、
|
||||
长输出截断与持久化。整个流程默认离线运行（关闭 LLM 总结）。
|
||||
"""
|
||||
# 演示放在独立临时工作区，避免污染当前目录。
|
||||
workspace = tempfile.mkdtemp(prefix="exec_tools_demo_")
|
||||
os.environ["WORKSPACE_DIR"] = workspace
|
||||
# 离线运行：关闭需要 LLM 的输出总结（截断持久化不依赖 LLM）。
|
||||
if "AUTO_SUMMARIZE_COMPLEX_OUTPUT" not in os.environ:
|
||||
os.environ["AUTO_SUMMARIZE_COMPLEX_OUTPUT"] = "false"
|
||||
|
||||
tools = _build_tools()
|
||||
file_tools = tools["file"]
|
||||
exec_tools = tools["exec"]
|
||||
|
||||
def section(title: str) -> None:
|
||||
print("\n" + "=" * 64)
|
||||
print(title)
|
||||
print("=" * 64)
|
||||
|
||||
print(f"演示工作区：{workspace}")
|
||||
print("（离线路径，无需 API key；如已配置 key，审批/总结将走真实 LLM）")
|
||||
|
||||
async def run() -> None:
|
||||
# 1. 写文件 + 自动 linter 校验（合法代码）
|
||||
section("1. file_write：写入词频统计脚本（自动语法校验）")
|
||||
script = textwrap.dedent('''\
|
||||
"""统计文本文件中的词频。"""
|
||||
import sys
|
||||
from collections import Counter
|
||||
|
||||
def word_count(path):
|
||||
with open(path, encoding="utf-8") as f:
|
||||
words = f.read().split()
|
||||
return Counter(words)
|
||||
|
||||
if __name__ == "__main__":
|
||||
for word, freq in word_count(sys.argv[1]).most_common(5):
|
||||
print(f"{word}\\t{freq}")
|
||||
''')
|
||||
r = await file_tools.write_file("wordcount.py", script, overwrite=True)
|
||||
print(f"结果：success={r['success']}, verification={r.get('verification')}")
|
||||
print(f"写入：{r.get('path')}")
|
||||
|
||||
# 2. linter 拦截语法错误的代码
|
||||
section("2. file_write：写入含语法错误的代码（linter 应拦截）")
|
||||
broken = "def broken(:\n return 1\n"
|
||||
r = await file_tools.write_file("broken.py", broken, overwrite=True)
|
||||
print(f"结果：success={r['success']}")
|
||||
print(f"校验反馈：{r.get('error')}")
|
||||
|
||||
# 3. 生成样本数据
|
||||
section("3. file_write：生成样本数据文件")
|
||||
sample = "apple banana apple cherry banana apple date cherry banana apple\n"
|
||||
r = await file_tools.write_file("data.txt", sample, overwrite=True)
|
||||
print(f"结果：success={r['success']}，写入 {r.get('bytes_written')} 字节")
|
||||
|
||||
# 4. code_interpreter：运行统计脚本
|
||||
section("4. code_interpreter：运行统计逻辑（Python 沙盒）")
|
||||
analysis = textwrap.dedent('''\
|
||||
from collections import Counter
|
||||
text = "apple banana apple cherry banana apple date cherry banana apple"
|
||||
for word, freq in Counter(text.split()).most_common(3):
|
||||
print(f"{word}: {freq}")
|
||||
''')
|
||||
r = await exec_tools.code_interpreter(code=analysis, language="python")
|
||||
print(f"结果：success={r['success']}, returncode={r.get('returncode')}")
|
||||
print("stdout:")
|
||||
print(textwrap.indent(r.get("stdout", ""), " "))
|
||||
|
||||
# 5. virtual_terminal：用 shell 校验数据文件
|
||||
section("5. virtual_terminal：用 shell 校验数据文件")
|
||||
r = await exec_tools.virtual_terminal(
|
||||
command=f"wc -w {workspace}/data.txt && echo '--- 词数统计完成 ---'"
|
||||
)
|
||||
print(f"结果：success={r['success']}, returncode={r.get('returncode')}")
|
||||
print("stdout:")
|
||||
print(textwrap.indent(r.get("stdout", ""), " "))
|
||||
|
||||
# 6. 长输出截断与持久化（离线，不需 LLM）
|
||||
section("6. code_interpreter：长输出自动截断并落盘")
|
||||
long_code = "for i in range(1000):\n print(f'line {i}: ' + 'x' * 20)\n"
|
||||
r = await exec_tools.code_interpreter(code=long_code, language="python")
|
||||
stdout = r.get("stdout", "")
|
||||
print(f"上下文中保留的输出行数：{len(stdout.splitlines())}（原始 1000 行）")
|
||||
print(f"完整输出落盘文件：{r.get('stdout_file')}")
|
||||
print("上下文中输出的尾部片段：")
|
||||
print(textwrap.indent("\n".join(stdout.splitlines()[-4:]), " "))
|
||||
|
||||
# 7. 危险命令的审批（离线 fail-safe / 在线交由真实 LLM 判断）
|
||||
section("7. virtual_terminal：危险命令触发审批")
|
||||
os.environ["REQUIRE_APPROVAL_FOR_DANGEROUS_OPS"] = "true"
|
||||
# 目标是不存在的临时路径，即便被执行也无副作用。
|
||||
danger = await exec_tools.virtual_terminal(
|
||||
command="rm -rf /tmp/exec_tools_demo_nonexistent_path_xyz"
|
||||
)
|
||||
print(f"结果：success={danger['success']}")
|
||||
if danger.get("error"):
|
||||
print(f"说明：{danger.get('error')}")
|
||||
print("（审批未通过：危险命令被拦截、未执行。离线无 LLM 时按 fail-safe 拒绝，"
|
||||
"在线时也可能被真实 LLM 判定为高风险而拒绝。）")
|
||||
else:
|
||||
print("（审批通过：已配置 API key，真实 LLM 判定该命令针对不存在路径、无副作用而放行。）")
|
||||
|
||||
section("演示完成")
|
||||
print("覆盖的安全机制：自动 linter 校验、危险命令审批、长输出截断持久化。")
|
||||
print(f"演示产物位于：{workspace}")
|
||||
|
||||
asyncio.run(run())
|
||||
return 0
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# 参数解析
|
||||
# ---------------------------------------------------------------------------
|
||||
def build_parser() -> argparse.ArgumentParser:
|
||||
parser = argparse.ArgumentParser(
|
||||
prog="cli.py",
|
||||
description="执行工具统一命令行入口（实验 4-3：执行工具 MCP 服务器）。",
|
||||
formatter_class=argparse.RawDescriptionHelpFormatter,
|
||||
epilog=textwrap.dedent("""\
|
||||
示例：
|
||||
python cli.py list 列出所有执行工具
|
||||
python cli.py demo 运行端到端离线演示
|
||||
python cli.py code --code "print(6*7)" 执行 Python 代码
|
||||
python cli.py shell "ls -la" 执行 shell 命令
|
||||
python cli.py write --path a.txt --content hi --overwrite
|
||||
python cli.py --no-approval shell "echo hello"
|
||||
|
||||
关闭 --no-approval / --no-summarize / --no-verify 后，
|
||||
code/shell/write/edit 等命令可完全离线运行，无需 API key。
|
||||
"""),
|
||||
)
|
||||
|
||||
# 全局开关
|
||||
parser.add_argument("--provider", help="LLM 提供商（覆盖 PROVIDER，如 dashscope/qwen/bailian/kimi/doubao/siliconflow/openrouter）")
|
||||
parser.add_argument("--workspace", help="工作目录（覆盖 WORKSPACE_DIR，文件操作被限制在此目录内）")
|
||||
parser.add_argument("--no-approval", action="store_true", help="关闭危险操作的 LLM 事前审批")
|
||||
parser.add_argument("--no-verify", action="store_true", help="关闭写文件/代码的自动语法校验")
|
||||
parser.add_argument("--no-summarize", action="store_true", help="关闭长输出的 LLM 总结（仍会截断持久化）")
|
||||
|
||||
sub = parser.add_subparsers(dest="command", metavar="<子命令>")
|
||||
|
||||
p = sub.add_parser("list", help="列出所有可用的执行工具")
|
||||
p.set_defaults(func=cmd_list)
|
||||
|
||||
p = sub.add_parser("demo", help="运行端到端离线演示（推荐先看这个）")
|
||||
p.set_defaults(func=cmd_demo)
|
||||
|
||||
p = sub.add_parser("code", help="调用 code_interpreter 执行代码")
|
||||
p.add_argument("--code", help="要执行的代码字符串")
|
||||
p.add_argument("--file", help="从文件读取要执行的代码")
|
||||
p.add_argument("--language", default="python",
|
||||
help="编程语言（python/javascript/typescript/go/java/cpp/rust/php/bash，默认 python）")
|
||||
p.add_argument("--timeout", type=float, default=30.0, help="执行超时秒数（默认 30）")
|
||||
p.add_argument("--stdin", help="可选的标准输入")
|
||||
p.set_defaults(func=cmd_code)
|
||||
|
||||
p = sub.add_parser("shell", help="调用 virtual_terminal 执行 shell 命令")
|
||||
p.add_argument("command", help="要执行的 shell 命令")
|
||||
p.add_argument("--timeout", type=int, default=30, help="超时秒数（默认 30）")
|
||||
p.set_defaults(func=cmd_shell)
|
||||
|
||||
p = sub.add_parser("write", help="调用 file_write 写文件")
|
||||
p.add_argument("--path", required=True, help="文件路径（相对工作目录或绝对路径）")
|
||||
p.add_argument("--content", help="文件内容")
|
||||
p.add_argument("--content-file", help="从文件读取要写入的内容")
|
||||
p.add_argument("--overwrite", action="store_true", help="允许覆盖已存在文件")
|
||||
p.set_defaults(func=cmd_write)
|
||||
|
||||
p = sub.add_parser("edit", help="调用 file_edit 按搜索-替换编辑文件")
|
||||
p.add_argument("--path", required=True, help="文件路径")
|
||||
p.add_argument("--search", required=True, help="要搜索的文本")
|
||||
p.add_argument("--replace", required=True, help="替换文本")
|
||||
p.set_defaults(func=cmd_edit)
|
||||
|
||||
p = sub.add_parser("calendar", help="调用 google_calendar_add 创建日历事件（需要凭据）")
|
||||
p.add_argument("--summary", required=True, help="事件标题")
|
||||
p.add_argument("--start", required=True, help="开始时间（ISO 8601，如 2025-10-01T10:00:00）")
|
||||
p.add_argument("--end", required=True, help="结束时间（ISO 8601）")
|
||||
p.add_argument("--description", help="事件描述")
|
||||
p.add_argument("--location", help="事件地点")
|
||||
p.set_defaults(func=cmd_calendar)
|
||||
|
||||
p = sub.add_parser("pr", help="调用 github_create_pr 创建 Pull Request（需要 token）")
|
||||
p.add_argument("--repo", required=True, help="仓库名（owner/repo 格式）")
|
||||
p.add_argument("--title", required=True, help="PR 标题")
|
||||
p.add_argument("--body", required=True, help="PR 描述")
|
||||
p.add_argument("--head", required=True, help="源分支")
|
||||
p.add_argument("--base", default="main", help="目标分支（默认 main）")
|
||||
p.set_defaults(func=cmd_pr)
|
||||
|
||||
return parser
|
||||
|
||||
|
||||
def main(argv=None) -> int:
|
||||
parser = build_parser()
|
||||
args = parser.parse_args(argv)
|
||||
|
||||
if not getattr(args, "command", None):
|
||||
parser.print_help()
|
||||
return 0
|
||||
|
||||
_apply_global_env(args)
|
||||
try:
|
||||
return args.func(args)
|
||||
except KeyboardInterrupt:
|
||||
print("\n已中断。", file=sys.stderr)
|
||||
return 130
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
sys.exit(main())
|
||||
@@ -0,0 +1,187 @@
|
||||
"""Configuration management for the execution tools MCP server."""
|
||||
|
||||
import os
|
||||
import sys
|
||||
from pathlib import Path
|
||||
from typing import Optional
|
||||
from dotenv import load_dotenv
|
||||
|
||||
# Load environment variables
|
||||
load_dotenv()
|
||||
|
||||
|
||||
def _env_int(name: str, default: int) -> int:
|
||||
"""Read an integer env var; fall back to default (with a warning) if malformed."""
|
||||
raw = os.getenv(name)
|
||||
if raw is None:
|
||||
return default
|
||||
try:
|
||||
return int(raw)
|
||||
except ValueError:
|
||||
print(f"Warning: invalid {name}={raw!r} (must be an integer); using default {default}",
|
||||
file=sys.stderr)
|
||||
return default
|
||||
|
||||
|
||||
def _env_float(name: str, default: float) -> float:
|
||||
"""Read a float env var; fall back to default (with a warning) if malformed."""
|
||||
raw = os.getenv(name)
|
||||
if raw is None:
|
||||
return default
|
||||
try:
|
||||
return float(raw)
|
||||
except ValueError:
|
||||
print(f"Warning: invalid {name}={raw!r} (must be a number); using default {default}",
|
||||
file=sys.stderr)
|
||||
return default
|
||||
|
||||
|
||||
class Config:
|
||||
"""Configuration for the MCP server."""
|
||||
|
||||
# LLM Configuration
|
||||
PROVIDER: str = os.getenv("PROVIDER", "kimi")
|
||||
|
||||
# API Keys
|
||||
DASHSCOPE_API_KEY: Optional[str] = os.getenv("DASHSCOPE_API_KEY")
|
||||
SILICONFLOW_API_KEY: Optional[str] = os.getenv("SILICONFLOW_API_KEY")
|
||||
DOUBAO_API_KEY: Optional[str] = os.getenv("DOUBAO_API_KEY")
|
||||
KIMI_API_KEY: Optional[str] = os.getenv("KIMI_API_KEY")
|
||||
MOONSHOT_API_KEY: Optional[str] = os.getenv("MOONSHOT_API_KEY")
|
||||
OPENROUTER_API_KEY: Optional[str] = os.getenv("OPENROUTER_API_KEY")
|
||||
DASHSCOPE_BASE_URL: str = os.getenv(
|
||||
"DASHSCOPE_BASE_URL", "https://dashscope.aliyuncs.com/compatible-mode/v1"
|
||||
)
|
||||
|
||||
# Model names (optional, defaults to provider defaults)
|
||||
MODEL: Optional[str] = os.getenv("MODEL")
|
||||
|
||||
# Model parameters
|
||||
TEMPERATURE: float = _env_float("TEMPERATURE", 0.7)
|
||||
MAX_TOKENS: int = _env_int("MAX_TOKENS", 4096)
|
||||
|
||||
# External Services
|
||||
GOOGLE_CALENDAR_CREDENTIALS_FILE: str = os.getenv(
|
||||
"GOOGLE_CALENDAR_CREDENTIALS_FILE",
|
||||
"credentials.json"
|
||||
)
|
||||
GITHUB_TOKEN: Optional[str] = os.getenv("GITHUB_TOKEN")
|
||||
|
||||
# Safety Settings
|
||||
REQUIRE_APPROVAL_FOR_DANGEROUS_OPS: bool = (
|
||||
os.getenv("REQUIRE_APPROVAL_FOR_DANGEROUS_OPS", "true").lower() == "true"
|
||||
)
|
||||
AUTO_SUMMARIZE_COMPLEX_OUTPUT: bool = (
|
||||
os.getenv("AUTO_SUMMARIZE_COMPLEX_OUTPUT", "true").lower() == "true"
|
||||
)
|
||||
AUTO_VERIFY_CODE: bool = (
|
||||
os.getenv("AUTO_VERIFY_CODE", "true").lower() == "true"
|
||||
)
|
||||
MAX_OUTPUT_LENGTH: int = _env_int("MAX_OUTPUT_LENGTH", 1000)
|
||||
|
||||
# Workspace Configuration
|
||||
WORKSPACE_DIR: Path = Path(os.getenv("WORKSPACE_DIR", os.getcwd()))
|
||||
|
||||
@classmethod
|
||||
def get_api_key(cls, provider: str) -> Optional[str]:
|
||||
"""Get API key for the specified provider."""
|
||||
provider = provider.lower()
|
||||
provider = {"qwen": "dashscope", "bailian": "dashscope"}.get(provider, provider)
|
||||
if provider == "dashscope":
|
||||
return cls.DASHSCOPE_API_KEY
|
||||
elif provider == "siliconflow":
|
||||
return cls.SILICONFLOW_API_KEY
|
||||
elif provider == "doubao":
|
||||
return cls.DOUBAO_API_KEY
|
||||
elif provider in ["kimi", "moonshot"]:
|
||||
return cls.KIMI_API_KEY or cls.MOONSHOT_API_KEY
|
||||
elif provider == "openrouter":
|
||||
return cls.OPENROUTER_API_KEY
|
||||
return None
|
||||
|
||||
@classmethod
|
||||
def effective_provider(cls) -> str:
|
||||
"""Resolve the provider actually used, applying the OpenRouter fallback.
|
||||
|
||||
Preserves default behavior when the configured provider's key is
|
||||
present. Otherwise, if an OPENROUTER_API_KEY is available, transparently
|
||||
fall back to 'openrouter' so the tools still run with only that key set.
|
||||
"""
|
||||
provider = cls.PROVIDER.lower()
|
||||
provider = {"qwen": "dashscope", "bailian": "dashscope"}.get(provider, provider)
|
||||
if cls.get_api_key(provider):
|
||||
return provider
|
||||
if cls.OPENROUTER_API_KEY:
|
||||
return "openrouter"
|
||||
return provider
|
||||
|
||||
@classmethod
|
||||
def validate(cls) -> None:
|
||||
"""Validate the configuration."""
|
||||
provider = cls.effective_provider()
|
||||
api_key = cls.get_api_key(provider)
|
||||
|
||||
if not api_key:
|
||||
raise ValueError(
|
||||
f"API key required for provider '{cls.PROVIDER.lower()}'. "
|
||||
f"Set one of {cls.PROVIDER.upper()}_API_KEY or OPENROUTER_API_KEY "
|
||||
f"(universal fallback)."
|
||||
)
|
||||
|
||||
@classmethod
|
||||
def get_llm_config(cls) -> dict:
|
||||
"""Get LLM configuration based on provider."""
|
||||
provider = cls.effective_provider()
|
||||
api_key = cls.get_api_key(provider)
|
||||
|
||||
if not api_key:
|
||||
raise ValueError(
|
||||
f"API key not found for provider '{cls.PROVIDER.lower()}'. "
|
||||
f"Set {cls.PROVIDER.upper()}_API_KEY or OPENROUTER_API_KEY."
|
||||
)
|
||||
|
||||
if provider == "dashscope":
|
||||
return {
|
||||
"provider": "dashscope",
|
||||
"api_key": api_key,
|
||||
"base_url": cls.DASHSCOPE_BASE_URL,
|
||||
"model": cls.MODEL or "qwen3.7-plus"
|
||||
}
|
||||
elif provider == "siliconflow":
|
||||
return {
|
||||
"provider": "siliconflow",
|
||||
"api_key": api_key,
|
||||
"base_url": "https://api.siliconflow.cn/v1",
|
||||
"model": cls.MODEL or "Qwen/Qwen3-235B-A22B-Thinking-2507"
|
||||
}
|
||||
elif provider == "doubao":
|
||||
return {
|
||||
"provider": "doubao",
|
||||
"api_key": api_key,
|
||||
"base_url": "https://ark.cn-beijing.volces.com/api/v3",
|
||||
"model": cls.MODEL or "doubao-seed-1-6-thinking-250715"
|
||||
}
|
||||
elif provider in ["kimi", "moonshot"]:
|
||||
return {
|
||||
"provider": "kimi",
|
||||
"api_key": api_key,
|
||||
"base_url": "https://api.moonshot.cn/v1",
|
||||
"model": cls.MODEL or "kimi-k3"
|
||||
}
|
||||
elif provider == "openrouter":
|
||||
return {
|
||||
"provider": "openrouter",
|
||||
"api_key": api_key,
|
||||
"base_url": "https://openrouter.ai/api/v1",
|
||||
"model": cls.MODEL or "google/gemini-3.5-flash"
|
||||
}
|
||||
else:
|
||||
raise ValueError(
|
||||
f"Unsupported provider: {provider}. "
|
||||
f"Use 'dashscope'/'qwen'/'bailian', 'siliconflow', 'doubao', 'kimi', 'moonshot', or 'openrouter'"
|
||||
)
|
||||
|
||||
|
||||
# Note: configuration is validated lazily when the LLM is actually used
|
||||
# (see LLMHelper), so that execution tools which do not require an LLM
|
||||
# (file write, code run, terminal) can be used offline without an API key.
|
||||
@@ -0,0 +1,35 @@
|
||||
# LLM Configuration (for safety checks and summarization)
|
||||
PROVIDER=kimi
|
||||
|
||||
# API Keys (set the one for your provider)
|
||||
DASHSCOPE_API_KEY=your_dashscope_key
|
||||
SILICONFLOW_API_KEY=your_siliconflow_key
|
||||
DOUBAO_API_KEY=your_doubao_key
|
||||
KIMI_API_KEY=your_kimi_key
|
||||
MOONSHOT_API_KEY=your_moonshot_key
|
||||
OPENROUTER_API_KEY=your_openrouter_key
|
||||
|
||||
# Universal OpenRouter fallback:
|
||||
# When the configured PROVIDER's key is missing but OPENROUTER_API_KEY is set,
|
||||
# Config.effective_provider() transparently switches to 'openrouter' so the LLM
|
||||
# steps (approval, summarization, error/syntax analysis) still run.
|
||||
# For OpenRouter, MODEL should be a provider/model id, e.g. openai/gpt-5.6-luna.
|
||||
|
||||
# Model (optional, defaults to provider's default model)
|
||||
# MODEL=kimi-k3
|
||||
# DashScope/Bailian defaults to qwen3.7-plus; set DASHSCOPE_BASE_URL to
|
||||
# https://dashscope-intl.aliyuncs.com/compatible-mode/v1 for international keys.
|
||||
|
||||
# Model parameters
|
||||
TEMPERATURE=0.7
|
||||
MAX_TOKENS=4096
|
||||
|
||||
# External Services (optional)
|
||||
GOOGLE_CALENDAR_CREDENTIALS_FILE=credentials.json
|
||||
GITHUB_TOKEN=your_github_token_here
|
||||
|
||||
# Safety Settings
|
||||
REQUIRE_APPROVAL_FOR_DANGEROUS_OPS=true
|
||||
AUTO_SUMMARIZE_COMPLEX_OUTPUT=true
|
||||
AUTO_VERIFY_CODE=true
|
||||
MAX_OUTPUT_LENGTH=1000
|
||||
@@ -0,0 +1,160 @@
|
||||
"""Example usage of the execution tools MCP server."""
|
||||
|
||||
import asyncio
|
||||
from llm_helper import LLMHelper
|
||||
from file_tools import FileTools
|
||||
from execution_tools import ExecutionTools
|
||||
from external_tools import ExternalTools
|
||||
|
||||
|
||||
async def example_file_operations():
|
||||
"""Example: File operations with verification."""
|
||||
print("=== File Operations Example ===\n")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
file_tools = FileTools(llm_helper)
|
||||
|
||||
# Write a Python file
|
||||
print("1. Writing a Python file...")
|
||||
result = await file_tools.write_file(
|
||||
path="test_script.py",
|
||||
content="""def greet(name):
|
||||
print(f"Hello, {name}!")
|
||||
|
||||
if __name__ == "__main__":
|
||||
greet("World")
|
||||
""",
|
||||
overwrite=True
|
||||
)
|
||||
print(f"Result: {result}\n")
|
||||
|
||||
# Edit the file
|
||||
print("2. Editing the file...")
|
||||
result = await file_tools.edit_file(
|
||||
path="test_script.py",
|
||||
search='greet("World")',
|
||||
replace='greet("MCP Server")'
|
||||
)
|
||||
print(f"Result: {result}\n")
|
||||
|
||||
|
||||
async def example_code_interpreter():
|
||||
"""Example: Code interpreter with analysis."""
|
||||
print("=== Code Interpreter Example ===\n")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
execution_tools = ExecutionTools(llm_helper)
|
||||
|
||||
# Execute valid code
|
||||
print("1. Executing valid code...")
|
||||
result = await execution_tools.code_interpreter(
|
||||
code="""
|
||||
import math
|
||||
|
||||
# Calculate factorial
|
||||
def factorial(n):
|
||||
if n <= 1:
|
||||
return 1
|
||||
return n * factorial(n - 1)
|
||||
|
||||
print(f"Factorial of 5: {factorial(5)}")
|
||||
print(f"Square root of 16: {math.sqrt(16)}")
|
||||
"""
|
||||
)
|
||||
print(f"Result: {result}\n")
|
||||
|
||||
# Execute code with error
|
||||
print("2. Executing code with error...")
|
||||
result = await execution_tools.code_interpreter(
|
||||
code="""
|
||||
# This will cause an error
|
||||
x = 10 / 0
|
||||
"""
|
||||
)
|
||||
print(f"Result: {result}\n")
|
||||
|
||||
|
||||
async def example_virtual_terminal():
|
||||
"""Example: Virtual terminal."""
|
||||
print("=== Virtual Terminal Example ===\n")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
execution_tools = ExecutionTools(llm_helper)
|
||||
|
||||
# Execute simple command
|
||||
print("1. Listing current directory...")
|
||||
result = await execution_tools.virtual_terminal(
|
||||
command="ls -la"
|
||||
)
|
||||
print(f"Result: {result}\n")
|
||||
|
||||
# Execute command with error
|
||||
print("2. Executing command that fails...")
|
||||
result = await execution_tools.virtual_terminal(
|
||||
command="cat nonexistent_file.txt"
|
||||
)
|
||||
print(f"Result: {result}\n")
|
||||
|
||||
|
||||
async def example_google_calendar():
|
||||
"""Example: Google Calendar integration."""
|
||||
print("=== Google Calendar Example ===\n")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
external_tools = ExternalTools(llm_helper)
|
||||
|
||||
print("Adding event to Google Calendar...")
|
||||
result = await external_tools.google_calendar_add(
|
||||
summary="Team Meeting",
|
||||
start_time="2025-10-01T10:00:00",
|
||||
end_time="2025-10-01T11:00:00",
|
||||
description="Quarterly planning meeting",
|
||||
location="Conference Room A"
|
||||
)
|
||||
print(f"Result: {result}\n")
|
||||
|
||||
|
||||
async def example_github_pr():
|
||||
"""Example: GitHub Pull Request creation."""
|
||||
print("=== GitHub PR Example ===\n")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
external_tools = ExternalTools(llm_helper)
|
||||
|
||||
print("Creating GitHub Pull Request...")
|
||||
result = await external_tools.github_create_pr(
|
||||
repo_name="owner/repository",
|
||||
title="Add new feature",
|
||||
body="This PR adds a new feature to improve performance.\n\n## Changes\n- Optimized algorithm\n- Added tests\n- Updated documentation",
|
||||
head_branch="feature/new-feature",
|
||||
base_branch="main"
|
||||
)
|
||||
print(f"Result: {result}\n")
|
||||
|
||||
|
||||
async def main():
|
||||
"""Run all examples."""
|
||||
try:
|
||||
# File operations
|
||||
await example_file_operations()
|
||||
|
||||
# Code interpreter
|
||||
await example_code_interpreter()
|
||||
|
||||
# Virtual terminal
|
||||
await example_virtual_terminal()
|
||||
|
||||
# External tools (commented out as they require credentials)
|
||||
# await example_google_calendar()
|
||||
# await example_github_pr()
|
||||
|
||||
print("All examples completed!")
|
||||
|
||||
except Exception as e:
|
||||
print(f"Error running examples: {e}")
|
||||
import traceback
|
||||
traceback.print_exc()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
asyncio.run(main())
|
||||
@@ -0,0 +1,280 @@
|
||||
"""Generic execution tools: code interpreter and virtual terminal."""
|
||||
|
||||
import os
|
||||
import subprocess
|
||||
import sys
|
||||
import io
|
||||
import tempfile
|
||||
import traceback
|
||||
from typing import Dict, Any, Optional, Tuple
|
||||
from contextlib import redirect_stdout, redirect_stderr
|
||||
from llm_helper import LLMHelper
|
||||
from config import Config
|
||||
from multilang_executor import LanguageExecutor, ExecutionStatus
|
||||
|
||||
# Long-output handling thresholds (see "长输出的截断与持久化" in chapter 4).
|
||||
# When output exceeds either threshold, keep the head and tail few lines in the
|
||||
# context and persist the full output to a temp file for later retrieval.
|
||||
MAX_OUTPUT_LINES = 200
|
||||
MAX_OUTPUT_CHARS = 10000
|
||||
HEAD_LINES = 50
|
||||
TAIL_LINES = 50
|
||||
|
||||
|
||||
def truncate_and_persist(
|
||||
text: str,
|
||||
tool_name: str = "execution",
|
||||
max_lines: int = MAX_OUTPUT_LINES,
|
||||
max_chars: int = MAX_OUTPUT_CHARS,
|
||||
head_lines: int = HEAD_LINES,
|
||||
tail_lines: int = TAIL_LINES,
|
||||
) -> Tuple[str, Optional[str]]:
|
||||
"""Truncate over-long output and persist the full text to a temp file.
|
||||
|
||||
Returns a tuple of (processed_text, saved_path). When the output is within
|
||||
both thresholds, it is returned unchanged with ``saved_path`` set to None.
|
||||
Otherwise only the first ``head_lines`` and last ``tail_lines`` lines are
|
||||
kept in context, with a middle marker pointing to the saved file. This
|
||||
keeps the agent's context bounded without discarding any information and
|
||||
requires no LLM call.
|
||||
"""
|
||||
if text is None:
|
||||
return text, None
|
||||
|
||||
lines = text.split("\n")
|
||||
if len(text) <= max_chars and len(lines) <= max_lines:
|
||||
return text, None
|
||||
|
||||
# Persist the complete output for later retrieval via read_file.
|
||||
fd, path = tempfile.mkstemp(prefix=f"{tool_name}_output_", suffix=".txt")
|
||||
with os.fdopen(fd, "w", encoding="utf-8") as f:
|
||||
f.write(text)
|
||||
|
||||
# lines[-0:] is the whole list in Python; treat 0 as "keep no tail".
|
||||
head_n = max(0, head_lines)
|
||||
tail_n = max(0, tail_lines)
|
||||
head_part = lines[:head_n] if head_n else []
|
||||
tail_part = lines[-tail_n:] if tail_n else []
|
||||
omitted = max(len(lines) - head_n - tail_n, 0)
|
||||
|
||||
guide = f"[如需完整输出，请使用 read_file 工具读取 {path}]"
|
||||
if omitted == 0:
|
||||
# Head+tail cover the file; do not concatenate overlapping slices.
|
||||
truncated = "\n".join(lines + [guide])
|
||||
else:
|
||||
middle = f"... [省略 {omitted} 行，完整输出已保存至 {path}] ..."
|
||||
truncated = "\n".join(head_part + [middle] + tail_part + [guide])
|
||||
return truncated, path
|
||||
|
||||
|
||||
class ExecutionTools:
|
||||
"""Generic execution tools with safety and result analysis."""
|
||||
|
||||
def __init__(self, llm_helper: LLMHelper):
|
||||
"""Initialize execution tools with LLM helper."""
|
||||
self.llm_helper = llm_helper
|
||||
self.lang_executor = LanguageExecutor(workspace_dir=Config.WORKSPACE_DIR)
|
||||
|
||||
async def code_interpreter(
|
||||
self,
|
||||
code: str,
|
||||
language: str = "python",
|
||||
timeout: float = 30.0,
|
||||
stdin: Optional[str] = None,
|
||||
files: Optional[Dict[str, str]] = None
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Execute code in a sandboxed environment with multi-language support.
|
||||
|
||||
Args:
|
||||
code: Code to execute
|
||||
language: Programming language (python, javascript, typescript, go, java, cpp, rust, php, bash)
|
||||
timeout: Execution timeout in seconds
|
||||
stdin: Optional stdin input
|
||||
files: Optional additional files
|
||||
|
||||
Returns:
|
||||
Result dictionary with output and analysis
|
||||
"""
|
||||
if language is None:
|
||||
language = "python"
|
||||
language = language.lower()
|
||||
|
||||
# Verify syntax first (only for Python for now)
|
||||
if Config.AUTO_VERIFY_CODE and language in ['python', 'python3']:
|
||||
is_valid, error_msg = self.llm_helper.verify_code_syntax(code, language)
|
||||
if not is_valid:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Syntax error: {error_msg}",
|
||||
"verification": "failed",
|
||||
"language": language
|
||||
}
|
||||
|
||||
# Check for dangerous operations
|
||||
if Config.REQUIRE_APPROVAL_FOR_DANGEROUS_OPS:
|
||||
dangerous_patterns = {
|
||||
'python': ['os.system', 'subprocess', 'eval', 'exec', 'open(', '__import__', 'compile'],
|
||||
'bash': ['rm -rf', 'dd if=', 'mkfs', '> /dev/', 'curl', 'wget'],
|
||||
'php': ['exec(', 'system(', 'shell_exec(', 'passthru(', 'eval('],
|
||||
}
|
||||
|
||||
patterns = dangerous_patterns.get(language, [])
|
||||
detected = [p for p in patterns if p in code]
|
||||
|
||||
if detected:
|
||||
approved, reason = self.llm_helper.request_approval(
|
||||
"code_execution",
|
||||
{
|
||||
"code": code,
|
||||
"language": language,
|
||||
"detected_patterns": detected
|
||||
}
|
||||
)
|
||||
|
||||
if not approved:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Execution not approved: {reason}",
|
||||
"language": language
|
||||
}
|
||||
|
||||
# Execute code using multi-language executor
|
||||
try:
|
||||
result = await self.lang_executor.execute_code(
|
||||
code=code,
|
||||
language=language,
|
||||
timeout=timeout,
|
||||
stdin=stdin,
|
||||
files=files
|
||||
)
|
||||
|
||||
# Convert status to success flag
|
||||
success = result.get('status') == ExecutionStatus.SUCCESS
|
||||
|
||||
# Long outputs: truncate head/tail and persist the full text to a
|
||||
# temp file (offline-safe), then optionally LLM-summarize whatever
|
||||
# still exceeds the char threshold.
|
||||
stdout = result.get('stdout', '')
|
||||
stderr = result.get('stderr', '')
|
||||
stdout, stdout_file = truncate_and_persist(stdout, "code_interpreter")
|
||||
stderr, stderr_file = truncate_and_persist(stderr, "code_interpreter")
|
||||
|
||||
if Config.AUTO_SUMMARIZE_COMPLEX_OUTPUT and len(stdout) > MAX_OUTPUT_CHARS:
|
||||
stdout = self.llm_helper.summarize_output("code_interpreter", stdout)
|
||||
if Config.AUTO_SUMMARIZE_COMPLEX_OUTPUT and len(stderr) > MAX_OUTPUT_CHARS:
|
||||
stderr = self.llm_helper.summarize_output("code_interpreter", stderr)
|
||||
|
||||
return {
|
||||
"success": success,
|
||||
"status": result.get('status'),
|
||||
"language": result.get('language', language),
|
||||
"stdout": stdout,
|
||||
"stderr": stderr,
|
||||
"stdout_file": stdout_file,
|
||||
"stderr_file": stderr_file,
|
||||
"returncode": result.get('returncode'),
|
||||
"error": result.get('error'),
|
||||
"compile_output": result.get('compile_output'),
|
||||
"phase": result.get('phase'),
|
||||
"execution_time": result.get('execution_time'),
|
||||
"sandbox": result.get('sandbox'),
|
||||
"verification": "passed" if Config.AUTO_VERIFY_CODE else "skipped"
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
error_output = f"{type(e).__name__}: {str(e)}\n{traceback.format_exc()}"
|
||||
return {
|
||||
"success": False,
|
||||
"error": error_output,
|
||||
"language": language
|
||||
}
|
||||
|
||||
async def virtual_terminal(
|
||||
self,
|
||||
command: str,
|
||||
timeout: int = 30
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Execute shell command in a virtual terminal.
|
||||
|
||||
Args:
|
||||
command: Shell command to execute
|
||||
timeout: Timeout in seconds
|
||||
|
||||
Returns:
|
||||
Result dictionary with output and analysis
|
||||
"""
|
||||
# Check for dangerous commands
|
||||
if Config.REQUIRE_APPROVAL_FOR_DANGEROUS_OPS:
|
||||
dangerous_commands = [
|
||||
'rm -rf', 'dd', 'mkfs', 'format',
|
||||
'> /dev/', 'chmod -R', 'chown -R'
|
||||
]
|
||||
|
||||
if any(dangerous in command for dangerous in dangerous_commands):
|
||||
approved, reason = self.llm_helper.request_approval(
|
||||
"terminal_command",
|
||||
{
|
||||
"command": command,
|
||||
"detected_patterns": [p for p in dangerous_commands if p in command]
|
||||
}
|
||||
)
|
||||
|
||||
if not approved:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Command execution not approved: {reason}"
|
||||
}
|
||||
|
||||
# Execute command
|
||||
try:
|
||||
result = subprocess.run(
|
||||
command,
|
||||
shell=True,
|
||||
capture_output=True,
|
||||
text=True,
|
||||
timeout=timeout,
|
||||
cwd=Config.WORKSPACE_DIR
|
||||
)
|
||||
|
||||
stdout = result.stdout
|
||||
stderr = result.stderr
|
||||
|
||||
# Long output: truncate head/tail and persist to a temp file, then
|
||||
# optionally LLM-summarize whatever still exceeds the char threshold.
|
||||
stdout, stdout_file = truncate_and_persist(stdout, "virtual_terminal")
|
||||
stderr, stderr_file = truncate_and_persist(stderr, "virtual_terminal")
|
||||
|
||||
if Config.AUTO_SUMMARIZE_COMPLEX_OUTPUT:
|
||||
if len(stdout) > MAX_OUTPUT_CHARS:
|
||||
stdout = self.llm_helper.summarize_output(
|
||||
"virtual_terminal",
|
||||
stdout
|
||||
)
|
||||
if len(stderr) > MAX_OUTPUT_CHARS:
|
||||
stderr = self.llm_helper.summarize_output(
|
||||
"virtual_terminal",
|
||||
stderr
|
||||
)
|
||||
|
||||
response = {
|
||||
"success": result.returncode == 0,
|
||||
"returncode": result.returncode,
|
||||
"stdout": stdout,
|
||||
"stderr": stderr,
|
||||
"stdout_file": stdout_file,
|
||||
"stderr_file": stderr_file
|
||||
}
|
||||
return response
|
||||
|
||||
except subprocess.TimeoutExpired:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Command timed out after {timeout} seconds"
|
||||
}
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Command execution failed: {str(e)}"
|
||||
}
|
||||
@@ -0,0 +1,20 @@
|
||||
{
|
||||
"experiment": "4-3",
|
||||
"authority": "book/chapter4.md:274",
|
||||
"required_categories": {
|
||||
"file_write_edit": ["python_linter", "javascript_linter", "structured_errors"],
|
||||
"terminal": ["timeout", "dangerous_command_review", "history_or_receipt"],
|
||||
"code_interpreter": ["real_sandbox", "dangerous_operation_gate", "long_output_persisted"],
|
||||
"data": ["excel_write", "formula", "screenshot"],
|
||||
"external": ["calendar", "github_pr", "email", "webhook"],
|
||||
"gui": ["browser", "virtual_desktop", "virtual_mobile"]
|
||||
},
|
||||
"safety": {
|
||||
"workspace_confinement": true,
|
||||
"automatic_linter": true,
|
||||
"llm_driven_danger_review": true,
|
||||
"long_output_head_tail_and_full_file": true,
|
||||
"credential_free_receipts": true
|
||||
},
|
||||
"completion_rule": "Every named manuscript category must have substantive real execution evidence; missing credentials or active GUI backends produce blocked, never passed."
|
||||
}
|
||||
@@ -0,0 +1,388 @@
|
||||
"""Real data, webhook, and browser execution tools for Experiment 4-3."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import hashlib
|
||||
import json
|
||||
import os
|
||||
import re
|
||||
import signal
|
||||
import shutil
|
||||
import subprocess
|
||||
import tempfile
|
||||
import time
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
import httpx
|
||||
from config import Config
|
||||
|
||||
|
||||
def _sha(path: Path) -> str:
|
||||
return hashlib.sha256(path.read_bytes()).hexdigest()
|
||||
|
||||
|
||||
def _safe_output(path: str) -> Path:
|
||||
candidate = Path(path)
|
||||
if not candidate.is_absolute():
|
||||
candidate = Path(Config.WORKSPACE_DIR) / candidate
|
||||
candidate = candidate.resolve()
|
||||
candidate.relative_to(Path(Config.WORKSPACE_DIR).resolve())
|
||||
candidate.parent.mkdir(parents=True, exist_ok=True)
|
||||
return candidate
|
||||
|
||||
|
||||
class ExtendedTools:
|
||||
async def excel_create_with_formula_and_screenshot(
|
||||
self, output_path: str, rows: list[dict[str, Any]]
|
||||
) -> dict[str, Any]:
|
||||
"""Create a real XLSX, apply formulas, and render a screenshot via LibreOffice."""
|
||||
from openpyxl import Workbook
|
||||
|
||||
target = _safe_output(output_path)
|
||||
workbook = Workbook()
|
||||
sheet = workbook.active
|
||||
sheet.title = "Invoice"
|
||||
sheet.append(["Item", "Quantity", "Unit price", "Total"])
|
||||
for index, row in enumerate(rows, 2):
|
||||
sheet.append([row["item"], float(row["quantity"]), float(row["unit_price"]),
|
||||
f"=B{index}*C{index}"])
|
||||
total_row = len(rows) + 2
|
||||
sheet.cell(total_row, 3, "Grand total")
|
||||
sheet.cell(total_row, 4, f"=SUM(D2:D{total_row - 1})")
|
||||
sheet.freeze_panes = "A2"
|
||||
sheet.column_dimensions["A"].width = 28
|
||||
for column in ("B", "C", "D"):
|
||||
sheet.column_dimensions[column].width = 16
|
||||
workbook.save(target)
|
||||
|
||||
soffice = shutil.which("soffice") or shutil.which("libreoffice")
|
||||
if not soffice:
|
||||
return {"success": False, "error": "LibreOffice is required for formula rendering"}
|
||||
started = time.perf_counter()
|
||||
process = subprocess.run(
|
||||
[soffice, "--headless", "--convert-to", "pdf", "--outdir",
|
||||
str(target.parent), str(target)],
|
||||
stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, timeout=120,
|
||||
)
|
||||
pdf = target.with_suffix(".pdf")
|
||||
if process.returncode != 0 or not pdf.is_file():
|
||||
return {"success": False, "error": process.stderr or process.stdout,
|
||||
"returncode": process.returncode}
|
||||
import fitz
|
||||
|
||||
document = fitz.open(pdf)
|
||||
screenshot = target.with_suffix(".png")
|
||||
document[0].get_pixmap(matrix=fitz.Matrix(1.5, 1.5), alpha=False).save(screenshot)
|
||||
document.close()
|
||||
return {
|
||||
"success": True,
|
||||
"xlsx": {"path": str(target), "bytes": target.stat().st_size, "sha256": _sha(target)},
|
||||
"pdf": {"path": str(pdf), "bytes": pdf.stat().st_size, "sha256": _sha(pdf)},
|
||||
"screenshot": {"path": str(screenshot), "bytes": screenshot.stat().st_size,
|
||||
"sha256": _sha(screenshot)},
|
||||
"formula_cells": [f"D{index}" for index in range(2, total_row + 1)],
|
||||
"rows": len(rows),
|
||||
"renderer": "LibreOffice headless + PyMuPDF",
|
||||
"latency_seconds": round(time.perf_counter() - started, 3),
|
||||
}
|
||||
|
||||
async def webhook_post(self, url: str, payload: dict[str, Any]) -> dict[str, Any]:
|
||||
"""POST JSON to a real HTTPS webhook and retain response evidence."""
|
||||
if not url.startswith("https://"):
|
||||
return {"success": False, "error": "Only HTTPS webhook URLs are allowed"}
|
||||
started = time.perf_counter()
|
||||
async with httpx.AsyncClient(timeout=30, follow_redirects=True) as client:
|
||||
response = await client.post(url, json=payload)
|
||||
try:
|
||||
body = response.json()
|
||||
except ValueError:
|
||||
body = {"text": response.text[:2000]}
|
||||
return {
|
||||
"success": response.is_success,
|
||||
"status": response.status_code,
|
||||
"url": str(response.url),
|
||||
"response": body,
|
||||
"response_sha256": hashlib.sha256(response.content).hexdigest(),
|
||||
"response_bytes": len(response.content),
|
||||
"latency_seconds": round(time.perf_counter() - started, 3),
|
||||
}
|
||||
|
||||
async def browser_navigate(self, url: str, screenshot_path: str) -> dict[str, Any]:
|
||||
"""Navigate with real headless Chromium, extract content, and retain pixels."""
|
||||
if not url.startswith("https://"):
|
||||
return {"success": False, "error": "Only HTTPS URLs are allowed"}
|
||||
target = _safe_output(screenshot_path)
|
||||
started = time.perf_counter()
|
||||
from playwright.async_api import async_playwright
|
||||
|
||||
async with async_playwright() as playwright:
|
||||
browser = await playwright.chromium.launch(headless=True)
|
||||
page = await browser.new_page(viewport={"width": 1280, "height": 720})
|
||||
response = await page.goto(url, wait_until="networkidle", timeout=60000)
|
||||
title = await page.title()
|
||||
text = (await page.locator("body").inner_text())[:4000]
|
||||
await page.screenshot(path=str(target), full_page=True)
|
||||
await browser.close()
|
||||
return {
|
||||
"success": bool(response and response.ok and target.is_file()),
|
||||
"url": url,
|
||||
"status": response.status if response else None,
|
||||
"title": title,
|
||||
"body_text": text,
|
||||
"screenshot": {"path": str(target), "bytes": target.stat().st_size,
|
||||
"sha256": _sha(target)},
|
||||
"browser": "Chromium via Playwright",
|
||||
"latency_seconds": round(time.perf_counter() - started, 3),
|
||||
}
|
||||
|
||||
async def virtual_desktop_execute(
|
||||
self, url: str, screenshot_path: str, expected_title: str | None = None
|
||||
) -> dict[str, Any]:
|
||||
"""Drive headful Chromium through X11 keyboard events and retain pixels."""
|
||||
if not url.startswith("https://"):
|
||||
return {"success": False, "error": "Only HTTPS URLs are allowed"}
|
||||
target = _safe_output(screenshot_path)
|
||||
required = {
|
||||
name: shutil.which(name)
|
||||
for name in ("Xvfb", "xdotool", "ffmpeg")
|
||||
}
|
||||
chromium = shutil.which("chromium") or shutil.which("chromium-browser")
|
||||
missing = [name for name, path in required.items() if not path]
|
||||
if not chromium:
|
||||
missing.append("chromium")
|
||||
if missing:
|
||||
return {"success": False, "error": f"Missing desktop executables: {missing}"}
|
||||
|
||||
display_number = next((
|
||||
number for number in range(90, 130)
|
||||
if not Path(f"/tmp/.X11-unix/X{number}").exists()
|
||||
and not Path(f"/tmp/.X{number}-lock").exists()
|
||||
), None)
|
||||
if display_number is None:
|
||||
return {"success": False, "error": "No free bounded X11 display number"}
|
||||
display = f":{display_number}"
|
||||
started = time.perf_counter()
|
||||
xvfb_process: subprocess.Popen[bytes] | None = None
|
||||
chromium_process: subprocess.Popen[bytes] | None = None
|
||||
|
||||
def stop(process: subprocess.Popen[bytes] | None) -> None:
|
||||
if process is None or process.poll() is not None:
|
||||
return
|
||||
try:
|
||||
os.killpg(process.pid, signal.SIGTERM)
|
||||
process.wait(timeout=5)
|
||||
except (ProcessLookupError, subprocess.TimeoutExpired):
|
||||
try:
|
||||
os.killpg(process.pid, signal.SIGKILL)
|
||||
except ProcessLookupError:
|
||||
pass
|
||||
|
||||
try:
|
||||
xvfb_process = subprocess.Popen(
|
||||
[required["Xvfb"], display, "-screen", "0", "1280x720x24", "-nolisten", "tcp"],
|
||||
stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL,
|
||||
start_new_session=True,
|
||||
)
|
||||
socket_path = Path(f"/tmp/.X11-unix/X{display_number}")
|
||||
for _ in range(50):
|
||||
if socket_path.exists():
|
||||
break
|
||||
if xvfb_process.poll() is not None:
|
||||
return {"success": False, "error": "Xvfb exited before creating its socket"}
|
||||
time.sleep(0.1)
|
||||
else:
|
||||
return {"success": False, "error": "Xvfb did not become ready"}
|
||||
|
||||
env = {**os.environ, "DISPLAY": display}
|
||||
with tempfile.TemporaryDirectory(prefix="exp4-computer-use-") as profile:
|
||||
chromium_process = subprocess.Popen(
|
||||
[chromium, "--no-sandbox", "--disable-gpu", "--disable-dev-shm-usage",
|
||||
f"--user-data-dir={profile}", "about:blank"],
|
||||
stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL,
|
||||
env=env, start_new_session=True,
|
||||
)
|
||||
window_id = ""
|
||||
for _ in range(100):
|
||||
search = subprocess.run(
|
||||
[required["xdotool"], "search", "--onlyvisible", "--class", "chromium"],
|
||||
stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True, env=env,
|
||||
)
|
||||
if search.stdout.strip():
|
||||
window_id = search.stdout.splitlines()[0].strip()
|
||||
break
|
||||
if chromium_process.poll() is not None:
|
||||
return {"success": False, "error": "Chromium exited before opening a window"}
|
||||
time.sleep(0.1)
|
||||
if not window_id:
|
||||
return {"success": False, "error": "No visible Chromium window appeared"}
|
||||
|
||||
subprocess.run(
|
||||
[required["xdotool"], "windowfocus", "--sync", window_id],
|
||||
stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL, env=env,
|
||||
)
|
||||
input_receipts = []
|
||||
for command in (
|
||||
[required["xdotool"], "key", "--window", window_id, "ctrl+l"],
|
||||
[required["xdotool"], "type", "--window", window_id, "--delay", "15", url],
|
||||
[required["xdotool"], "key", "--window", window_id, "Return"],
|
||||
):
|
||||
completed = subprocess.run(
|
||||
command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, env=env,
|
||||
)
|
||||
input_receipts.append({"operation": command[1], "returncode": completed.returncode})
|
||||
if completed.returncode != 0:
|
||||
return {"success": False, "error": completed.stderr.strip(),
|
||||
"input_receipts": input_receipts}
|
||||
|
||||
title = ""
|
||||
for _ in range(100):
|
||||
title_result = subprocess.run(
|
||||
[required["xdotool"], "getwindowname", window_id],
|
||||
stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True, env=env,
|
||||
)
|
||||
title = title_result.stdout.strip()
|
||||
if title and (not expected_title or expected_title in title):
|
||||
break
|
||||
time.sleep(0.1)
|
||||
title_matched = bool(title and (not expected_title or expected_title in title))
|
||||
capture = subprocess.run(
|
||||
[required["ffmpeg"], "-nostdin", "-loglevel", "error", "-f", "x11grab",
|
||||
"-video_size", "1280x720", "-i", display, "-frames:v", "1", "-y", str(target)],
|
||||
stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, env=env, timeout=30,
|
||||
)
|
||||
png_valid = target.is_file() and target.read_bytes().startswith(b"\x89PNG\r\n\x1a\n")
|
||||
return {
|
||||
"success": capture.returncode == 0 and png_valid and title_matched,
|
||||
"backend": "Xvfb + headful Chromium + xdotool",
|
||||
"versions": {
|
||||
"chromium": subprocess.run(
|
||||
[chromium, "--version"], stdout=subprocess.PIPE,
|
||||
stderr=subprocess.STDOUT, text=True,
|
||||
).stdout.strip(),
|
||||
"xdotool": subprocess.run(
|
||||
[required["xdotool"], "-v"], stdout=subprocess.PIPE,
|
||||
stderr=subprocess.STDOUT, text=True,
|
||||
).stdout.strip(),
|
||||
"ffmpeg": subprocess.run(
|
||||
[required["ffmpeg"], "-version"], stdout=subprocess.PIPE,
|
||||
stderr=subprocess.STDOUT, text=True,
|
||||
).stdout.splitlines()[0],
|
||||
},
|
||||
"display": display,
|
||||
"window_id": window_id,
|
||||
"url_entered_via_os_keyboard": url,
|
||||
"window_title": title,
|
||||
"expected_title": expected_title,
|
||||
"expected_title_matched": title_matched,
|
||||
"input_receipts": input_receipts,
|
||||
"screenshot": ({"path": str(target), "bytes": target.stat().st_size,
|
||||
"sha256": _sha(target)} if png_valid else None),
|
||||
"capture_returncode": capture.returncode,
|
||||
"capture_error": capture.stderr.strip() or None,
|
||||
"latency_seconds": round(time.perf_counter() - started, 3),
|
||||
}
|
||||
finally:
|
||||
stop(chromium_process)
|
||||
stop(xvfb_process)
|
||||
|
||||
async def virtual_mobile_execute(
|
||||
self, container_name: str, screenshot_path: str
|
||||
) -> dict[str, Any]:
|
||||
"""Operate a real AndroidWorld emulator through ADB inside its container."""
|
||||
if not re.fullmatch(r"[A-Za-z0-9_.-]{1,128}", container_name):
|
||||
return {"success": False, "error": "Invalid Docker container name"}
|
||||
if not shutil.which("docker"):
|
||||
return {"success": False, "error": "Docker is required"}
|
||||
target = _safe_output(screenshot_path)
|
||||
started = time.perf_counter()
|
||||
|
||||
def adb(*arguments: str, binary: bool = False) -> subprocess.CompletedProcess[Any]:
|
||||
return subprocess.run(
|
||||
["docker", "exec", container_name, "adb", *arguments],
|
||||
stdout=subprocess.PIPE, stderr=subprocess.PIPE,
|
||||
text=not binary, timeout=30,
|
||||
)
|
||||
|
||||
inspect = subprocess.run(
|
||||
["docker", "inspect", "--format", "{{.State.Running}} {{.Image}}", container_name],
|
||||
stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True,
|
||||
)
|
||||
inspect_parts = inspect.stdout.strip().split(maxsplit=1)
|
||||
if inspect.returncode != 0 or not inspect_parts or inspect_parts[0] != "true":
|
||||
return {"success": False, "error": "AndroidWorld container is not running",
|
||||
"container": container_name}
|
||||
boot = adb("shell", "getprop", "sys.boot_completed")
|
||||
devices = adb("devices", "-l")
|
||||
if boot.stdout.strip() != "1" or " device " not in f" {devices.stdout} ":
|
||||
return {"success": False, "error": "Android emulator is not boot-complete",
|
||||
"container": container_name, "devices": devices.stdout.strip()}
|
||||
|
||||
model = adb("shell", "getprop", "ro.product.model").stdout.strip()
|
||||
sdk = adb("shell", "getprop", "ro.build.version.sdk").stdout.strip()
|
||||
focus_before = adb("shell", "dumpsys", "window").stdout
|
||||
launch = adb("shell", "am", "start", "-W", "-a", "android.settings.WIFI_SETTINGS")
|
||||
focus_settings = adb("shell", "dumpsys", "window").stdout
|
||||
screenshot = adb("exec-out", "screencap", "-p", binary=True)
|
||||
if screenshot.returncode == 0:
|
||||
target.write_bytes(screenshot.stdout)
|
||||
home = adb("shell", "input", "keyevent", "KEYCODE_HOME")
|
||||
focus_home = adb("shell", "dumpsys", "window").stdout
|
||||
png_valid = target.is_file() and target.read_bytes().startswith(b"\x89PNG\r\n\x1a\n")
|
||||
settings_focused = "com.android.settings" in focus_settings
|
||||
launcher_focused = "launcher" in focus_home.lower()
|
||||
return {
|
||||
"success": all((launch.returncode == 0, settings_focused, home.returncode == 0,
|
||||
launcher_focused, png_valid)),
|
||||
"backend": "AndroidWorld Docker emulator + ADB",
|
||||
"container": container_name,
|
||||
"container_image_id": inspect_parts[1] if len(inspect_parts) > 1 else None,
|
||||
"devices": devices.stdout.strip().splitlines(),
|
||||
"boot_completed": boot.stdout.strip(),
|
||||
"model": model,
|
||||
"api_level": sdk,
|
||||
"focus_before": next((line.strip() for line in focus_before.splitlines()
|
||||
if "mCurrentFocus=" in line), None),
|
||||
"settings_launch_returncode": launch.returncode,
|
||||
"settings_activity": next((line.strip() for line in launch.stdout.splitlines()
|
||||
if line.strip().startswith("Activity:")), None),
|
||||
"settings_focus": next((line.strip() for line in focus_settings.splitlines()
|
||||
if "mCurrentFocus=" in line), None),
|
||||
"home_input_returncode": home.returncode,
|
||||
"home_focus": next((line.strip() for line in focus_home.splitlines()
|
||||
if "mCurrentFocus=" in line), None),
|
||||
"screenshot": ({"path": str(target), "bytes": target.stat().st_size,
|
||||
"sha256": _sha(target)} if png_valid else None),
|
||||
"latency_seconds": round(time.perf_counter() - started, 3),
|
||||
}
|
||||
|
||||
async def environment_capabilities(self) -> dict[str, Any]:
|
||||
"""Report, without simulation, whether desktop/mobile backends are actually usable."""
|
||||
docker_image = subprocess.run(
|
||||
["docker", "image", "inspect",
|
||||
"ghcr.io/anthropics/anthropic-quickstarts:computer-use-demo-latest"],
|
||||
stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL,
|
||||
).returncode == 0 if shutil.which("docker") else False
|
||||
android_container = os.getenv("ANDROID_WORLD_CONTAINER", "")
|
||||
active_devices: list[str] = []
|
||||
if android_container and re.fullmatch(r"[A-Za-z0-9_.-]{1,128}", android_container):
|
||||
devices = subprocess.run(
|
||||
["docker", "exec", android_container, "adb", "devices", "-l"],
|
||||
stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True,
|
||||
) if shutil.which("docker") else None
|
||||
if devices and devices.returncode == 0:
|
||||
active_devices = [line for line in devices.stdout.splitlines()[1:]
|
||||
if " device " in f" {line} "]
|
||||
return {
|
||||
"success": True,
|
||||
"computer_use_container_image_present": docker_image,
|
||||
"computer_use_host_stack_present": all(shutil.which(name) for name in
|
||||
("Xvfb", "xdotool", "ffmpeg"))
|
||||
and bool(shutil.which("chromium") or
|
||||
shutil.which("chromium-browser")),
|
||||
"computer_use_active_session": False,
|
||||
"android_world_container": android_container or None,
|
||||
"android_world_adb_present": bool(active_devices),
|
||||
"android_active_devices": active_devices,
|
||||
"note": "Availability probe only; execution gates are established by the dedicated desktop and mobile action receipts.",
|
||||
}
|
||||
@@ -0,0 +1,323 @@
|
||||
"""External system integration tools: Google Calendar and GitHub."""
|
||||
|
||||
import os
|
||||
import json
|
||||
from typing import Dict, Any, Optional
|
||||
from datetime import datetime, timedelta
|
||||
from pathlib import Path
|
||||
|
||||
from llm_helper import LLMHelper
|
||||
from config import Config
|
||||
|
||||
# Google Calendar imports (optional)
|
||||
try:
|
||||
from google.oauth2.credentials import Credentials
|
||||
from google_auth_oauthlib.flow import InstalledAppFlow
|
||||
from google.auth.transport.requests import Request
|
||||
from googleapiclient.discovery import build
|
||||
GOOGLE_AVAILABLE = True
|
||||
except ImportError:
|
||||
GOOGLE_AVAILABLE = False
|
||||
|
||||
# GitHub imports (optional)
|
||||
try:
|
||||
from github import Github, GithubException
|
||||
GITHUB_AVAILABLE = True
|
||||
except ImportError:
|
||||
GITHUB_AVAILABLE = False
|
||||
|
||||
|
||||
class ExternalTools:
|
||||
"""External system integration tools."""
|
||||
|
||||
def __init__(self, llm_helper: LLMHelper):
|
||||
"""Initialize external tools with LLM helper."""
|
||||
self.llm_helper = llm_helper
|
||||
self._google_service = None
|
||||
self._github_client = None
|
||||
|
||||
def _get_google_calendar_service(self):
|
||||
"""Get or create Google Calendar service."""
|
||||
if not GOOGLE_AVAILABLE:
|
||||
raise ImportError("Google Calendar libraries not installed")
|
||||
|
||||
if self._google_service:
|
||||
return self._google_service
|
||||
|
||||
SCOPES = ['https://www.googleapis.com/auth/calendar']
|
||||
creds = None
|
||||
|
||||
token_path = Path('token.json')
|
||||
creds_path = Path(Config.GOOGLE_CALENDAR_CREDENTIALS_FILE)
|
||||
|
||||
# Load token if exists
|
||||
if token_path.exists():
|
||||
creds = Credentials.from_authorized_user_file(str(token_path), SCOPES)
|
||||
|
||||
# Refresh or get new token
|
||||
if not creds or not creds.valid:
|
||||
if creds and creds.expired and creds.refresh_token:
|
||||
creds.refresh(Request())
|
||||
else:
|
||||
if not creds_path.exists():
|
||||
raise FileNotFoundError(f"Credentials file not found: {creds_path}")
|
||||
flow = InstalledAppFlow.from_client_secrets_file(str(creds_path), SCOPES)
|
||||
creds = flow.run_local_server(port=0)
|
||||
|
||||
# Save token
|
||||
token_path.write_text(creds.to_json(), encoding="utf-8")
|
||||
|
||||
self._google_service = build('calendar', 'v3', credentials=creds)
|
||||
return self._google_service
|
||||
|
||||
def _get_github_client(self):
|
||||
"""Get or create GitHub client."""
|
||||
if not GITHUB_AVAILABLE:
|
||||
raise ImportError("GitHub library not installed")
|
||||
|
||||
if self._github_client:
|
||||
return self._github_client
|
||||
|
||||
if not Config.GITHUB_TOKEN:
|
||||
raise ValueError("GitHub token not configured")
|
||||
|
||||
self._github_client = Github(Config.GITHUB_TOKEN)
|
||||
return self._github_client
|
||||
|
||||
async def google_calendar_add(
|
||||
self,
|
||||
summary: str,
|
||||
start_time: str,
|
||||
end_time: str,
|
||||
description: Optional[str] = None,
|
||||
location: Optional[str] = None
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Add event to Google Calendar.
|
||||
|
||||
Args:
|
||||
summary: Event title
|
||||
start_time: Start time (ISO 8601 format or natural language)
|
||||
end_time: End time (ISO 8601 format or natural language)
|
||||
description: Event description
|
||||
location: Event location
|
||||
|
||||
Returns:
|
||||
Result dictionary with event details
|
||||
"""
|
||||
try:
|
||||
service = self._get_google_calendar_service()
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to initialize Google Calendar: {str(e)}"
|
||||
}
|
||||
|
||||
# Parse times
|
||||
try:
|
||||
start_dt = self._parse_datetime(start_time)
|
||||
end_dt = self._parse_datetime(end_time)
|
||||
except ValueError as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Invalid datetime format: {str(e)}"
|
||||
}
|
||||
|
||||
# Validate times
|
||||
if end_dt <= start_dt:
|
||||
return {
|
||||
"success": False,
|
||||
"error": "End time must be after start time"
|
||||
}
|
||||
|
||||
# Request approval
|
||||
if Config.REQUIRE_APPROVAL_FOR_DANGEROUS_OPS:
|
||||
approved, reason = self.llm_helper.request_approval(
|
||||
"google_calendar_add",
|
||||
{
|
||||
"summary": summary,
|
||||
"start_time": start_dt.isoformat(),
|
||||
"end_time": end_dt.isoformat(),
|
||||
"description": description
|
||||
}
|
||||
)
|
||||
|
||||
if not approved:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Calendar event creation not approved: {reason}"
|
||||
}
|
||||
|
||||
# Create event
|
||||
event = {
|
||||
'summary': summary,
|
||||
'start': {
|
||||
'dateTime': start_dt.isoformat(),
|
||||
'timeZone': 'UTC'
|
||||
},
|
||||
'end': {
|
||||
'dateTime': end_dt.isoformat(),
|
||||
'timeZone': 'UTC'
|
||||
}
|
||||
}
|
||||
|
||||
if description:
|
||||
event['description'] = description
|
||||
if location:
|
||||
event['location'] = location
|
||||
|
||||
try:
|
||||
created_event = service.events().insert(
|
||||
calendarId='primary',
|
||||
body=event
|
||||
).execute()
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"event_id": created_event['id'],
|
||||
"event_link": created_event.get('htmlLink'),
|
||||
"summary": summary,
|
||||
"start_time": start_dt.isoformat(),
|
||||
"end_time": end_dt.isoformat()
|
||||
}
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to create calendar event: {str(e)}"
|
||||
}
|
||||
|
||||
async def github_create_pr(
|
||||
self,
|
||||
repo_name: str,
|
||||
title: str,
|
||||
body: str,
|
||||
head_branch: str,
|
||||
base_branch: str = "main"
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Create a GitHub Pull Request.
|
||||
|
||||
Args:
|
||||
repo_name: Repository name (format: owner/repo)
|
||||
title: PR title
|
||||
body: PR description
|
||||
head_branch: Source branch
|
||||
base_branch: Target branch
|
||||
|
||||
Returns:
|
||||
Result dictionary with PR details
|
||||
"""
|
||||
try:
|
||||
github = self._get_github_client()
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to initialize GitHub client: {str(e)}"
|
||||
}
|
||||
|
||||
# Validate repository name
|
||||
if '/' not in repo_name:
|
||||
return {
|
||||
"success": False,
|
||||
"error": "Repository name must be in format: owner/repo"
|
||||
}
|
||||
|
||||
# Request approval
|
||||
if Config.REQUIRE_APPROVAL_FOR_DANGEROUS_OPS:
|
||||
approved, reason = self.llm_helper.request_approval(
|
||||
"github_create_pr",
|
||||
{
|
||||
"repo": repo_name,
|
||||
"title": title,
|
||||
"head": head_branch,
|
||||
"base": base_branch,
|
||||
"body_preview": body[:200]
|
||||
}
|
||||
)
|
||||
|
||||
if not approved:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"PR creation not approved: {reason}"
|
||||
}
|
||||
|
||||
try:
|
||||
# Get repository
|
||||
repo = github.get_repo(repo_name)
|
||||
|
||||
# Verify branches exist
|
||||
try:
|
||||
repo.get_branch(head_branch)
|
||||
repo.get_branch(base_branch)
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Branch verification failed: {str(e)}"
|
||||
}
|
||||
|
||||
# Query before mutation so a retry cannot create a duplicate PR.
|
||||
# This implements the idempotency rule described immediately before
|
||||
# Experiment 4-3 in the manuscript.
|
||||
owner = repo_name.split("/", 1)[0]
|
||||
existing = repo.get_pulls(
|
||||
state="open", head=f"{owner}:{head_branch}", base=base_branch
|
||||
)
|
||||
for pr in existing:
|
||||
if pr.head.ref == head_branch and pr.base.ref == base_branch:
|
||||
return {
|
||||
"success": True,
|
||||
"pr_number": pr.number,
|
||||
"pr_url": pr.html_url,
|
||||
"title": pr.title,
|
||||
"state": pr.state,
|
||||
"created_at": pr.created_at.isoformat(),
|
||||
"idempotent_reuse": True,
|
||||
}
|
||||
|
||||
# Create pull request
|
||||
pr = repo.create_pull(
|
||||
title=title,
|
||||
body=body,
|
||||
head=head_branch,
|
||||
base=base_branch
|
||||
)
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"pr_number": pr.number,
|
||||
"pr_url": pr.html_url,
|
||||
"title": title,
|
||||
"state": pr.state,
|
||||
"created_at": pr.created_at.isoformat(),
|
||||
"idempotent_reuse": False,
|
||||
}
|
||||
|
||||
except GithubException as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"GitHub API error: {e.data.get('message', str(e))}"
|
||||
}
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to create pull request: {str(e)}"
|
||||
}
|
||||
|
||||
def _parse_datetime(self, time_str: str) -> datetime:
|
||||
"""Parse datetime string in various formats."""
|
||||
# Try ISO 8601 format first
|
||||
formats = [
|
||||
'%Y-%m-%dT%H:%M:%S',
|
||||
'%Y-%m-%d %H:%M:%S',
|
||||
'%Y-%m-%d %H:%M',
|
||||
'%Y-%m-%d'
|
||||
]
|
||||
|
||||
for fmt in formats:
|
||||
try:
|
||||
return datetime.strptime(time_str, fmt)
|
||||
except ValueError:
|
||||
continue
|
||||
|
||||
# If no format works, raise error
|
||||
raise ValueError(f"Could not parse datetime: {time_str}")
|
||||
@@ -0,0 +1,210 @@
|
||||
"""File system tools with safety mechanisms."""
|
||||
|
||||
import itertools
|
||||
import os
|
||||
from pathlib import Path
|
||||
from typing import Dict, Any
|
||||
from llm_helper import LLMHelper
|
||||
from config import Config
|
||||
|
||||
|
||||
class FileTools:
|
||||
"""File system tools with verification and safety checks."""
|
||||
|
||||
def __init__(self, llm_helper: LLMHelper):
|
||||
"""Initialize file tools with LLM helper."""
|
||||
self.llm_helper = llm_helper
|
||||
self.workspace_dir = Config.WORKSPACE_DIR
|
||||
|
||||
def _resolve_path(self, path: str) -> Path:
|
||||
"""Resolve path relative to workspace."""
|
||||
path_obj = Path(path)
|
||||
if not path_obj.is_absolute():
|
||||
path_obj = self.workspace_dir / path_obj
|
||||
return path_obj.resolve()
|
||||
|
||||
def _is_safe_path(self, path: Path) -> bool:
|
||||
"""Check if path is within workspace."""
|
||||
try:
|
||||
path.resolve().relative_to(self.workspace_dir.resolve())
|
||||
return True
|
||||
except ValueError:
|
||||
return False
|
||||
|
||||
async def write_file(
|
||||
self,
|
||||
path: str,
|
||||
content: str,
|
||||
overwrite: bool = False
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Write content to a file with safety checks.
|
||||
|
||||
Args:
|
||||
path: File path (relative to workspace or absolute)
|
||||
content: Content to write
|
||||
overwrite: Whether to overwrite existing files
|
||||
|
||||
Returns:
|
||||
Result dictionary with status and details
|
||||
"""
|
||||
resolved_path = self._resolve_path(path)
|
||||
|
||||
# Safety check: ensure path is within workspace
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Path {path} is outside workspace directory"
|
||||
}
|
||||
|
||||
# Check if file exists and overwrite is not allowed
|
||||
if resolved_path.exists() and not overwrite:
|
||||
# Request approval for overwriting
|
||||
if Config.REQUIRE_APPROVAL_FOR_DANGEROUS_OPS:
|
||||
approved, reason = self.llm_helper.request_approval(
|
||||
"file_overwrite",
|
||||
{
|
||||
"path": str(resolved_path),
|
||||
"existing_size": resolved_path.stat().st_size,
|
||||
"new_content_size": len(content)
|
||||
}
|
||||
)
|
||||
|
||||
if not approved:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Overwrite not approved: {reason}"
|
||||
}
|
||||
|
||||
# Verify code syntax if it's a code file
|
||||
if Config.AUTO_VERIFY_CODE and resolved_path.suffix in ['.py', '.js', '.ts']:
|
||||
language = {'.py': 'python', '.js': 'javascript', '.ts': 'typescript'}[resolved_path.suffix]
|
||||
is_valid, error_msg = self.llm_helper.verify_code_syntax(content, language)
|
||||
|
||||
if not is_valid:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Syntax validation failed: {error_msg}",
|
||||
"verification": "failed"
|
||||
}
|
||||
|
||||
# Write the file
|
||||
try:
|
||||
resolved_path.parent.mkdir(parents=True, exist_ok=True)
|
||||
resolved_path.write_text(content, encoding="utf-8")
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"path": str(resolved_path),
|
||||
"bytes_written": len(content),
|
||||
"verification": "passed" if Config.AUTO_VERIFY_CODE else "skipped"
|
||||
}
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to write file: {str(e)}"
|
||||
}
|
||||
|
||||
async def edit_file(
|
||||
self,
|
||||
path: str,
|
||||
search: str,
|
||||
replace: str
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Edit a file by searching and replacing content.
|
||||
|
||||
Args:
|
||||
path: File path
|
||||
search: Text to search for
|
||||
replace: Replacement text
|
||||
|
||||
Returns:
|
||||
Result dictionary with status and details
|
||||
"""
|
||||
resolved_path = self._resolve_path(path)
|
||||
|
||||
# Safety check
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Path {path} is outside workspace directory"
|
||||
}
|
||||
|
||||
# Check if file exists
|
||||
if not resolved_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {path} does not exist"
|
||||
}
|
||||
|
||||
# Read current content
|
||||
try:
|
||||
current_content = resolved_path.read_text(encoding="utf-8")
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to read file: {str(e)}"
|
||||
}
|
||||
|
||||
# Empty search matches everywhere; reject instead of inserting at start.
|
||||
if search == "":
|
||||
return {
|
||||
"success": False,
|
||||
"error": "Search text cannot be empty"
|
||||
}
|
||||
|
||||
# Check if search text exists
|
||||
if search not in current_content:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Search text not found in file"
|
||||
}
|
||||
|
||||
# Perform replacement
|
||||
new_content = current_content.replace(search, replace, 1)
|
||||
|
||||
# Generate diff preview
|
||||
diff_preview = self._generate_diff(current_content, new_content)
|
||||
|
||||
# Verify new content if it's code
|
||||
if Config.AUTO_VERIFY_CODE and resolved_path.suffix in ['.py', '.js', '.ts']:
|
||||
language = {'.py': 'python', '.js': 'javascript', '.ts': 'typescript'}[resolved_path.suffix]
|
||||
is_valid, error_msg = self.llm_helper.verify_code_syntax(new_content, language)
|
||||
|
||||
if not is_valid:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Syntax validation failed after edit: {error_msg}",
|
||||
"diff_preview": diff_preview
|
||||
}
|
||||
|
||||
# Write the modified content
|
||||
try:
|
||||
resolved_path.write_text(new_content, encoding="utf-8")
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"path": str(resolved_path),
|
||||
"diff_preview": diff_preview,
|
||||
"verification": "passed" if Config.AUTO_VERIFY_CODE else "skipped"
|
||||
}
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to write file: {str(e)}"
|
||||
}
|
||||
|
||||
def _generate_diff(self, old_content: str, new_content: str) -> str:
|
||||
"""Generate a simple diff preview."""
|
||||
old_lines = old_content.split('\n')
|
||||
new_lines = new_content.split('\n')
|
||||
|
||||
diff_lines = []
|
||||
for i, (old, new) in enumerate(itertools.zip_longest(old_lines, new_lines, fillvalue=''), 1):
|
||||
if old != new:
|
||||
diff_lines.append(f"Line {i}:")
|
||||
diff_lines.append(f" - {old}")
|
||||
diff_lines.append(f" + {new}")
|
||||
|
||||
return '\n'.join(diff_lines[:20]) # Limit to 20 lines
|
||||
@@ -0,0 +1,676 @@
|
||||
"""
|
||||
Enhanced filesystem tools based on AWorld filesystem-server.
|
||||
Provides comprehensive file and directory operations with safety checks.
|
||||
"""
|
||||
import json
|
||||
import os
|
||||
import shutil
|
||||
import traceback
|
||||
from pathlib import Path
|
||||
from typing import Dict, Any, List
|
||||
|
||||
from config import Config
|
||||
|
||||
|
||||
class FilesystemEnhanced:
|
||||
"""Enhanced filesystem operations with safety and validation."""
|
||||
|
||||
def __init__(self):
|
||||
self.workspace_dir = Path(Config.WORKSPACE_DIR).resolve()
|
||||
self.allowed_directories = [self.workspace_dir]
|
||||
|
||||
def _resolve_path(self, path: str) -> Path:
|
||||
"""Resolve path relative to workspace."""
|
||||
path_obj = Path(path)
|
||||
if not path_obj.is_absolute():
|
||||
path_obj = self.workspace_dir / path_obj
|
||||
return path_obj.resolve()
|
||||
|
||||
def _is_safe_path(self, path: Path) -> bool:
|
||||
"""Check if path is within allowed directories."""
|
||||
try:
|
||||
resolved = path.resolve()
|
||||
for allowed in self.allowed_directories:
|
||||
try:
|
||||
resolved.relative_to(allowed.resolve())
|
||||
return True
|
||||
except ValueError:
|
||||
continue
|
||||
return False
|
||||
except Exception:
|
||||
return False
|
||||
|
||||
async def read_text_file(
|
||||
self,
|
||||
file_path: str,
|
||||
encoding: str = "utf-8",
|
||||
max_size_mb: int = 10
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Read a text file with size limits.
|
||||
|
||||
Args:
|
||||
file_path: Path to the file
|
||||
encoding: File encoding
|
||||
max_size_mb: Maximum file size in MB
|
||||
|
||||
Returns:
|
||||
Dictionary with file content and metadata
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(file_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Path {file_path} is outside allowed directories"
|
||||
}
|
||||
|
||||
if not resolved_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {file_path} does not exist"
|
||||
}
|
||||
|
||||
# Check file size
|
||||
file_size = resolved_path.stat().st_size
|
||||
max_size_bytes = max_size_mb * 1024 * 1024
|
||||
|
||||
if file_size > max_size_bytes:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File too large: {file_size / (1024*1024):.2f}MB (max: {max_size_mb}MB)"
|
||||
}
|
||||
|
||||
# Read file
|
||||
content = resolved_path.read_text(encoding=encoding)
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"content": content,
|
||||
"file_path": str(resolved_path),
|
||||
"file_size": file_size,
|
||||
"encoding": encoding,
|
||||
"lines": len(content.splitlines())
|
||||
}
|
||||
|
||||
except UnicodeDecodeError:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File is not a valid text file with encoding {encoding}"
|
||||
}
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to read file: {str(e)}"
|
||||
}
|
||||
|
||||
async def read_multiple_files(
|
||||
self,
|
||||
file_paths: List[str],
|
||||
encoding: str = "utf-8"
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Read multiple files at once.
|
||||
|
||||
Args:
|
||||
file_paths: List of file paths
|
||||
encoding: File encoding
|
||||
|
||||
Returns:
|
||||
Dictionary with all file contents
|
||||
"""
|
||||
results = {}
|
||||
errors = []
|
||||
|
||||
for file_path in file_paths:
|
||||
result = await self.read_text_file(file_path, encoding)
|
||||
|
||||
if result["success"]:
|
||||
results[file_path] = {
|
||||
"content": result["content"],
|
||||
"size": result["file_size"],
|
||||
"lines": result["lines"]
|
||||
}
|
||||
else:
|
||||
errors.append({
|
||||
"file": file_path,
|
||||
"error": result["error"]
|
||||
})
|
||||
|
||||
return {
|
||||
"success": len(results) > 0,
|
||||
"files_read": len(results),
|
||||
"files_failed": len(errors),
|
||||
"results": results,
|
||||
"errors": errors
|
||||
}
|
||||
|
||||
async def list_directory_with_sizes(
|
||||
self,
|
||||
directory_path: str = "."
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
List directory contents with file sizes.
|
||||
|
||||
Args:
|
||||
directory_path: Path to directory
|
||||
|
||||
Returns:
|
||||
Dictionary with directory contents and sizes
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(directory_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Path {directory_path} is outside allowed directories"
|
||||
}
|
||||
|
||||
if not resolved_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Directory {directory_path} does not exist"
|
||||
}
|
||||
|
||||
if not resolved_path.is_dir():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"{directory_path} is not a directory"
|
||||
}
|
||||
|
||||
# List contents with sizes
|
||||
contents = []
|
||||
total_size = 0
|
||||
|
||||
for item in sorted(resolved_path.iterdir()):
|
||||
try:
|
||||
is_dir = item.is_dir()
|
||||
size = 0 if is_dir else item.stat().st_size
|
||||
total_size += size
|
||||
|
||||
contents.append({
|
||||
"name": item.name,
|
||||
"type": "directory" if is_dir else "file",
|
||||
"size": size,
|
||||
"size_human": self._format_size(size),
|
||||
"modified": item.stat().st_mtime
|
||||
})
|
||||
except Exception as e:
|
||||
contents.append({
|
||||
"name": item.name,
|
||||
"type": "unknown",
|
||||
"error": str(e)
|
||||
})
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"directory": str(resolved_path),
|
||||
"total_items": len(contents),
|
||||
"total_size": total_size,
|
||||
"total_size_human": self._format_size(total_size),
|
||||
"contents": contents
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to list directory: {str(e)}"
|
||||
}
|
||||
|
||||
async def directory_tree(
|
||||
self,
|
||||
directory_path: str = ".",
|
||||
max_depth: int = 3,
|
||||
show_hidden: bool = False
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Generate a tree structure of directory contents.
|
||||
|
||||
Args:
|
||||
directory_path: Path to directory
|
||||
max_depth: Maximum depth to traverse
|
||||
show_hidden: Whether to show hidden files
|
||||
|
||||
Returns:
|
||||
Dictionary with directory tree structure
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(directory_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Path {directory_path} is outside allowed directories"
|
||||
}
|
||||
|
||||
def build_tree(path: Path, current_depth: int = 0) -> Dict[str, Any]:
|
||||
"""Recursively build tree structure."""
|
||||
if current_depth >= max_depth:
|
||||
return {"name": path.name, "type": "directory", "truncated": True}
|
||||
|
||||
if not path.is_dir():
|
||||
return {
|
||||
"name": path.name,
|
||||
"type": "file",
|
||||
"size": path.stat().st_size
|
||||
}
|
||||
|
||||
children = []
|
||||
try:
|
||||
for item in sorted(path.iterdir()):
|
||||
# Skip hidden files if needed
|
||||
if not show_hidden and item.name.startswith('.'):
|
||||
continue
|
||||
|
||||
children.append(build_tree(item, current_depth + 1))
|
||||
except PermissionError:
|
||||
return {
|
||||
"name": path.name,
|
||||
"type": "directory",
|
||||
"error": "Permission denied"
|
||||
}
|
||||
|
||||
return {
|
||||
"name": path.name,
|
||||
"type": "directory",
|
||||
"children": children,
|
||||
"count": len(children)
|
||||
}
|
||||
|
||||
tree = build_tree(resolved_path)
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"root": str(resolved_path),
|
||||
"tree": tree,
|
||||
"max_depth": max_depth
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to generate directory tree: {str(e)}"
|
||||
}
|
||||
|
||||
async def search_files(
|
||||
self,
|
||||
pattern: str,
|
||||
directory_path: str = ".",
|
||||
recursive: bool = True,
|
||||
case_sensitive: bool = False
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Search for files matching a pattern.
|
||||
|
||||
Args:
|
||||
pattern: Glob pattern (e.g., "*.py", "test_*.txt")
|
||||
directory_path: Directory to search in
|
||||
recursive: Search recursively
|
||||
case_sensitive: Case-sensitive matching
|
||||
|
||||
Returns:
|
||||
Dictionary with matching files
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(directory_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Path {directory_path} is outside allowed directories"
|
||||
}
|
||||
|
||||
if not resolved_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Directory {directory_path} does not exist"
|
||||
}
|
||||
|
||||
# Search for files
|
||||
if recursive:
|
||||
matches = list(resolved_path.rglob(pattern))
|
||||
else:
|
||||
matches = list(resolved_path.glob(pattern))
|
||||
|
||||
# Filter to files only
|
||||
files = [m for m in matches if m.is_file()]
|
||||
|
||||
results = []
|
||||
for file_path in sorted(files):
|
||||
try:
|
||||
stat = file_path.stat()
|
||||
results.append({
|
||||
"path": str(file_path.relative_to(resolved_path)),
|
||||
"absolute_path": str(file_path),
|
||||
"size": stat.st_size,
|
||||
"size_human": self._format_size(stat.st_size),
|
||||
"modified": stat.st_mtime
|
||||
})
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"pattern": pattern,
|
||||
"directory": str(resolved_path),
|
||||
"recursive": recursive,
|
||||
"matches": len(results),
|
||||
"files": results
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to search files: {str(e)}"
|
||||
}
|
||||
|
||||
async def get_file_info(
|
||||
self,
|
||||
file_path: str
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Get detailed information about a file.
|
||||
|
||||
Args:
|
||||
file_path: Path to the file
|
||||
|
||||
Returns:
|
||||
Dictionary with file information
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(file_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Path {file_path} is outside allowed directories"
|
||||
}
|
||||
|
||||
if not resolved_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {file_path} does not exist"
|
||||
}
|
||||
|
||||
stat = resolved_path.stat()
|
||||
|
||||
info = {
|
||||
"path": str(resolved_path),
|
||||
"name": resolved_path.name,
|
||||
"extension": resolved_path.suffix,
|
||||
"size": stat.st_size,
|
||||
"size_human": self._format_size(stat.st_size),
|
||||
"is_file": resolved_path.is_file(),
|
||||
"is_directory": resolved_path.is_dir(),
|
||||
"is_symlink": resolved_path.is_symlink(),
|
||||
"created": stat.st_ctime,
|
||||
"modified": stat.st_mtime,
|
||||
"accessed": stat.st_atime,
|
||||
"permissions": oct(stat.st_mode)[-3:]
|
||||
}
|
||||
|
||||
# Add parent directory info
|
||||
info["parent"] = str(resolved_path.parent)
|
||||
|
||||
# For text files, add line count
|
||||
if resolved_path.is_file() and resolved_path.suffix in ['.txt', '.py', '.md', '.json', '.yaml', '.yml']:
|
||||
try:
|
||||
content = resolved_path.read_text(encoding="utf-8")
|
||||
info["lines"] = len(content.splitlines())
|
||||
info["characters"] = len(content)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"file_info": info
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to get file info: {str(e)}"
|
||||
}
|
||||
|
||||
async def move_file(
|
||||
self,
|
||||
source: str,
|
||||
destination: str,
|
||||
overwrite: bool = False
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Move or rename a file/directory.
|
||||
|
||||
Args:
|
||||
source: Source path
|
||||
destination: Destination path
|
||||
overwrite: Whether to overwrite existing destination
|
||||
|
||||
Returns:
|
||||
Dictionary with operation result
|
||||
"""
|
||||
try:
|
||||
source_path = self._resolve_path(source)
|
||||
dest_path = self._resolve_path(destination)
|
||||
|
||||
if not self._is_safe_path(source_path) or not self._is_safe_path(dest_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": "Paths must be within allowed directories"
|
||||
}
|
||||
|
||||
if not source_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Source {source} does not exist"
|
||||
}
|
||||
|
||||
if dest_path.exists() and not overwrite:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Destination {destination} already exists. Use overwrite=True to replace."
|
||||
}
|
||||
|
||||
# Perform move
|
||||
if dest_path.exists():
|
||||
if dest_path.is_dir():
|
||||
shutil.rmtree(dest_path)
|
||||
else:
|
||||
dest_path.unlink()
|
||||
|
||||
shutil.move(str(source_path), str(dest_path))
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"source": str(source_path),
|
||||
"destination": str(dest_path),
|
||||
"message": f"Moved {source} to {destination}"
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to move file: {str(e)}"
|
||||
}
|
||||
|
||||
async def copy_file(
|
||||
self,
|
||||
source: str,
|
||||
destination: str,
|
||||
overwrite: bool = False
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Copy a file or directory.
|
||||
|
||||
Args:
|
||||
source: Source path
|
||||
destination: Destination path
|
||||
overwrite: Whether to overwrite existing destination
|
||||
|
||||
Returns:
|
||||
Dictionary with operation result
|
||||
"""
|
||||
try:
|
||||
source_path = self._resolve_path(source)
|
||||
dest_path = self._resolve_path(destination)
|
||||
|
||||
if not self._is_safe_path(source_path) or not self._is_safe_path(dest_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": "Paths must be within allowed directories"
|
||||
}
|
||||
|
||||
if not source_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Source {source} does not exist"
|
||||
}
|
||||
|
||||
if dest_path.exists() and not overwrite:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Destination {destination} already exists"
|
||||
}
|
||||
|
||||
# Perform copy
|
||||
if source_path.is_dir():
|
||||
if dest_path.exists():
|
||||
shutil.rmtree(dest_path)
|
||||
shutil.copytree(source_path, dest_path)
|
||||
else:
|
||||
dest_path.parent.mkdir(parents=True, exist_ok=True)
|
||||
shutil.copy2(source_path, dest_path)
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"source": str(source_path),
|
||||
"destination": str(dest_path),
|
||||
"message": f"Copied {source} to {destination}"
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to copy file: {str(e)}"
|
||||
}
|
||||
|
||||
async def delete_file(
|
||||
self,
|
||||
file_path: str,
|
||||
recursive: bool = False
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Delete a file or directory.
|
||||
|
||||
Args:
|
||||
file_path: Path to delete
|
||||
recursive: For directories, delete recursively
|
||||
|
||||
Returns:
|
||||
Dictionary with operation result
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(file_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Path {file_path} is outside allowed directories"
|
||||
}
|
||||
|
||||
if not resolved_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Path {file_path} does not exist"
|
||||
}
|
||||
|
||||
# Delete
|
||||
if resolved_path.is_dir():
|
||||
if not recursive:
|
||||
return {
|
||||
"success": False,
|
||||
"error": "Cannot delete directory without recursive=True"
|
||||
}
|
||||
shutil.rmtree(resolved_path)
|
||||
else:
|
||||
resolved_path.unlink()
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"deleted": str(resolved_path),
|
||||
"message": f"Deleted {file_path}"
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to delete: {str(e)}"
|
||||
}
|
||||
|
||||
async def create_directory(
|
||||
self,
|
||||
directory_path: str,
|
||||
parents: bool = True
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Create a new directory.
|
||||
|
||||
Args:
|
||||
directory_path: Path for new directory
|
||||
parents: Create parent directories if needed
|
||||
|
||||
Returns:
|
||||
Dictionary with operation result
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(directory_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Path {directory_path} is outside allowed directories"
|
||||
}
|
||||
|
||||
if resolved_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Directory {directory_path} already exists"
|
||||
}
|
||||
|
||||
# Create directory
|
||||
resolved_path.mkdir(parents=parents, exist_ok=False)
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"directory": str(resolved_path),
|
||||
"message": f"Created directory {directory_path}"
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to create directory: {str(e)}"
|
||||
}
|
||||
|
||||
async def list_allowed_directories(self) -> Dict[str, Any]:
|
||||
"""
|
||||
List directories that are accessible.
|
||||
|
||||
Returns:
|
||||
Dictionary with allowed directories
|
||||
"""
|
||||
return {
|
||||
"success": True,
|
||||
"allowed_directories": [str(d) for d in self.allowed_directories],
|
||||
"count": len(self.allowed_directories)
|
||||
}
|
||||
|
||||
def _format_size(self, size_bytes: int) -> str:
|
||||
"""Format file size in human-readable format."""
|
||||
for unit in ['B', 'KB', 'MB', 'GB', 'TB']:
|
||||
if size_bytes < 1024.0:
|
||||
return f"{size_bytes:.2f} {unit}"
|
||||
size_bytes /= 1024.0
|
||||
return f"{size_bytes:.2f} PB"
|
||||
@@ -0,0 +1,9 @@
|
||||
#!/usr/bin/env python3
|
||||
"""A simple greeting script."""
|
||||
|
||||
def main():
|
||||
name = "World"
|
||||
print(f"Hello, {name}!")
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
@@ -0,0 +1,346 @@
|
||||
"""LLM helper for safety checks, approval, and summarization."""
|
||||
|
||||
import json
|
||||
import datetime as dt
|
||||
import os
|
||||
import subprocess
|
||||
import time
|
||||
from pathlib import Path
|
||||
from typing import Optional, Dict, Any
|
||||
from openai import OpenAI
|
||||
from config import Config
|
||||
|
||||
|
||||
def _reasoning_safe_temperature(model, requested=1.0):
|
||||
"""Reasoning models (Kimi K3, GPT-5, ...) only accept temperature=1.
|
||||
Return 1 for those; otherwise the requested value so non-reasoning
|
||||
providers (Doubao, DeepSeek, older Moonshot) are unchanged."""
|
||||
m = str(model or "").lower().replace("/", "-")
|
||||
return 1 if ("kimi-k3" in m or "gpt-5" in m) else requested
|
||||
|
||||
|
||||
def _parse_json_response(content):
|
||||
"""Parse a JSON object out of an LLM reply, tolerating markdown fences.
|
||||
|
||||
Reasoning models (notably kimi-k3) reliably return valid JSON but wrap it
|
||||
in a ```json ... ``` code fence, so a bare json.loads() fails with
|
||||
"Expecting value: line 1 column 1". Strip an optional fence and, as a last
|
||||
resort, slice from the first '{' to the last '}' before parsing."""
|
||||
text = (content or "").strip()
|
||||
if text.startswith("```"):
|
||||
# Drop the opening fence line (``` or ```json) and the closing fence.
|
||||
text = text.split("\n", 1)[1] if "\n" in text else ""
|
||||
if text.rstrip().endswith("```"):
|
||||
text = text.rstrip()[:-3]
|
||||
text = text.strip()
|
||||
try:
|
||||
return json.loads(text)
|
||||
except json.JSONDecodeError:
|
||||
start, end = text.find("{"), text.rfind("}")
|
||||
if start != -1 and end != -1 and end > start:
|
||||
return json.loads(text[start:end + 1])
|
||||
raise
|
||||
|
||||
|
||||
class LLMHelper:
|
||||
"""Helper class for LLM-based operations."""
|
||||
|
||||
def __init__(self):
|
||||
"""Initialize the LLM helper.
|
||||
|
||||
The OpenAI-compatible client is created lazily on first use so that
|
||||
execution tools which do not need an LLM (e.g. Python code execution
|
||||
with local syntax checking, terminal commands, file writes) work
|
||||
offline without any API key configured. Methods that actually call
|
||||
the LLM (approval, summarization, non-Python syntax check) will raise
|
||||
or fail-safe if no key is available.
|
||||
"""
|
||||
self.client = None
|
||||
self.model = None
|
||||
self.provider = None
|
||||
|
||||
def _record_receipt(self, purpose: str, request: dict, response, latency: float) -> None:
|
||||
"""Checkpoint credential-free raw provider evidence after every call."""
|
||||
target = os.getenv("EXECUTION_LLM_RECEIPT_PATH")
|
||||
if not target:
|
||||
return
|
||||
path = Path(target)
|
||||
path.parent.mkdir(parents=True, exist_ok=True)
|
||||
usage = getattr(response, "usage", None)
|
||||
choice = response.choices[0]
|
||||
row = {
|
||||
"purpose": purpose,
|
||||
"called_at_utc": dt.datetime.now(dt.timezone.utc).isoformat(),
|
||||
"provider": self.provider,
|
||||
"request": request,
|
||||
"response": {
|
||||
"id": getattr(response, "id", None),
|
||||
"model": getattr(response, "model", None),
|
||||
"finish_reason": getattr(choice, "finish_reason", None),
|
||||
"content": choice.message.content,
|
||||
},
|
||||
"usage": {
|
||||
"prompt_tokens": getattr(usage, "prompt_tokens", None),
|
||||
"completion_tokens": getattr(usage, "completion_tokens", None),
|
||||
"total_tokens": getattr(usage, "total_tokens", None),
|
||||
},
|
||||
"latency_seconds": round(latency, 3),
|
||||
}
|
||||
existing = json.loads(path.read_text(encoding="utf-8")) if path.is_file() else []
|
||||
existing.append(row)
|
||||
temporary = path.with_suffix(path.suffix + ".tmp")
|
||||
temporary.write_text(json.dumps(existing, ensure_ascii=False, indent=2), encoding="utf-8")
|
||||
temporary.replace(path)
|
||||
|
||||
def _ensure_client(self) -> None:
|
||||
"""Create the LLM client on first use (raises if no API key)."""
|
||||
if self.client is None:
|
||||
llm_config = Config.get_llm_config()
|
||||
# All providers use OpenAI-compatible API
|
||||
self.client = OpenAI(
|
||||
api_key=llm_config["api_key"],
|
||||
base_url=llm_config.get("base_url")
|
||||
)
|
||||
self.model = llm_config["model"]
|
||||
self.provider = llm_config["provider"]
|
||||
|
||||
def request_approval(
|
||||
self,
|
||||
operation: str,
|
||||
details: Dict[str, Any]
|
||||
) -> tuple[bool, str]:
|
||||
"""
|
||||
Request LLM approval for a dangerous operation.
|
||||
|
||||
Args:
|
||||
operation: The operation name
|
||||
details: Details about the operation
|
||||
|
||||
Returns:
|
||||
Tuple of (approved, reason)
|
||||
"""
|
||||
prompt = f"""You are a safety reviewer for an AI agent execution system.
|
||||
Review the following operation and determine if it should be approved.
|
||||
|
||||
Operation: {operation}
|
||||
Details: {json.dumps(details, indent=2)}
|
||||
|
||||
Analyze the operation for:
|
||||
1. Potential data loss or destructive actions
|
||||
2. Security risks
|
||||
3. Resource consumption concerns
|
||||
4. Compliance with best practices
|
||||
|
||||
Respond in JSON format:
|
||||
{{
|
||||
"approved": true/false,
|
||||
"reason": "Brief explanation of your decision",
|
||||
"risk_level": "low/medium/high",
|
||||
"recommendations": ["List of recommendations if any"]
|
||||
}}
|
||||
"""
|
||||
|
||||
try:
|
||||
self._ensure_client()
|
||||
request = {
|
||||
"model": self.model,
|
||||
"messages": [
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are a cautious safety reviewer. Approve operations that are safe and reject risky ones."
|
||||
},
|
||||
{"role": "user", "content": prompt}
|
||||
],
|
||||
"temperature": _reasoning_safe_temperature(self.model, 0.1),
|
||||
"max_tokens": Config.MAX_TOKENS,
|
||||
}
|
||||
started = time.perf_counter()
|
||||
response = self.client.chat.completions.create(**request)
|
||||
self._record_receipt("dangerous_operation_review", request, response,
|
||||
time.perf_counter() - started)
|
||||
|
||||
result = _parse_json_response(response.choices[0].message.content)
|
||||
return result["approved"], result["reason"]
|
||||
|
||||
except Exception as e:
|
||||
# If approval check fails, default to rejection for safety
|
||||
return False, f"Approval check failed: {str(e)}"
|
||||
|
||||
def summarize_output(
|
||||
self,
|
||||
tool_name: str,
|
||||
output: str
|
||||
) -> str:
|
||||
"""
|
||||
Summarize complex tool output.
|
||||
|
||||
Args:
|
||||
tool_name: Name of the tool that produced the output
|
||||
output: The output to summarize
|
||||
|
||||
Returns:
|
||||
Summarized output
|
||||
"""
|
||||
|
||||
prompt = f"""Summarize the following output from the '{tool_name}' tool.
|
||||
Focus on:
|
||||
1. Key results or findings
|
||||
2. Errors or warnings
|
||||
3. Important patterns or insights
|
||||
4. Actionable information
|
||||
|
||||
Output to summarize:
|
||||
{output[:5000]} # Limit input to avoid token limits
|
||||
|
||||
Provide a concise summary that captures the essential information."""
|
||||
|
||||
try:
|
||||
self._ensure_client()
|
||||
response = self.client.chat.completions.create(
|
||||
model=self.model,
|
||||
messages=[
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are an expert at summarizing technical output. Be concise and focus on actionable information."
|
||||
},
|
||||
{"role": "user", "content": prompt}
|
||||
],
|
||||
temperature=_reasoning_safe_temperature(self.model, 0.1),
|
||||
max_tokens=Config.MAX_TOKENS
|
||||
)
|
||||
|
||||
summary = response.choices[0].message.content
|
||||
return f"[SUMMARIZED OUTPUT]\n{summary}\n\n[Original output length: {len(output)} characters]"
|
||||
|
||||
except Exception as e:
|
||||
return f"[SUMMARIZATION FAILED: {str(e)}]\n\n{output[:Config.MAX_OUTPUT_LENGTH]}..."
|
||||
|
||||
def analyze_error(
|
||||
self,
|
||||
tool_name: str,
|
||||
command: str,
|
||||
error_output: str
|
||||
) -> str:
|
||||
"""
|
||||
Analyze error output and provide suggestions.
|
||||
|
||||
Args:
|
||||
tool_name: Name of the tool that produced the error
|
||||
command: The command or code that failed
|
||||
error_output: The error output
|
||||
|
||||
Returns:
|
||||
Analysis with suggestions
|
||||
"""
|
||||
prompt = f"""Analyze the following error from the '{tool_name}' tool:
|
||||
|
||||
Command/Code:
|
||||
{command}
|
||||
|
||||
Error Output:
|
||||
{error_output[:3000]}
|
||||
|
||||
Provide:
|
||||
1. Root cause analysis
|
||||
2. Suggested fixes
|
||||
3. Prevention strategies
|
||||
|
||||
Be concise and practical."""
|
||||
|
||||
try:
|
||||
self._ensure_client()
|
||||
response = self.client.chat.completions.create(
|
||||
model=self.model,
|
||||
messages=[
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are an expert debugger. Analyze errors and provide clear, actionable solutions."
|
||||
},
|
||||
{"role": "user", "content": prompt}
|
||||
],
|
||||
temperature=_reasoning_safe_temperature(self.model, 0.2),
|
||||
max_tokens=Config.MAX_TOKENS
|
||||
)
|
||||
|
||||
return response.choices[0].message.content
|
||||
|
||||
except Exception as e:
|
||||
return f"Error analysis failed: {str(e)}"
|
||||
|
||||
def verify_code_syntax(
|
||||
self,
|
||||
code: str,
|
||||
language: str = "python"
|
||||
) -> tuple[bool, Optional[str]]:
|
||||
"""
|
||||
Verify code syntax and provide feedback.
|
||||
|
||||
Args:
|
||||
code: The code to verify
|
||||
language: Programming language
|
||||
|
||||
Returns:
|
||||
Tuple of (is_valid, error_message)
|
||||
"""
|
||||
# For Python, we can do actual syntax checking
|
||||
if language == "python":
|
||||
try:
|
||||
compile(code, "<string>", "exec")
|
||||
return True, None
|
||||
except SyntaxError as e:
|
||||
return False, f"Syntax error at line {e.lineno}: {e.msg}"
|
||||
|
||||
# JavaScript gets a real deterministic parser/linter rather than an
|
||||
# LLM opinion. Node's --check performs syntax validation without
|
||||
# executing the program.
|
||||
if language in {"javascript", "js"}:
|
||||
try:
|
||||
process = subprocess.run(
|
||||
["node", "--check", "-"], input=code, text=True,
|
||||
stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=10,
|
||||
)
|
||||
except (OSError, subprocess.SubprocessError) as exc:
|
||||
return False, f"JavaScript linter unavailable: {exc}"
|
||||
if process.returncode == 0:
|
||||
return True, None
|
||||
return False, process.stderr.strip() or "JavaScript syntax check failed"
|
||||
|
||||
# For other languages, use LLM for basic validation
|
||||
prompt = f"""Check the following {language} code for syntax errors:
|
||||
|
||||
```{language}
|
||||
{code}
|
||||
```
|
||||
|
||||
Respond in JSON format:
|
||||
{{
|
||||
"valid": true/false,
|
||||
"errors": ["List of syntax errors if any"],
|
||||
"warnings": ["List of warnings if any"]
|
||||
}}
|
||||
"""
|
||||
|
||||
try:
|
||||
self._ensure_client()
|
||||
response = self.client.chat.completions.create(
|
||||
model=self.model,
|
||||
messages=[
|
||||
{
|
||||
"role": "system",
|
||||
"content": f"You are a {language} syntax validator. Check code for syntax errors."
|
||||
},
|
||||
{"role": "user", "content": prompt}
|
||||
],
|
||||
temperature=_reasoning_safe_temperature(self.model, 0.1),
|
||||
max_tokens=Config.MAX_TOKENS
|
||||
)
|
||||
|
||||
result = _parse_json_response(response.choices[0].message.content)
|
||||
if result["valid"]:
|
||||
return True, None
|
||||
else:
|
||||
return False, "; ".join(result["errors"])
|
||||
|
||||
except Exception as e:
|
||||
# If validation fails, allow the code through
|
||||
return True, None
|
||||
@@ -0,0 +1,639 @@
|
||||
"""Multi-language code execution support inspired by SandboxFusion."""
|
||||
|
||||
import asyncio
|
||||
import subprocess
|
||||
import tempfile
|
||||
import os
|
||||
import shutil
|
||||
import time
|
||||
import base64
|
||||
import psutil
|
||||
import shlex
|
||||
from typing import Dict, Any, Optional, List
|
||||
from enum import Enum
|
||||
import logging
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
def try_decode(s: bytes) -> str:
|
||||
"""Safely decode bytes to string."""
|
||||
try:
|
||||
return s.decode('utf-8', errors='replace')
|
||||
except Exception as e:
|
||||
return f'[DecodeError] {e}'
|
||||
|
||||
|
||||
async def get_all_output(stream) -> str:
|
||||
"""Read stream until EOF. Call after the process has exited or been killed."""
|
||||
if stream is None:
|
||||
return ""
|
||||
try:
|
||||
result = await stream.read()
|
||||
return try_decode(result)
|
||||
except Exception as e:
|
||||
logger.debug(f"Error reading output: {e}")
|
||||
return ""
|
||||
|
||||
|
||||
def kill_process_tree(pid: int):
|
||||
"""Kill process and all its children."""
|
||||
try:
|
||||
parent = psutil.Process(pid)
|
||||
children = parent.children(recursive=True)
|
||||
|
||||
# Kill children first
|
||||
for child in children:
|
||||
try:
|
||||
child.kill()
|
||||
except psutil.NoSuchProcess:
|
||||
pass
|
||||
|
||||
# Kill parent
|
||||
try:
|
||||
parent.kill()
|
||||
except psutil.NoSuchProcess:
|
||||
pass
|
||||
|
||||
except psutil.NoSuchProcess:
|
||||
pass
|
||||
except Exception as e:
|
||||
logger.warning(f'Error killing process tree: {e}')
|
||||
|
||||
|
||||
class ExecutionStatus(str, Enum):
|
||||
"""Execution status."""
|
||||
SUCCESS = "success"
|
||||
FAILED = "failed"
|
||||
TIMEOUT = "timeout"
|
||||
ERROR = "error"
|
||||
|
||||
|
||||
class LanguageExecutor:
|
||||
"""Multi-language code executor."""
|
||||
|
||||
def __init__(self, workspace_dir: str = None):
|
||||
"""Initialize executor."""
|
||||
self.workspace_dir = workspace_dir or os.getcwd()
|
||||
|
||||
async def execute_code(
|
||||
self,
|
||||
code: str,
|
||||
language: str,
|
||||
timeout: float = 30.0,
|
||||
compile_timeout: float = 10.0,
|
||||
stdin: Optional[str] = None,
|
||||
files: Optional[Dict[str, str]] = None
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Execute code in the specified language.
|
||||
|
||||
Args:
|
||||
code: Code to execute
|
||||
language: Programming language
|
||||
timeout: Execution timeout in seconds
|
||||
compile_timeout: Compilation timeout in seconds
|
||||
stdin: Optional stdin input
|
||||
files: Optional additional files (name -> content)
|
||||
|
||||
Returns:
|
||||
Execution result dictionary
|
||||
"""
|
||||
if language is None:
|
||||
language = "python"
|
||||
language = language.lower()
|
||||
|
||||
# Map language to executor
|
||||
executors = {
|
||||
'python': self._run_python,
|
||||
'python3': self._run_python,
|
||||
'javascript': self._run_javascript,
|
||||
'js': self._run_javascript,
|
||||
'typescript': self._run_typescript,
|
||||
'ts': self._run_typescript,
|
||||
'go': self._run_go,
|
||||
'java': self._run_java,
|
||||
'cpp': self._run_cpp,
|
||||
'c++': self._run_cpp,
|
||||
'rust': self._run_rust,
|
||||
'php': self._run_php,
|
||||
'bash': self._run_bash,
|
||||
'shell': self._run_bash,
|
||||
'sh': self._run_bash,
|
||||
'nodejs': self._run_javascript,
|
||||
'node': self._run_javascript,
|
||||
}
|
||||
|
||||
executor = executors.get(language)
|
||||
if not executor:
|
||||
return {
|
||||
"status": ExecutionStatus.ERROR,
|
||||
"error": f"Unsupported language: {language}. Supported: {', '.join(sorted(set(executors.keys())))}"
|
||||
}
|
||||
|
||||
try:
|
||||
return await executor(code, timeout, compile_timeout, stdin, files or {})
|
||||
except Exception as e:
|
||||
logger.exception(f"Error executing {language} code")
|
||||
return {
|
||||
"status": ExecutionStatus.ERROR,
|
||||
"error": f"Execution failed: {str(e)}"
|
||||
}
|
||||
|
||||
async def _run_command(
|
||||
self,
|
||||
command: str,
|
||||
timeout: float,
|
||||
stdin: Optional[str] = None,
|
||||
cwd: Optional[str] = None,
|
||||
shell: bool = True
|
||||
) -> Dict[str, Any]:
|
||||
"""Run a shell command and return results with proper process management."""
|
||||
process = None
|
||||
try:
|
||||
logger.debug(f'Running command: {command[:100]}...')
|
||||
|
||||
process = await asyncio.create_subprocess_shell(
|
||||
command,
|
||||
stdin=asyncio.subprocess.PIPE if stdin else None,
|
||||
stdout=asyncio.subprocess.PIPE,
|
||||
stderr=asyncio.subprocess.PIPE,
|
||||
cwd=cwd,
|
||||
executable='/bin/bash'
|
||||
)
|
||||
|
||||
# Write stdin if provided
|
||||
if stdin and process.stdin:
|
||||
try:
|
||||
process.stdin.write(stdin.encode())
|
||||
await process.stdin.drain()
|
||||
process.stdin.close()
|
||||
except Exception as e:
|
||||
logger.warning(f"Failed to write stdin: {e}")
|
||||
|
||||
start_time = time.time()
|
||||
|
||||
# Drain both pipes concurrently with the wait. Reading only *after*
|
||||
# process.wait() deadlocks as soon as the child fills the OS pipe
|
||||
# buffer (~256 KB here): the child blocks in write(), so it never
|
||||
# exits and wait() never returns, turning a fast program with large
|
||||
# stdout into a bogus TIMEOUT.
|
||||
stdout_task = asyncio.ensure_future(get_all_output(process.stdout))
|
||||
stderr_task = asyncio.ensure_future(get_all_output(process.stderr))
|
||||
|
||||
try:
|
||||
# Wait for process with timeout
|
||||
await asyncio.wait_for(process.wait(), timeout=timeout)
|
||||
execution_time = time.time() - start_time
|
||||
|
||||
stdout = await stdout_task
|
||||
stderr = await stderr_task
|
||||
|
||||
logger.debug(f'Command completed in {execution_time:.2f}s')
|
||||
|
||||
return {
|
||||
"status": ExecutionStatus.SUCCESS if process.returncode == 0 else ExecutionStatus.FAILED,
|
||||
"returncode": process.returncode,
|
||||
"stdout": stdout,
|
||||
"stderr": stderr,
|
||||
"execution_time": execution_time
|
||||
}
|
||||
|
||||
except asyncio.TimeoutError:
|
||||
execution_time = time.time() - start_time
|
||||
|
||||
# Kill first so pipes close, then drain remaining output
|
||||
if psutil.pid_exists(process.pid):
|
||||
kill_process_tree(process.pid)
|
||||
logger.info(f'Process {process.pid} killed due to timeout')
|
||||
|
||||
stdout = await stdout_task
|
||||
stderr = await stderr_task
|
||||
|
||||
return {
|
||||
"status": ExecutionStatus.TIMEOUT,
|
||||
"error": f"Execution timed out after {timeout} seconds",
|
||||
"stdout": stdout,
|
||||
"stderr": stderr,
|
||||
"execution_time": execution_time
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
logger.exception(f"Error running command: {command[:100]}")
|
||||
return {
|
||||
"status": ExecutionStatus.ERROR,
|
||||
"error": f"Command execution failed: {str(e)}"
|
||||
}
|
||||
finally:
|
||||
# Cleanup: ensure process is terminated
|
||||
if process and psutil.pid_exists(process.pid):
|
||||
kill_process_tree(process.pid)
|
||||
|
||||
def _write_files(self, tmp_dir: str, files: Dict[str, str]):
|
||||
"""Write additional files to tmp directory."""
|
||||
for filename, content in files.items():
|
||||
if not content or "IGNORE_THIS_FILE" in filename:
|
||||
continue
|
||||
|
||||
filepath = os.path.join(tmp_dir, filename)
|
||||
dirpath = os.path.dirname(filepath)
|
||||
|
||||
if dirpath:
|
||||
os.makedirs(dirpath, exist_ok=True)
|
||||
|
||||
# Handle base64 encoded content
|
||||
try:
|
||||
if self._is_base64(content):
|
||||
with open(filepath, 'wb') as f:
|
||||
f.write(base64.b64decode(content))
|
||||
else:
|
||||
with open(filepath, 'w', encoding='utf-8') as f:
|
||||
f.write(content)
|
||||
except Exception as e:
|
||||
logger.warning(f"Failed to write file {filename}: {e}")
|
||||
|
||||
def _is_base64(self, s: str) -> bool:
|
||||
"""Check if string is base64 encoded."""
|
||||
try:
|
||||
if len(s) % 4 != 0 or not all(c in 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/=' for c in s):
|
||||
return False
|
||||
base64.b64decode(s, validate=True)
|
||||
return True
|
||||
except Exception:
|
||||
return False
|
||||
|
||||
async def _run_python(
|
||||
self,
|
||||
code: str,
|
||||
timeout: float,
|
||||
compile_timeout: float,
|
||||
stdin: Optional[str],
|
||||
files: Dict[str, str]
|
||||
) -> Dict[str, Any]:
|
||||
"""Execute Python code."""
|
||||
with tempfile.TemporaryDirectory(prefix='python_', ignore_cleanup_errors=True) as tmp_dir:
|
||||
self._write_files(tmp_dir, files)
|
||||
code_file = os.path.join(tmp_dir, 'main.py')
|
||||
with open(code_file, 'w', encoding='utf-8') as f:
|
||||
f.write(code)
|
||||
|
||||
# Run untrusted Python in a real container boundary when Docker is
|
||||
# available: no network, read-only rootfs, bounded memory/CPU/PIDs,
|
||||
# and only the one ephemeral work directory mounted writable.
|
||||
if shutil.which("docker"):
|
||||
mount = shlex.quote(f"{tmp_dir}:/workspace:rw")
|
||||
command = (
|
||||
"docker run --rm --network none --memory 256m --cpus 1 "
|
||||
"--pids-limit 64 --read-only "
|
||||
"--tmpfs /tmp:rw,nosuid,nodev,noexec,size=16m "
|
||||
f"-v {mount} -w /workspace python:3.11-slim "
|
||||
"python -I -B -u main.py"
|
||||
)
|
||||
result = await self._run_command(command, timeout, stdin, tmp_dir)
|
||||
result["sandbox"] = {
|
||||
"kind": "docker",
|
||||
"image": "python:3.11-slim",
|
||||
"network": "none",
|
||||
"rootfs": "read-only",
|
||||
"memory": "256m",
|
||||
"cpus": 1,
|
||||
"pids_limit": 64,
|
||||
}
|
||||
else:
|
||||
result = await self._run_command(
|
||||
f'python3 -I -B -u {shlex.quote(code_file)}',
|
||||
timeout,
|
||||
stdin,
|
||||
tmp_dir
|
||||
)
|
||||
result["sandbox"] = {"kind": "local-process", "degraded": True}
|
||||
result['language'] = 'python'
|
||||
return result
|
||||
|
||||
async def _run_javascript(
|
||||
self,
|
||||
code: str,
|
||||
timeout: float,
|
||||
compile_timeout: float,
|
||||
stdin: Optional[str],
|
||||
files: Dict[str, str]
|
||||
) -> Dict[str, Any]:
|
||||
"""Execute JavaScript code with Node.js."""
|
||||
with tempfile.TemporaryDirectory(prefix='js_', ignore_cleanup_errors=True) as tmp_dir:
|
||||
self._write_files(tmp_dir, files)
|
||||
|
||||
# Create package.json if not exists to enable ES modules
|
||||
if 'package.json' not in files:
|
||||
package_json = {
|
||||
"type": "module",
|
||||
"dependencies": {}
|
||||
}
|
||||
with open(os.path.join(tmp_dir, 'package.json'), 'w') as f:
|
||||
import json
|
||||
json.dump(package_json, f)
|
||||
|
||||
code_file = os.path.join(tmp_dir, 'main.js')
|
||||
with open(code_file, 'w', encoding='utf-8') as f:
|
||||
f.write(code)
|
||||
|
||||
result = await self._run_command(
|
||||
f'node {code_file}',
|
||||
timeout,
|
||||
stdin,
|
||||
tmp_dir
|
||||
)
|
||||
result['language'] = 'javascript'
|
||||
return result
|
||||
|
||||
async def _run_typescript(
|
||||
self,
|
||||
code: str,
|
||||
timeout: float,
|
||||
compile_timeout: float,
|
||||
stdin: Optional[str],
|
||||
files: Dict[str, str]
|
||||
) -> Dict[str, Any]:
|
||||
"""Execute TypeScript code with tsx."""
|
||||
with tempfile.TemporaryDirectory(prefix='ts_', ignore_cleanup_errors=True) as tmp_dir:
|
||||
self._write_files(tmp_dir, files)
|
||||
code_file = os.path.join(tmp_dir, 'main.ts')
|
||||
with open(code_file, 'w', encoding='utf-8') as f:
|
||||
f.write(code)
|
||||
|
||||
# Check if tsx is available, fallback to ts-node
|
||||
check_tsx = await self._run_command('which tsx 2>/dev/null', 1.0)
|
||||
cmd = 'tsx' if check_tsx['status'] == ExecutionStatus.SUCCESS else 'ts-node'
|
||||
|
||||
result = await self._run_command(
|
||||
f'{cmd} {code_file}',
|
||||
timeout,
|
||||
stdin,
|
||||
tmp_dir
|
||||
)
|
||||
result['language'] = 'typescript'
|
||||
return result
|
||||
|
||||
async def _run_go(
|
||||
self,
|
||||
code: str,
|
||||
timeout: float,
|
||||
compile_timeout: float,
|
||||
stdin: Optional[str],
|
||||
files: Dict[str, str]
|
||||
) -> Dict[str, Any]:
|
||||
"""Execute Go code."""
|
||||
with tempfile.TemporaryDirectory(prefix='go_', ignore_cleanup_errors=True) as tmp_dir:
|
||||
self._write_files(tmp_dir, files)
|
||||
|
||||
# Initialize go module (ignore errors if already exists)
|
||||
await self._run_command('go mod init main 2>/dev/null || true', 2.0, cwd=tmp_dir)
|
||||
|
||||
code_file = os.path.join(tmp_dir, 'main.go')
|
||||
with open(code_file, 'w', encoding='utf-8') as f:
|
||||
f.write(code)
|
||||
|
||||
# Compile
|
||||
compile_result = await self._run_command(
|
||||
'go build -o main main.go',
|
||||
compile_timeout,
|
||||
cwd=tmp_dir
|
||||
)
|
||||
|
||||
if compile_result['status'] != ExecutionStatus.SUCCESS:
|
||||
return {
|
||||
"status": ExecutionStatus.FAILED,
|
||||
"language": "go",
|
||||
"phase": "compilation",
|
||||
"returncode": compile_result.get('returncode', 1),
|
||||
"stdout": compile_result.get('stdout', ''),
|
||||
"stderr": compile_result.get('stderr', ''),
|
||||
"error": "Compilation failed"
|
||||
}
|
||||
|
||||
# Run
|
||||
result = await self._run_command(
|
||||
'./main',
|
||||
timeout,
|
||||
stdin,
|
||||
tmp_dir
|
||||
)
|
||||
result['language'] = 'go'
|
||||
result['compile_stdout'] = compile_result.get('stdout', '')
|
||||
result['compile_stderr'] = compile_result.get('stderr', '')
|
||||
return result
|
||||
|
||||
async def _run_java(
|
||||
self,
|
||||
code: str,
|
||||
timeout: float,
|
||||
compile_timeout: float,
|
||||
stdin: Optional[str],
|
||||
files: Dict[str, str]
|
||||
) -> Dict[str, Any]:
|
||||
"""Execute Java code."""
|
||||
with tempfile.TemporaryDirectory(prefix='java_', ignore_cleanup_errors=True) as tmp_dir:
|
||||
self._write_files(tmp_dir, files)
|
||||
|
||||
# Extract class name from public class declaration
|
||||
class_name = 'Main'
|
||||
import re
|
||||
match = re.search(r'public\s+class\s+(\w+)', code)
|
||||
if match:
|
||||
class_name = match.group(1)
|
||||
|
||||
code_file = os.path.join(tmp_dir, f'{class_name}.java')
|
||||
with open(code_file, 'w', encoding='utf-8') as f:
|
||||
f.write(code)
|
||||
|
||||
# Prepare classpath for additional jars
|
||||
jars = [f for f in files.keys() if f.endswith('.jar')]
|
||||
classpath = '.:' + ':'.join(jars) if jars else '.'
|
||||
|
||||
# Compile
|
||||
compile_result = await self._run_command(
|
||||
f'javac -cp {classpath} {class_name}.java',
|
||||
compile_timeout,
|
||||
cwd=tmp_dir
|
||||
)
|
||||
|
||||
if compile_result['status'] != ExecutionStatus.SUCCESS:
|
||||
return {
|
||||
"status": ExecutionStatus.FAILED,
|
||||
"language": "java",
|
||||
"phase": "compilation",
|
||||
"returncode": compile_result.get('returncode', 1),
|
||||
"stdout": compile_result.get('stdout', ''),
|
||||
"stderr": compile_result.get('stderr', ''),
|
||||
"error": "Compilation failed"
|
||||
}
|
||||
|
||||
# Run with assertions enabled
|
||||
result = await self._run_command(
|
||||
f'java -cp {classpath} -ea {class_name}',
|
||||
timeout,
|
||||
stdin,
|
||||
tmp_dir
|
||||
)
|
||||
result['language'] = 'java'
|
||||
result['compile_stdout'] = compile_result.get('stdout', '')
|
||||
result['compile_stderr'] = compile_result.get('stderr', '')
|
||||
return result
|
||||
|
||||
async def _run_cpp(
|
||||
self,
|
||||
code: str,
|
||||
timeout: float,
|
||||
compile_timeout: float,
|
||||
stdin: Optional[str],
|
||||
files: Dict[str, str]
|
||||
) -> Dict[str, Any]:
|
||||
"""Execute C++ code."""
|
||||
with tempfile.TemporaryDirectory(prefix='cpp_', ignore_cleanup_errors=True) as tmp_dir:
|
||||
self._write_files(tmp_dir, files)
|
||||
code_file = os.path.join(tmp_dir, 'main.cpp')
|
||||
with open(code_file, 'w', encoding='utf-8') as f:
|
||||
f.write(code)
|
||||
|
||||
# Compile with commonly needed flags
|
||||
# Try with optional libraries (crypto, ssl, pthread)
|
||||
compile_flags = '-std=c++17 -O2'
|
||||
optional_libs = []
|
||||
|
||||
# Check if we need pthread
|
||||
if '#include <thread>' in code or 'std::thread' in code:
|
||||
optional_libs.append('-lpthread')
|
||||
|
||||
libs = ' '.join(optional_libs)
|
||||
compile_result = await self._run_command(
|
||||
f'g++ {compile_flags} main.cpp -o main {libs}',
|
||||
compile_timeout,
|
||||
cwd=tmp_dir
|
||||
)
|
||||
|
||||
if compile_result['status'] != ExecutionStatus.SUCCESS:
|
||||
return {
|
||||
"status": ExecutionStatus.FAILED,
|
||||
"language": "cpp",
|
||||
"phase": "compilation",
|
||||
"returncode": compile_result.get('returncode', 1),
|
||||
"stdout": compile_result.get('stdout', ''),
|
||||
"stderr": compile_result.get('stderr', ''),
|
||||
"error": "Compilation failed"
|
||||
}
|
||||
|
||||
# Run
|
||||
result = await self._run_command(
|
||||
'./main',
|
||||
timeout,
|
||||
stdin,
|
||||
tmp_dir
|
||||
)
|
||||
result['language'] = 'cpp'
|
||||
result['compile_stdout'] = compile_result.get('stdout', '')
|
||||
result['compile_stderr'] = compile_result.get('stderr', '')
|
||||
return result
|
||||
|
||||
async def _run_rust(
|
||||
self,
|
||||
code: str,
|
||||
timeout: float,
|
||||
compile_timeout: float,
|
||||
stdin: Optional[str],
|
||||
files: Dict[str, str]
|
||||
) -> Dict[str, Any]:
|
||||
"""Execute Rust code."""
|
||||
with tempfile.TemporaryDirectory(prefix='rust_', ignore_cleanup_errors=True) as tmp_dir:
|
||||
self._write_files(tmp_dir, files)
|
||||
code_file = os.path.join(tmp_dir, 'main.rs')
|
||||
with open(code_file, 'w', encoding='utf-8') as f:
|
||||
f.write(code)
|
||||
|
||||
# Compile with optimizations
|
||||
compile_result = await self._run_command(
|
||||
'rustc -O main.rs -o main',
|
||||
compile_timeout,
|
||||
cwd=tmp_dir
|
||||
)
|
||||
|
||||
if compile_result['status'] != ExecutionStatus.SUCCESS:
|
||||
return {
|
||||
"status": ExecutionStatus.FAILED,
|
||||
"language": "rust",
|
||||
"phase": "compilation",
|
||||
"returncode": compile_result.get('returncode', 1),
|
||||
"stdout": compile_result.get('stdout', ''),
|
||||
"stderr": compile_result.get('stderr', ''),
|
||||
"error": "Compilation failed"
|
||||
}
|
||||
|
||||
# Run
|
||||
result = await self._run_command(
|
||||
'./main',
|
||||
timeout,
|
||||
stdin,
|
||||
tmp_dir
|
||||
)
|
||||
result['language'] = 'rust'
|
||||
result['compile_stdout'] = compile_result.get('stdout', '')
|
||||
result['compile_stderr'] = compile_result.get('stderr', '')
|
||||
return result
|
||||
|
||||
async def _run_php(
|
||||
self,
|
||||
code: str,
|
||||
timeout: float,
|
||||
compile_timeout: float,
|
||||
stdin: Optional[str],
|
||||
files: Dict[str, str]
|
||||
) -> Dict[str, Any]:
|
||||
"""Execute PHP code."""
|
||||
with tempfile.TemporaryDirectory(prefix='php_', ignore_cleanup_errors=True) as tmp_dir:
|
||||
self._write_files(tmp_dir, files)
|
||||
|
||||
# Ensure PHP tags
|
||||
code_clean = code.strip()
|
||||
if not code_clean.startswith('<?php') and not code_clean.startswith('<?'):
|
||||
code = '<?php\n' + code
|
||||
|
||||
code_file = os.path.join(tmp_dir, 'main.php')
|
||||
with open(code_file, 'w', encoding='utf-8') as f:
|
||||
f.write(code)
|
||||
|
||||
result = await self._run_command(
|
||||
f'php -f {code_file}',
|
||||
timeout,
|
||||
stdin,
|
||||
tmp_dir
|
||||
)
|
||||
result['language'] = 'php'
|
||||
return result
|
||||
|
||||
async def _run_bash(
|
||||
self,
|
||||
code: str,
|
||||
timeout: float,
|
||||
compile_timeout: float,
|
||||
stdin: Optional[str],
|
||||
files: Dict[str, str]
|
||||
) -> Dict[str, Any]:
|
||||
"""Execute Bash script."""
|
||||
with tempfile.TemporaryDirectory(prefix='bash_', ignore_cleanup_errors=True) as tmp_dir:
|
||||
self._write_files(tmp_dir, files)
|
||||
code_file = os.path.join(tmp_dir, 'script.sh')
|
||||
with open(code_file, 'w', encoding='utf-8') as f:
|
||||
# Add shebang if not present
|
||||
if not code.startswith('#!'):
|
||||
f.write('#!/bin/bash\n')
|
||||
f.write(code)
|
||||
|
||||
# Make executable
|
||||
os.chmod(code_file, 0o755)
|
||||
|
||||
result = await self._run_command(
|
||||
f'bash {code_file}',
|
||||
timeout,
|
||||
stdin,
|
||||
tmp_dir
|
||||
)
|
||||
result['language'] = 'bash'
|
||||
return result
|
||||
@@ -0,0 +1,95 @@
|
||||
"""Quick start guide for the execution tools MCP server."""
|
||||
|
||||
import asyncio
|
||||
from llm_helper import LLMHelper
|
||||
from file_tools import FileTools
|
||||
from execution_tools import ExecutionTools
|
||||
|
||||
|
||||
async def quickstart():
|
||||
"""Quick demonstration of the execution tools."""
|
||||
print("=== Execution Tools MCP Server - Quick Start ===\n")
|
||||
|
||||
# Initialize
|
||||
print("Initializing tools...")
|
||||
llm_helper = LLMHelper()
|
||||
file_tools = FileTools(llm_helper)
|
||||
execution_tools = ExecutionTools(llm_helper)
|
||||
|
||||
# 1. File operations
|
||||
print("\n1. File Operations Demo")
|
||||
print("-" * 50)
|
||||
|
||||
print("\nWriting a Python script...")
|
||||
result = await file_tools.write_file(
|
||||
path="hello.py",
|
||||
content="""#!/usr/bin/env python3
|
||||
\"\"\"A simple greeting script.\"\"\"
|
||||
|
||||
def main():
|
||||
name = "World"
|
||||
print(f"Hello, {name}!")
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
""",
|
||||
overwrite=True
|
||||
)
|
||||
print(f"Status: {'✓ Success' if result['success'] else '✗ Failed'}")
|
||||
if result['success']:
|
||||
print(f"Written to: {result['path']}")
|
||||
print(f"Verification: {result['verification']}")
|
||||
|
||||
# 2. Code execution
|
||||
print("\n2. Code Interpreter Demo")
|
||||
print("-" * 50)
|
||||
|
||||
print("\nExecuting Python code...")
|
||||
result = await execution_tools.code_interpreter(
|
||||
code="""
|
||||
# Calculate fibonacci sequence
|
||||
def fibonacci(n):
|
||||
if n <= 1:
|
||||
return n
|
||||
return fibonacci(n-1) + fibonacci(n-2)
|
||||
|
||||
print("Fibonacci sequence (first 10 numbers):")
|
||||
for i in range(10):
|
||||
print(f"F({i}) = {fibonacci(i)}")
|
||||
"""
|
||||
)
|
||||
print(f"Status: {'✓ Success' if result['success'] else '✗ Failed'}")
|
||||
if result['success']:
|
||||
print("Output:")
|
||||
print(result['stdout'][:500]) # Print first 500 chars
|
||||
|
||||
# 3. Terminal execution
|
||||
print("\n3. Virtual Terminal Demo")
|
||||
print("-" * 50)
|
||||
|
||||
print("\nExecuting shell command...")
|
||||
result = await execution_tools.virtual_terminal(
|
||||
command="python --version && echo 'Current directory:' && pwd"
|
||||
)
|
||||
print(f"Status: {'✓ Success' if result['success'] else '✗ Failed'}")
|
||||
if result['success']:
|
||||
print("Output:")
|
||||
print(result['stdout'])
|
||||
|
||||
# Summary
|
||||
print("\n" + "=" * 50)
|
||||
print("Quick start completed!")
|
||||
print("\nKey Features:")
|
||||
print(" • File operations with automatic syntax verification")
|
||||
print(" • Code execution with error analysis")
|
||||
print(" • Shell commands with result summarization")
|
||||
print(" • LLM-based approval for dangerous operations")
|
||||
print(" • External integrations (Google Calendar, GitHub)")
|
||||
print("\nNext Steps:")
|
||||
print(" • Run 'python examples.py' for more examples")
|
||||
print(" • Run 'python server.py' to start the MCP server")
|
||||
print(" • See README.md for full documentation")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
asyncio.run(quickstart())
|
||||
@@ -0,0 +1,64 @@
|
||||
# MCP and Core Dependencies
|
||||
mcp>=0.9.0
|
||||
openai>=1.0.0
|
||||
httpx>=0.24.0
|
||||
python-dotenv>=1.0.0
|
||||
|
||||
# External API Integration
|
||||
google-auth>=2.0.0
|
||||
google-auth-oauthlib>=1.0.0
|
||||
google-auth-httplib2>=0.1.0
|
||||
google-api-python-client>=2.0.0
|
||||
PyGithub>=2.0.0
|
||||
|
||||
# Process Management
|
||||
subprocess-tee>=0.4.0
|
||||
psutil>=5.9.8
|
||||
|
||||
# Scientific Computing - Core
|
||||
numpy>=1.26.0
|
||||
scipy>=1.11.0
|
||||
pandas>=2.2.0
|
||||
matplotlib>=3.8.0
|
||||
seaborn>=0.13.0
|
||||
|
||||
# Machine Learning & AI
|
||||
scikit-learn>=1.4.0
|
||||
xgboost>=2.0.0
|
||||
lightgbm>=4.3.0
|
||||
|
||||
# Deep Learning (Optional, comment out if not needed)
|
||||
# torch>=2.1.0
|
||||
# tensorflow>=2.14.0
|
||||
|
||||
# Data Processing & Analysis
|
||||
openpyxl>=3.1.0
|
||||
xlrd>=2.0.0
|
||||
XlsxWriter>=3.1.0
|
||||
beautifulsoup4>=4.12.0
|
||||
lxml>=5.1.0
|
||||
|
||||
# Scientific Libraries
|
||||
statsmodels>=0.14.0
|
||||
scikit-image>=0.22.0
|
||||
pillow>=10.2.0
|
||||
opencv-python>=4.9.0
|
||||
plotly>=5.18.0
|
||||
|
||||
# Document Processing
|
||||
PyPDF2>=3.0.0
|
||||
python-docx>=1.1.0
|
||||
PyMuPDF>=1.24.0
|
||||
|
||||
# Testing & Development
|
||||
pytest>=7.4.0
|
||||
pytest-asyncio>=0.23.0
|
||||
pytest-cov>=5.0.0
|
||||
pytest-mock>=3.12.0
|
||||
|
||||
# Utilities
|
||||
requests>=2.31.0
|
||||
pydantic>=2.6.0
|
||||
structlog>=23.1.0
|
||||
tabulate>=0.9.0
|
||||
colorama>=0.4.4
|
||||
@@ -0,0 +1,278 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Run the manuscript-scope Experiment 4-3 campaign over real MCP stdio."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
import asyncio
|
||||
import hashlib
|
||||
import json
|
||||
import os
|
||||
import re
|
||||
import shutil
|
||||
import sys
|
||||
import time
|
||||
from datetime import datetime, timezone
|
||||
from pathlib import Path
|
||||
from typing import Any
|
||||
|
||||
from mcp import ClientSession, StdioServerParameters
|
||||
from mcp.client.stdio import stdio_client
|
||||
|
||||
HERE = Path(__file__).resolve().parent
|
||||
PROTOCOL = HERE / "experiment_protocol.json"
|
||||
SERVER = HERE / "server.py"
|
||||
VALIDATION = HERE / "validation" / "experiment_4_3"
|
||||
CREDENTIAL = re.compile(r"\b(?:sk|gh[opusr])-[A-Za-z0-9_-]{12,}\b")
|
||||
|
||||
|
||||
def sha(path: Path) -> str:
|
||||
return hashlib.sha256(path.read_bytes()).hexdigest()
|
||||
|
||||
|
||||
def write_json(path: Path, value: Any) -> None:
|
||||
text = json.dumps(value, ensure_ascii=False, indent=2, default=str) + "\n"
|
||||
if CREDENTIAL.search(text):
|
||||
raise ValueError(f"credential-shaped string in {path}")
|
||||
path.parent.mkdir(parents=True, exist_ok=True)
|
||||
path.write_text(text, encoding="utf-8")
|
||||
|
||||
|
||||
def unwrap(result: Any) -> Any:
|
||||
structured = getattr(result, "structuredContent", None) or getattr(result, "structured_content", None)
|
||||
if structured:
|
||||
return structured
|
||||
texts = [getattr(item, "text", None) for item in getattr(result, "content", [])]
|
||||
texts = [item for item in texts if item]
|
||||
if len(texts) == 1:
|
||||
try:
|
||||
return json.loads(texts[0])
|
||||
except json.JSONDecodeError:
|
||||
return texts[0]
|
||||
return texts
|
||||
|
||||
|
||||
async def run(
|
||||
campaign_id: str,
|
||||
android_container: str,
|
||||
github_head_branch: str,
|
||||
github_base_branch: str,
|
||||
) -> Path:
|
||||
run_dir = VALIDATION / campaign_id
|
||||
run_dir.mkdir(parents=True, exist_ok=False)
|
||||
workspace = run_dir / "workspace"
|
||||
workspace.mkdir()
|
||||
outside = run_dir / "outside-witness.txt"
|
||||
outside.write_text("MUST-NOT-CHANGE\n", encoding="utf-8")
|
||||
outside_before = sha(outside)
|
||||
write_json(run_dir / "protocol.json", json.loads(PROTOCOL.read_text(encoding="utf-8")))
|
||||
|
||||
env = os.environ.copy()
|
||||
if env.get("KIMI_API_KEY") or env.get("MOONSHOT_API_KEY"):
|
||||
review_provider = "kimi"
|
||||
review_model = "kimi-k3"
|
||||
else:
|
||||
review_provider = "openrouter"
|
||||
review_model = "openai/gpt-4.1-mini"
|
||||
env.update({
|
||||
"WORKSPACE_DIR": str(workspace),
|
||||
"REQUIRE_APPROVAL_FOR_DANGEROUS_OPS": "true",
|
||||
"AUTO_VERIFY_CODE": "true",
|
||||
"AUTO_SUMMARIZE_COMPLEX_OUTPUT": "false",
|
||||
"EXECUTION_LLM_RECEIPT_PATH": str(run_dir / "llm_receipts.checkpoint.json"),
|
||||
"PROVIDER": review_provider,
|
||||
"MODEL": review_model,
|
||||
"ANDROID_WORLD_CONTAINER": android_container,
|
||||
})
|
||||
parameters = StdioServerParameters(command=sys.executable, args=[str(SERVER)], env=env)
|
||||
receipts: list[dict[str, Any]] = []
|
||||
|
||||
async with stdio_client(parameters) as (read, write):
|
||||
async with ClientSession(read, write) as session:
|
||||
initialized = await session.initialize()
|
||||
listed = await session.list_tools()
|
||||
schemas = [tool.model_dump(by_alias=True, exclude_none=True, mode="json") for tool in listed.tools]
|
||||
write_json(run_dir / "catalog.json", {
|
||||
"transport": "mcp-stdio", "server_name": initialized.serverInfo.name,
|
||||
"server_version": initialized.serverInfo.version, "schemas": schemas,
|
||||
"schema_sha256": hashlib.sha256(json.dumps(schemas, sort_keys=True).encode()).hexdigest(),
|
||||
})
|
||||
|
||||
async def call(case: str, tool: str, arguments: dict[str, Any]) -> dict[str, Any]:
|
||||
started = time.perf_counter()
|
||||
try:
|
||||
result = await session.call_tool(tool, arguments=arguments)
|
||||
payload = unwrap(result)
|
||||
is_error = bool(getattr(result, "isError", False) or getattr(result, "is_error", False))
|
||||
except Exception as exc:
|
||||
payload, is_error = {"success": False, "error": f"{type(exc).__name__}: {exc}"}, True
|
||||
row = {
|
||||
"case": case, "tool": tool, "arguments": arguments,
|
||||
"transport": "mcp-stdio", "mcp_result_is_error": is_error,
|
||||
"payload": payload,
|
||||
"latency_seconds": round(time.perf_counter() - started, 3),
|
||||
}
|
||||
receipts.append(row)
|
||||
write_json(run_dir / "receipts" / f"{len(receipts):02d}_{case}.json", row)
|
||||
return row
|
||||
|
||||
await call("python_valid_write", "file_write", {
|
||||
"path": "valid.py", "content": "def add(a, b):\n return a + b\n", "overwrite": True})
|
||||
await call("python_invalid_rejected", "file_write", {
|
||||
"path": "invalid.py", "content": "def broken(:\n pass\n", "overwrite": True})
|
||||
await call("javascript_valid_write", "file_write", {
|
||||
"path": "valid.js", "content": "const answer = 42;\nconsole.log(answer);\n", "overwrite": True})
|
||||
await call("javascript_invalid_rejected", "file_write", {
|
||||
"path": "invalid.js", "content": "const broken = ;\n", "overwrite": True})
|
||||
await call("verified_edit", "file_edit", {
|
||||
"path": "valid.py", "search": "a + b", "replace": "a - b"})
|
||||
await call("path_escape_rejected", "file_write", {
|
||||
"path": "../../escape.py", "content": "print('escape')\n", "overwrite": True})
|
||||
await call("terminal_safe", "virtual_terminal", {"command": "pwd && printf SAFE", "timeout": 10})
|
||||
await call("terminal_timeout", "virtual_terminal", {"command": "sleep 2", "timeout": 1})
|
||||
await call("terminal_danger_rejected", "virtual_terminal", {
|
||||
"command": "rm -rf ./should-never-execute", "timeout": 10})
|
||||
await call("python_docker_sandbox", "code_interpreter", {
|
||||
"language": "python", "timeout": 30,
|
||||
"code": "import os, json\nprint(json.dumps({'root': os.listdir('/'), 'network_proxy': os.environ.get('HTTPS_PROXY')}))\n"})
|
||||
await call("python_network_denied", "code_interpreter", {
|
||||
"language": "python", "timeout": 30,
|
||||
"code": "import urllib.request\ntry:\n print(urllib.request.urlopen('https://example.com', timeout=3).status)\nexcept Exception as e:\n print(type(e).__name__, str(e))\n"})
|
||||
await call("long_output_persisted", "code_interpreter", {
|
||||
"language": "python", "timeout": 30,
|
||||
"code": "for i in range(260): print(f'LINE-{i:03d}')\n"})
|
||||
await call("excel_formula_screenshot", "excel_create_with_formula_and_screenshot", {
|
||||
"output_path": "invoice.xlsx", "rows": [
|
||||
{"item": "Compute", "quantity": 2, "unit_price": 12.5},
|
||||
{"item": "Storage", "quantity": 3, "unit_price": 7.0}]})
|
||||
await call("real_webhook", "webhook_post", {
|
||||
"url": "https://postman-echo.com/post",
|
||||
"payload": {"experiment": "4-3", "marker": "REAL-WEBHOOK-RECEIPT"}})
|
||||
await call("real_browser", "browser_navigate", {
|
||||
"url": "https://example.com", "screenshot_path": "browser-example.png"})
|
||||
await call("calendar_preflight", "google_calendar_add", {
|
||||
"summary": "Experiment 4-3", "start_time": "2026-08-01T10:00:00+00:00",
|
||||
"end_time": "2026-08-01T10:30:00+00:00"})
|
||||
await call("github_pr_preflight", "github_create_pr", {
|
||||
"repo_name": "bojieli/ai-agent-book",
|
||||
"title": "feat(ch4): build Experiment 4-3 GUI environments",
|
||||
"body": "Experiment 4-3 evidence: real Android and X11 Computer Use execution.",
|
||||
"head_branch": github_head_branch, "base_branch": github_base_branch})
|
||||
await call("real_virtual_desktop", "virtual_desktop_execute", {
|
||||
"url": "https://example.com", "screenshot_path": "computer-use-example.png",
|
||||
"expected_title": "Example Domain"})
|
||||
await call("real_virtual_mobile", "virtual_mobile_execute", {
|
||||
"container_name": android_container, "screenshot_path": "android-wifi-settings.png"})
|
||||
await call("desktop_mobile_capabilities", "environment_capabilities", {})
|
||||
|
||||
by_case = {row["case"]: row["payload"] for row in receipts}
|
||||
llm_path = run_dir / "llm_receipts.checkpoint.json"
|
||||
llm_receipts = json.loads(llm_path.read_text(encoding="utf-8")) if llm_path.is_file() else []
|
||||
write_json(run_dir / "llm_receipts.json", llm_receipts)
|
||||
long_path = by_case.get("long_output_persisted", {}).get("stdout_file")
|
||||
long_file = Path(long_path) if long_path else None
|
||||
retained_long_file = run_dir / "artifacts" / "long_output.full.txt"
|
||||
if long_file and long_file.is_file():
|
||||
retained_long_file.parent.mkdir(parents=True, exist_ok=True)
|
||||
shutil.copyfile(long_file, retained_long_file)
|
||||
long_evidence = {
|
||||
"path": str(retained_long_file.relative_to(run_dir)),
|
||||
"bytes": retained_long_file.stat().st_size,
|
||||
"sha256": sha(retained_long_file),
|
||||
"source_temp_path_sha256": hashlib.sha256(str(long_file).encode()).hexdigest(),
|
||||
}
|
||||
else:
|
||||
long_evidence = None
|
||||
caps = by_case.get("desktop_mobile_capabilities", {})
|
||||
gates = {
|
||||
"real_mcp_catalog_and_calls": len(schemas) >= 12 and len(receipts) == 20,
|
||||
"python_and_javascript_linter": (
|
||||
by_case["python_valid_write"].get("verification") == "passed"
|
||||
and by_case["javascript_valid_write"].get("verification") == "passed"
|
||||
and by_case["python_invalid_rejected"].get("success") is False
|
||||
and by_case["javascript_invalid_rejected"].get("success") is False),
|
||||
"file_edit_verified_and_escape_rejected": (
|
||||
by_case["verified_edit"].get("success") is True
|
||||
and by_case["path_escape_rejected"].get("success") is False
|
||||
and outside_before == sha(outside)),
|
||||
"terminal_timeout_and_llm_danger_review": (
|
||||
by_case["terminal_safe"].get("success") is True
|
||||
and by_case["terminal_timeout"].get("success") is False
|
||||
and by_case["terminal_danger_rejected"].get("success") is False
|
||||
and any(row.get("purpose") == "dangerous_operation_review"
|
||||
and row.get("response", {}).get("id") and row.get("usage", {}).get("total_tokens")
|
||||
and row.get("latency_seconds") is not None for row in llm_receipts)),
|
||||
"real_python_container_sandbox": (
|
||||
by_case["python_docker_sandbox"].get("success") is True
|
||||
and by_case["python_docker_sandbox"].get("sandbox", {}).get("kind") == "docker"
|
||||
and by_case["python_network_denied"].get("success") is True
|
||||
and "URLError" in by_case["python_network_denied"].get("stdout", "")),
|
||||
"long_output_truncated_and_persisted": bool(
|
||||
long_evidence and "省略" in by_case["long_output_persisted"].get("stdout", "")),
|
||||
"real_excel_formula_and_screenshot": by_case["excel_formula_screenshot"].get("success") is True,
|
||||
"real_webhook": by_case["real_webhook"].get("success") is True,
|
||||
"real_browser": by_case["real_browser"].get("success") is True,
|
||||
"real_calendar_mutation": by_case["calendar_preflight"].get("success") is True,
|
||||
"real_github_pr_mutation": by_case["github_pr_preflight"].get("success") is True,
|
||||
"real_email_mutation": False,
|
||||
"real_virtual_desktop_session": bool(
|
||||
by_case["real_virtual_desktop"].get("success") is True
|
||||
and by_case["real_virtual_desktop"].get("expected_title_matched") is True
|
||||
and by_case["real_virtual_desktop"].get("screenshot", {}).get("sha256")),
|
||||
"real_virtual_mobile_session": bool(
|
||||
by_case["real_virtual_mobile"].get("success") is True
|
||||
and by_case["real_virtual_mobile"].get("settings_activity")
|
||||
and by_case["real_virtual_mobile"].get("screenshot", {}).get("sha256")
|
||||
and caps.get("android_active_devices")),
|
||||
"credential_free_usage_latency_receipts": bool(llm_receipts) and all(
|
||||
row.get("response", {}).get("id") and row.get("usage", {}).get("total_tokens") is not None
|
||||
and row.get("latency_seconds") is not None for row in llm_receipts),
|
||||
}
|
||||
core_names = [name for name in gates if name not in {
|
||||
"real_calendar_mutation", "real_github_pr_mutation", "real_email_mutation",
|
||||
"real_virtual_desktop_session", "real_virtual_mobile_session"}]
|
||||
status = "passed" if all(gates.values()) else (
|
||||
"blocked" if all(gates[name] for name in core_names) else "failed")
|
||||
summary = {
|
||||
"experiment": "4-3", "campaign_id": campaign_id,
|
||||
"generated_at": datetime.now(timezone.utc).isoformat(),
|
||||
"status": status, "official_complete": status == "passed",
|
||||
"gates": gates, "long_output_full_file": long_evidence,
|
||||
"blockers": [name for name, value in gates.items() if not value],
|
||||
"receipt_count": len(receipts), "llm_call_count": len(llm_receipts),
|
||||
}
|
||||
write_json(run_dir / "summary.json", summary)
|
||||
files = []
|
||||
for path in sorted(run_dir.rglob("*")):
|
||||
if path.is_file() and path.name != "manifest.json":
|
||||
files.append({"path": str(path.relative_to(run_dir)), "bytes": path.stat().st_size,
|
||||
"sha256": sha(path)})
|
||||
manifest = {"experiment": "4-3", "campaign_id": campaign_id,
|
||||
"status": status, "official_complete": status == "passed", "files": files}
|
||||
write_json(run_dir / "manifest.json", manifest)
|
||||
write_json(VALIDATION / "latest.json", {
|
||||
"experiment": "4-3", "campaign_id": campaign_id, "status": status,
|
||||
"official_complete": status == "passed",
|
||||
"manifest": str((run_dir / "manifest.json").relative_to(HERE)),
|
||||
"manifest_sha256": sha(run_dir / "manifest.json")})
|
||||
return run_dir
|
||||
|
||||
|
||||
def main() -> int:
|
||||
parser = argparse.ArgumentParser()
|
||||
parser.add_argument("--campaign-id", default=datetime.now(timezone.utc).strftime("real_mcp_%Y%m%dT%H%M%SZ"))
|
||||
parser.add_argument("--android-container", default=os.getenv("ANDROID_WORLD_CONTAINER", "exp4-3-android"))
|
||||
parser.add_argument("--github-head-branch", default="nonexistent-exp4-3")
|
||||
parser.add_argument("--github-base-branch", default="main")
|
||||
args = parser.parse_args()
|
||||
path = asyncio.run(run(
|
||||
args.campaign_id, args.android_container,
|
||||
args.github_head_branch, args.github_base_branch,
|
||||
))
|
||||
print(path)
|
||||
status = json.loads((path / "summary.json").read_text(encoding="utf-8"))["status"]
|
||||
return 0 if status in {"passed", "blocked"} else 1
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
raise SystemExit(main())
|
||||
@@ -0,0 +1,363 @@
|
||||
"""MCP server for execution tools."""
|
||||
|
||||
import asyncio
|
||||
import json
|
||||
from typing import Any
|
||||
from mcp.server import Server, NotificationOptions
|
||||
from mcp.server.models import InitializationOptions
|
||||
import mcp.server.stdio
|
||||
import mcp.types as types
|
||||
|
||||
from config import Config
|
||||
from llm_helper import LLMHelper
|
||||
from file_tools import FileTools
|
||||
from execution_tools import ExecutionTools
|
||||
from external_tools import ExternalTools
|
||||
from extended_tools import ExtendedTools
|
||||
|
||||
|
||||
# Initialize server
|
||||
server = Server("execution-tools")
|
||||
|
||||
# Initialize tools
|
||||
llm_helper = LLMHelper()
|
||||
file_tools = FileTools(llm_helper)
|
||||
execution_tools = ExecutionTools(llm_helper)
|
||||
external_tools = ExternalTools(llm_helper)
|
||||
extended_tools = ExtendedTools()
|
||||
|
||||
|
||||
@server.list_tools()
|
||||
async def handle_list_tools() -> list[types.Tool]:
|
||||
"""List available tools."""
|
||||
return [
|
||||
types.Tool(
|
||||
name="file_write",
|
||||
description="Write content to a file with automatic syntax verification",
|
||||
inputSchema={
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"path": {
|
||||
"type": "string",
|
||||
"description": "File path (relative to workspace or absolute)"
|
||||
},
|
||||
"content": {
|
||||
"type": "string",
|
||||
"description": "Content to write"
|
||||
},
|
||||
"overwrite": {
|
||||
"type": "boolean",
|
||||
"description": "Whether to overwrite existing files",
|
||||
"default": False
|
||||
}
|
||||
},
|
||||
"required": ["path", "content"]
|
||||
}
|
||||
),
|
||||
types.Tool(
|
||||
name="file_edit",
|
||||
description="Edit an existing file by searching and replacing content",
|
||||
inputSchema={
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"path": {
|
||||
"type": "string",
|
||||
"description": "File path"
|
||||
},
|
||||
"search": {
|
||||
"type": "string",
|
||||
"description": "Text to search for"
|
||||
},
|
||||
"replace": {
|
||||
"type": "string",
|
||||
"description": "Replacement text"
|
||||
}
|
||||
},
|
||||
"required": ["path", "search", "replace"]
|
||||
}
|
||||
),
|
||||
types.Tool(
|
||||
name="code_interpreter",
|
||||
description="Execute code in multiple programming languages in a sandboxed environment with result analysis. Supports: Python, JavaScript, TypeScript, Go, Java, C++, Rust, PHP, Bash",
|
||||
inputSchema={
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"code": {
|
||||
"type": "string",
|
||||
"description": "Code to execute"
|
||||
},
|
||||
"language": {
|
||||
"type": "string",
|
||||
"description": "Programming language (python, javascript, typescript, go, java, cpp, rust, php, bash)",
|
||||
"default": "python"
|
||||
},
|
||||
"timeout": {
|
||||
"type": "number",
|
||||
"description": "Execution timeout in seconds",
|
||||
"default": 30.0
|
||||
},
|
||||
"stdin": {
|
||||
"type": "string",
|
||||
"description": "Optional stdin input for the program"
|
||||
},
|
||||
"files": {
|
||||
"type": "object",
|
||||
"description": "Optional additional files (filename -> content mapping)",
|
||||
"additionalProperties": {"type": "string"}
|
||||
}
|
||||
},
|
||||
"required": ["code"]
|
||||
}
|
||||
),
|
||||
types.Tool(
|
||||
name="virtual_terminal",
|
||||
description="Execute shell commands with error summarization",
|
||||
inputSchema={
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"command": {
|
||||
"type": "string",
|
||||
"description": "Shell command to execute"
|
||||
},
|
||||
"timeout": {
|
||||
"type": "integer",
|
||||
"description": "Timeout in seconds",
|
||||
"default": 30
|
||||
}
|
||||
},
|
||||
"required": ["command"]
|
||||
}
|
||||
),
|
||||
types.Tool(
|
||||
name="google_calendar_add",
|
||||
description="Add an event to Google Calendar",
|
||||
inputSchema={
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"summary": {
|
||||
"type": "string",
|
||||
"description": "Event title"
|
||||
},
|
||||
"start_time": {
|
||||
"type": "string",
|
||||
"description": "Start time (ISO 8601 format, e.g., 2024-01-01T10:00:00)"
|
||||
},
|
||||
"end_time": {
|
||||
"type": "string",
|
||||
"description": "End time (ISO 8601 format)"
|
||||
},
|
||||
"description": {
|
||||
"type": "string",
|
||||
"description": "Event description"
|
||||
},
|
||||
"location": {
|
||||
"type": "string",
|
||||
"description": "Event location"
|
||||
}
|
||||
},
|
||||
"required": ["summary", "start_time", "end_time"]
|
||||
}
|
||||
),
|
||||
types.Tool(
|
||||
name="github_create_pr",
|
||||
description="Create a GitHub Pull Request",
|
||||
inputSchema={
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"repo_name": {
|
||||
"type": "string",
|
||||
"description": "Repository name (format: owner/repo)"
|
||||
},
|
||||
"title": {
|
||||
"type": "string",
|
||||
"description": "PR title"
|
||||
},
|
||||
"body": {
|
||||
"type": "string",
|
||||
"description": "PR description"
|
||||
},
|
||||
"head_branch": {
|
||||
"type": "string",
|
||||
"description": "Source branch"
|
||||
},
|
||||
"base_branch": {
|
||||
"type": "string",
|
||||
"description": "Target branch",
|
||||
"default": "main"
|
||||
}
|
||||
},
|
||||
"required": ["repo_name", "title", "body", "head_branch"]
|
||||
}
|
||||
),
|
||||
types.Tool(
|
||||
name="excel_create_with_formula_and_screenshot",
|
||||
description="Create an XLSX workbook, apply formulas, and render a real screenshot with LibreOffice",
|
||||
inputSchema={
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"output_path": {"type": "string"},
|
||||
"rows": {
|
||||
"type": "array",
|
||||
"items": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"item": {"type": "string"},
|
||||
"quantity": {"type": "number"},
|
||||
"unit_price": {"type": "number"},
|
||||
},
|
||||
"required": ["item", "quantity", "unit_price"],
|
||||
},
|
||||
},
|
||||
},
|
||||
"required": ["output_path", "rows"],
|
||||
}
|
||||
),
|
||||
types.Tool(
|
||||
name="webhook_post",
|
||||
description="POST JSON to a real HTTPS webhook endpoint",
|
||||
inputSchema={"type": "object", "properties": {
|
||||
"url": {"type": "string"}, "payload": {"type": "object"}},
|
||||
"required": ["url", "payload"]}
|
||||
),
|
||||
types.Tool(
|
||||
name="browser_navigate",
|
||||
description="Navigate with real headless Chromium, extract page content, and save a screenshot",
|
||||
inputSchema={"type": "object", "properties": {
|
||||
"url": {"type": "string"}, "screenshot_path": {"type": "string"}},
|
||||
"required": ["url", "screenshot_path"]}
|
||||
),
|
||||
types.Tool(
|
||||
name="virtual_desktop_execute",
|
||||
description="Drive a headful Chromium desktop through X11 keyboard events and retain a screenshot",
|
||||
inputSchema={"type": "object", "properties": {
|
||||
"url": {"type": "string"},
|
||||
"screenshot_path": {"type": "string"},
|
||||
"expected_title": {"type": ["string", "null"]}},
|
||||
"required": ["url", "screenshot_path"]}
|
||||
),
|
||||
types.Tool(
|
||||
name="virtual_mobile_execute",
|
||||
description="Operate a running AndroidWorld emulator through ADB and retain a screenshot",
|
||||
inputSchema={"type": "object", "properties": {
|
||||
"container_name": {"type": "string"},
|
||||
"screenshot_path": {"type": "string"}},
|
||||
"required": ["container_name", "screenshot_path"]}
|
||||
),
|
||||
types.Tool(
|
||||
name="environment_capabilities",
|
||||
description="Inspect real Computer Use container and Android device availability",
|
||||
inputSchema={"type": "object", "properties": {}}
|
||||
)
|
||||
]
|
||||
|
||||
|
||||
@server.call_tool()
|
||||
async def handle_call_tool(
|
||||
name: str,
|
||||
arguments: dict[str, Any] | None
|
||||
) -> list[types.TextContent]:
|
||||
"""Handle tool calls."""
|
||||
if arguments is None:
|
||||
arguments = {}
|
||||
|
||||
try:
|
||||
# Route to appropriate tool
|
||||
if name == "file_write":
|
||||
result = await file_tools.write_file(
|
||||
path=arguments["path"],
|
||||
content=arguments["content"],
|
||||
overwrite=arguments.get("overwrite", False)
|
||||
)
|
||||
elif name == "file_edit":
|
||||
result = await file_tools.edit_file(
|
||||
path=arguments["path"],
|
||||
search=arguments["search"],
|
||||
replace=arguments["replace"]
|
||||
)
|
||||
elif name == "code_interpreter":
|
||||
result = await execution_tools.code_interpreter(
|
||||
code=arguments["code"],
|
||||
language=arguments.get("language") or "python",
|
||||
timeout=arguments.get("timeout", 30.0),
|
||||
stdin=arguments.get("stdin"),
|
||||
files=arguments.get("files")
|
||||
)
|
||||
elif name == "virtual_terminal":
|
||||
result = await execution_tools.virtual_terminal(
|
||||
command=arguments["command"],
|
||||
timeout=arguments.get("timeout", 30)
|
||||
)
|
||||
elif name == "google_calendar_add":
|
||||
result = await external_tools.google_calendar_add(
|
||||
summary=arguments["summary"],
|
||||
start_time=arguments["start_time"],
|
||||
end_time=arguments["end_time"],
|
||||
description=arguments.get("description"),
|
||||
location=arguments.get("location")
|
||||
)
|
||||
elif name == "github_create_pr":
|
||||
result = await external_tools.github_create_pr(
|
||||
repo_name=arguments["repo_name"],
|
||||
title=arguments["title"],
|
||||
body=arguments["body"],
|
||||
head_branch=arguments["head_branch"],
|
||||
base_branch=arguments.get("base_branch", "main")
|
||||
)
|
||||
elif name == "excel_create_with_formula_and_screenshot":
|
||||
result = await extended_tools.excel_create_with_formula_and_screenshot(
|
||||
arguments["output_path"], arguments["rows"])
|
||||
elif name == "webhook_post":
|
||||
result = await extended_tools.webhook_post(arguments["url"], arguments["payload"])
|
||||
elif name == "browser_navigate":
|
||||
result = await extended_tools.browser_navigate(
|
||||
arguments["url"], arguments["screenshot_path"])
|
||||
elif name == "virtual_desktop_execute":
|
||||
result = await extended_tools.virtual_desktop_execute(
|
||||
arguments["url"], arguments["screenshot_path"], arguments.get("expected_title"))
|
||||
elif name == "virtual_mobile_execute":
|
||||
result = await extended_tools.virtual_mobile_execute(
|
||||
arguments["container_name"], arguments["screenshot_path"])
|
||||
elif name == "environment_capabilities":
|
||||
result = await extended_tools.environment_capabilities()
|
||||
else:
|
||||
raise ValueError(f"Unknown tool: {name}")
|
||||
|
||||
# Format result
|
||||
return [
|
||||
types.TextContent(
|
||||
type="text",
|
||||
text=json.dumps(result, indent=2)
|
||||
)
|
||||
]
|
||||
|
||||
except Exception as e:
|
||||
return [
|
||||
types.TextContent(
|
||||
type="text",
|
||||
text=json.dumps({
|
||||
"success": False,
|
||||
"error": f"Tool execution failed: {str(e)}"
|
||||
}, indent=2)
|
||||
)
|
||||
]
|
||||
|
||||
|
||||
async def main():
|
||||
"""Run the MCP server."""
|
||||
async with mcp.server.stdio.stdio_server() as (read_stream, write_stream):
|
||||
await server.run(
|
||||
read_stream,
|
||||
write_stream,
|
||||
InitializationOptions(
|
||||
server_name="execution-tools",
|
||||
server_version="1.0.0",
|
||||
capabilities=server.get_capabilities(
|
||||
notification_options=NotificationOptions(),
|
||||
experimental_capabilities={}
|
||||
)
|
||||
)
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
asyncio.run(main())
|
||||
@@ -0,0 +1,504 @@
|
||||
"""
|
||||
Terminal Controller with integrated file operations.
|
||||
Based on AWorld terminal-controller implementation.
|
||||
Provides command execution with directory navigation and file editing.
|
||||
"""
|
||||
import os
|
||||
import subprocess
|
||||
import traceback
|
||||
from pathlib import Path
|
||||
from typing import Dict, Any
|
||||
|
||||
from config import Config
|
||||
|
||||
|
||||
class TerminalController:
|
||||
"""Terminal controller with directory navigation and file operations."""
|
||||
|
||||
def __init__(self):
|
||||
self.workspace_dir = Path(Config.WORKSPACE_DIR).resolve()
|
||||
self.current_directory = self.workspace_dir
|
||||
self.command_history = []
|
||||
self.max_history = 100
|
||||
|
||||
def _is_safe_path(self, path: Path) -> bool:
|
||||
"""Check if path is within workspace."""
|
||||
try:
|
||||
path.resolve().relative_to(self.workspace_dir)
|
||||
return True
|
||||
except ValueError:
|
||||
return False
|
||||
|
||||
def _resolve_path(self, path: str) -> Path:
|
||||
"""Resolve path relative to current directory."""
|
||||
path_obj = Path(path)
|
||||
if not path_obj.is_absolute():
|
||||
path_obj = self.current_directory / path_obj
|
||||
return path_obj.resolve()
|
||||
|
||||
async def execute_command(
|
||||
self,
|
||||
command: str,
|
||||
timeout: int = 30
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Execute a shell command in current directory.
|
||||
|
||||
Args:
|
||||
command: Command to execute
|
||||
timeout: Timeout in seconds
|
||||
|
||||
Returns:
|
||||
Dictionary with command output
|
||||
"""
|
||||
try:
|
||||
# Add to history
|
||||
self.command_history.append(command)
|
||||
if len(self.command_history) > self.max_history:
|
||||
self.command_history.pop(0)
|
||||
|
||||
# Execute command
|
||||
result = subprocess.run(
|
||||
command,
|
||||
shell=True,
|
||||
capture_output=True,
|
||||
text=True,
|
||||
timeout=timeout,
|
||||
cwd=str(self.current_directory)
|
||||
)
|
||||
|
||||
return {
|
||||
"success": result.returncode == 0,
|
||||
"command": command,
|
||||
"stdout": result.stdout,
|
||||
"stderr": result.stderr,
|
||||
"returncode": result.returncode,
|
||||
"cwd": str(self.current_directory)
|
||||
}
|
||||
|
||||
except subprocess.TimeoutExpired:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Command timed out after {timeout} seconds",
|
||||
"command": command
|
||||
}
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Command execution failed: {str(e)}",
|
||||
"command": command
|
||||
}
|
||||
|
||||
async def get_current_directory(self) -> Dict[str, Any]:
|
||||
"""
|
||||
Get the current working directory.
|
||||
|
||||
Returns:
|
||||
Dictionary with current directory path
|
||||
"""
|
||||
return {
|
||||
"success": True,
|
||||
"current_directory": str(self.current_directory),
|
||||
"workspace": str(self.workspace_dir)
|
||||
}
|
||||
|
||||
async def change_directory(
|
||||
self,
|
||||
directory: str
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Change the current working directory.
|
||||
|
||||
Args:
|
||||
directory: Directory to change to
|
||||
|
||||
Returns:
|
||||
Dictionary with new directory
|
||||
"""
|
||||
try:
|
||||
new_dir = self._resolve_path(directory)
|
||||
|
||||
if not self._is_safe_path(new_dir):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Directory {directory} is outside workspace"
|
||||
}
|
||||
|
||||
if not new_dir.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Directory {directory} does not exist"
|
||||
}
|
||||
|
||||
if not new_dir.is_dir():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"{directory} is not a directory"
|
||||
}
|
||||
|
||||
self.current_directory = new_dir
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"current_directory": str(self.current_directory),
|
||||
"message": f"Changed to {directory}"
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to change directory: {str(e)}"
|
||||
}
|
||||
|
||||
async def list_directory(
|
||||
self,
|
||||
directory: str = "."
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
List contents of a directory.
|
||||
|
||||
Args:
|
||||
directory: Directory to list (relative to current)
|
||||
|
||||
Returns:
|
||||
Dictionary with directory contents
|
||||
"""
|
||||
try:
|
||||
dir_path = self._resolve_path(directory)
|
||||
|
||||
if not self._is_safe_path(dir_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Directory {directory} is outside workspace"
|
||||
}
|
||||
|
||||
if not dir_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Directory {directory} does not exist"
|
||||
}
|
||||
|
||||
contents = []
|
||||
for item in sorted(dir_path.iterdir()):
|
||||
contents.append({
|
||||
"name": item.name,
|
||||
"type": "directory" if item.is_dir() else "file",
|
||||
"size": 0 if item.is_dir() else item.stat().st_size
|
||||
})
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"directory": str(dir_path),
|
||||
"contents": contents,
|
||||
"count": len(contents)
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to list directory: {str(e)}"
|
||||
}
|
||||
|
||||
async def read_file(
|
||||
self,
|
||||
file_path: str,
|
||||
encoding: str = "utf-8"
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Read a file from current directory.
|
||||
|
||||
Args:
|
||||
file_path: File path relative to current directory
|
||||
encoding: File encoding
|
||||
|
||||
Returns:
|
||||
Dictionary with file content
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(file_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {file_path} is outside workspace"
|
||||
}
|
||||
|
||||
if not resolved_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {file_path} does not exist"
|
||||
}
|
||||
|
||||
content = resolved_path.read_text(encoding=encoding)
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"file_path": str(resolved_path),
|
||||
"content": content,
|
||||
"size": len(content),
|
||||
"lines": len(content.splitlines())
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to read file: {str(e)}"
|
||||
}
|
||||
|
||||
async def write_file(
|
||||
self,
|
||||
file_path: str,
|
||||
content: str,
|
||||
mode: str = "w"
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Write content to a file.
|
||||
|
||||
Args:
|
||||
file_path: File path relative to current directory
|
||||
content: Content to write
|
||||
mode: Write mode ('w' for write, 'a' for append)
|
||||
|
||||
Returns:
|
||||
Dictionary with write result
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(file_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {file_path} is outside workspace"
|
||||
}
|
||||
|
||||
# Create parent directories if needed
|
||||
resolved_path.parent.mkdir(parents=True, exist_ok=True)
|
||||
|
||||
# Write file
|
||||
if mode == "a":
|
||||
with open(resolved_path, 'a', encoding='utf-8') as f:
|
||||
f.write(content)
|
||||
else:
|
||||
resolved_path.write_text(content, encoding='utf-8')
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"file_path": str(resolved_path),
|
||||
"bytes_written": len(content),
|
||||
"mode": mode
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to write file: {str(e)}"
|
||||
}
|
||||
|
||||
async def insert_file_content(
|
||||
self,
|
||||
file_path: str,
|
||||
content: str,
|
||||
line_number: int
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Insert content at specific line in file.
|
||||
|
||||
Args:
|
||||
file_path: File path
|
||||
content: Content to insert
|
||||
line_number: Line number to insert at (1-indexed)
|
||||
|
||||
Returns:
|
||||
Dictionary with operation result
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(file_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {file_path} is outside workspace"
|
||||
}
|
||||
|
||||
if not resolved_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {file_path} does not exist"
|
||||
}
|
||||
|
||||
# Read current content
|
||||
lines = resolved_path.read_text(encoding="utf-8").splitlines()
|
||||
|
||||
# Insert content
|
||||
if line_number < 1 or line_number > len(lines) + 1:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Line number {line_number} out of range (1-{len(lines)+1})"
|
||||
}
|
||||
|
||||
lines.insert(line_number - 1, content)
|
||||
|
||||
# Write back
|
||||
resolved_path.write_text('\n'.join(lines) + '\n', encoding="utf-8")
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"file_path": str(resolved_path),
|
||||
"line_number": line_number,
|
||||
"total_lines": len(lines)
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to insert content: {str(e)}"
|
||||
}
|
||||
|
||||
async def delete_file_content(
|
||||
self,
|
||||
file_path: str,
|
||||
start_line: int,
|
||||
end_line: int
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Delete lines from file.
|
||||
|
||||
Args:
|
||||
file_path: File path
|
||||
start_line: Start line number (1-indexed, inclusive)
|
||||
end_line: End line number (1-indexed, inclusive)
|
||||
|
||||
Returns:
|
||||
Dictionary with operation result
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(file_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {file_path} is outside workspace"
|
||||
}
|
||||
|
||||
if not resolved_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {file_path} does not exist"
|
||||
}
|
||||
|
||||
# Read lines
|
||||
lines = resolved_path.read_text(encoding="utf-8").splitlines()
|
||||
|
||||
# Validate range
|
||||
if start_line < 1 or end_line > len(lines) or start_line > end_line:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Invalid line range: {start_line}-{end_line} (file has {len(lines)} lines)"
|
||||
}
|
||||
|
||||
# Delete lines
|
||||
del lines[start_line - 1:end_line]
|
||||
|
||||
# Write back
|
||||
resolved_path.write_text('\n'.join(lines) + '\n' if lines else '', encoding="utf-8")
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"file_path": str(resolved_path),
|
||||
"deleted_lines": end_line - start_line + 1,
|
||||
"remaining_lines": len(lines)
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to delete content: {str(e)}"
|
||||
}
|
||||
|
||||
async def update_file_content(
|
||||
self,
|
||||
file_path: str,
|
||||
line_number: int,
|
||||
new_content: str
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Update a specific line in file.
|
||||
|
||||
Args:
|
||||
file_path: File path
|
||||
line_number: Line number to update (1-indexed)
|
||||
new_content: New content for the line
|
||||
|
||||
Returns:
|
||||
Dictionary with operation result
|
||||
"""
|
||||
try:
|
||||
resolved_path = self._resolve_path(file_path)
|
||||
|
||||
if not self._is_safe_path(resolved_path):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {file_path} is outside workspace"
|
||||
}
|
||||
|
||||
if not resolved_path.exists():
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"File {file_path} does not exist"
|
||||
}
|
||||
|
||||
# Read lines
|
||||
lines = resolved_path.read_text(encoding="utf-8").splitlines()
|
||||
|
||||
if line_number < 1 or line_number > len(lines):
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Line number {line_number} out of range (1-{len(lines)})"
|
||||
}
|
||||
|
||||
# Update line
|
||||
old_content = lines[line_number - 1]
|
||||
lines[line_number - 1] = new_content
|
||||
|
||||
# Write back
|
||||
resolved_path.write_text('\n'.join(lines) + '\n', encoding="utf-8")
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"file_path": str(resolved_path),
|
||||
"line_number": line_number,
|
||||
"old_content": old_content,
|
||||
"new_content": new_content
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
return {
|
||||
"success": False,
|
||||
"error": f"Failed to update content: {str(e)}"
|
||||
}
|
||||
|
||||
async def get_command_history(
|
||||
self,
|
||||
count: int = 10
|
||||
) -> Dict[str, Any]:
|
||||
"""
|
||||
Get recent command history.
|
||||
|
||||
Args:
|
||||
count: Number of recent commands
|
||||
|
||||
Returns:
|
||||
Dictionary with command history
|
||||
"""
|
||||
# count<=0 → []; history[-0:] would return the full list.
|
||||
if count <= 0 or not self.command_history:
|
||||
recent = []
|
||||
else:
|
||||
recent = self.command_history[-count:]
|
||||
|
||||
return {
|
||||
"success": True,
|
||||
"history": recent,
|
||||
"count": len(recent),
|
||||
"total": len(self.command_history)
|
||||
}
|
||||
@@ -0,0 +1,31 @@
|
||||
"""get_command_history(count=0) must return an empty list, not the full history."""
|
||||
import pytest
|
||||
|
||||
from config import Config
|
||||
from terminal_controller import TerminalController
|
||||
|
||||
|
||||
@pytest.fixture
|
||||
def tc(tmp_path, monkeypatch):
|
||||
monkeypatch.setattr(Config, "WORKSPACE_DIR", tmp_path)
|
||||
controller = TerminalController()
|
||||
controller.command_history = ["cmd0", "cmd1", "cmd2", "cmd3", "cmd4"]
|
||||
return controller
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_count_zero_returns_empty_history(tc):
|
||||
result = await tc.get_command_history(count=0)
|
||||
assert result["success"] is True
|
||||
assert result["history"] == []
|
||||
assert result["count"] == 0
|
||||
assert result["total"] == 5
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_positive_count_still_returns_recent(tc):
|
||||
result = await tc.get_command_history(count=2)
|
||||
assert result["success"] is True
|
||||
assert result["history"] == ["cmd3", "cmd4"]
|
||||
assert result["count"] == 2
|
||||
assert result["total"] == 5
|
||||
@@ -0,0 +1,48 @@
|
||||
"""Regression test: malformed numeric env vars must not crash config import.
|
||||
|
||||
TEMPERATURE / MAX_TOKENS / MAX_OUTPUT_LENGTH were parsed with bare
|
||||
float()/int() at import time, so e.g. MAX_TOKENS=abc crashed every tool with
|
||||
ValueError. They now fall back to defaults with a warning.
|
||||
"""
|
||||
import importlib
|
||||
import os
|
||||
from pathlib import Path
|
||||
|
||||
import config as cfg
|
||||
|
||||
|
||||
def test_env_int_falls_back_on_malformed(monkeypatch, capsys):
|
||||
monkeypatch.setenv("MAX_TOKENS", "abc")
|
||||
assert cfg._env_int("MAX_TOKENS", 4096) == 4096
|
||||
assert "invalid MAX_TOKENS" in capsys.readouterr().err
|
||||
|
||||
|
||||
def test_env_int_parses_valid_value(monkeypatch):
|
||||
monkeypatch.setenv("MAX_TOKENS", "123")
|
||||
assert cfg._env_int("MAX_TOKENS", 4096) == 123
|
||||
|
||||
|
||||
def test_env_float_falls_back_on_malformed(monkeypatch, capsys):
|
||||
monkeypatch.setenv("TEMPERATURE", "hot")
|
||||
assert cfg._env_float("TEMPERATURE", 0.7) == 0.7
|
||||
assert "invalid TEMPERATURE" in capsys.readouterr().err
|
||||
|
||||
|
||||
def test_env_float_parses_valid_value(monkeypatch):
|
||||
monkeypatch.setenv("TEMPERATURE", "0.2")
|
||||
assert cfg._env_float("TEMPERATURE", 0.7) == 0.2
|
||||
|
||||
|
||||
def test_module_import_survives_malformed_env(monkeypatch):
|
||||
"""Import-time class attributes must not raise on malformed env values."""
|
||||
monkeypatch.setenv("MAX_OUTPUT_LENGTH", "lots")
|
||||
# Execute a fresh copy under a unique name without replacing the shared
|
||||
# config module that implementation modules imported during collection.
|
||||
spec = importlib.util.spec_from_file_location(
|
||||
"_execution_tools_config_malformed_test",
|
||||
Path(cfg.__file__),
|
||||
)
|
||||
assert spec is not None and spec.loader is not None
|
||||
fresh = importlib.util.module_from_spec(spec)
|
||||
spec.loader.exec_module(fresh)
|
||||
assert fresh.Config.MAX_OUTPUT_LENGTH == 1000
|
||||
@@ -0,0 +1,39 @@
|
||||
"""Regression: edit_file must reject empty search text (not insert at start)."""
|
||||
import tempfile
|
||||
from pathlib import Path
|
||||
|
||||
import pytest
|
||||
|
||||
from llm_helper import LLMHelper
|
||||
from file_tools import FileTools
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_empty_search_rejected():
|
||||
with tempfile.TemporaryDirectory() as td:
|
||||
root = Path(td).resolve()
|
||||
tools = FileTools(LLMHelper())
|
||||
tools.workspace_dir = root
|
||||
target = root / "note.txt"
|
||||
target.write_text("hello world\n", encoding="utf-8")
|
||||
|
||||
result = await tools.edit_file(path="note.txt", search="", replace="INJECT")
|
||||
assert result["success"] is False
|
||||
assert "empty" in result["error"].lower()
|
||||
assert target.read_text(encoding="utf-8") == "hello world\n"
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_normal_edit_still_works():
|
||||
with tempfile.TemporaryDirectory() as td:
|
||||
root = Path(td).resolve()
|
||||
tools = FileTools(LLMHelper())
|
||||
tools.workspace_dir = root
|
||||
target = root / "note.txt"
|
||||
target.write_text("hello world\n", encoding="utf-8")
|
||||
|
||||
result = await tools.edit_file(
|
||||
path="note.txt", search="hello", replace="hi"
|
||||
)
|
||||
assert result["success"] is True
|
||||
assert target.read_text(encoding="utf-8") == "hi world\n"
|
||||
@@ -0,0 +1,96 @@
|
||||
"""Test execution tools."""
|
||||
|
||||
import asyncio
|
||||
from llm_helper import LLMHelper
|
||||
from execution_tools import ExecutionTools
|
||||
|
||||
|
||||
async def test_code_interpreter():
|
||||
"""Test code interpreter functionality."""
|
||||
print("Testing code interpreter...")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
execution_tools = ExecutionTools(llm_helper)
|
||||
|
||||
# Test valid code
|
||||
result = await execution_tools.code_interpreter(
|
||||
code='print("Test successful")\nresult = 2 + 2\nprint(f"2 + 2 = {result}")'
|
||||
)
|
||||
|
||||
assert result["success"], f"Code execution failed: {result.get('error')}"
|
||||
assert "Test successful" in result["stdout"]
|
||||
print(f"✓ Code execution successful: {result}")
|
||||
|
||||
# Test error handling
|
||||
result = await execution_tools.code_interpreter(
|
||||
code='x = 1 / 0'
|
||||
)
|
||||
|
||||
assert not result["success"], "Should fail with division by zero"
|
||||
assert "error_analysis" in result
|
||||
print(f"✓ Error handling works: {result['error'][:100]}...")
|
||||
|
||||
|
||||
async def test_virtual_terminal():
|
||||
"""Test virtual terminal functionality."""
|
||||
print("\nTesting virtual terminal...")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
execution_tools = ExecutionTools(llm_helper)
|
||||
|
||||
# Test successful command
|
||||
result = await execution_tools.virtual_terminal(
|
||||
command='echo "Terminal test"'
|
||||
)
|
||||
|
||||
assert result["success"], f"Command failed: {result.get('error')}"
|
||||
assert "Terminal test" in result["stdout"]
|
||||
print(f"✓ Command execution successful: {result}")
|
||||
|
||||
# Test failed command
|
||||
result = await execution_tools.virtual_terminal(
|
||||
command='ls /nonexistent_directory_12345'
|
||||
)
|
||||
|
||||
assert not result["success"], "Should fail with non-existent directory"
|
||||
assert "error_analysis" in result
|
||||
print(f"✓ Error handling works: returncode={result['returncode']}")
|
||||
|
||||
|
||||
async def test_syntax_verification():
|
||||
"""Test syntax verification."""
|
||||
print("\nTesting syntax verification...")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
execution_tools = ExecutionTools(llm_helper)
|
||||
|
||||
# Test syntax error detection
|
||||
result = await execution_tools.code_interpreter(
|
||||
code='print("Unclosed string'
|
||||
)
|
||||
|
||||
assert not result["success"], "Should detect syntax error"
|
||||
print(f"✓ Syntax verification works: {result['error'][:100]}...")
|
||||
|
||||
|
||||
async def main():
|
||||
"""Run all tests."""
|
||||
print("=== Execution Tools Tests ===\n")
|
||||
|
||||
try:
|
||||
await test_code_interpreter()
|
||||
await test_virtual_terminal()
|
||||
await test_syntax_verification()
|
||||
|
||||
print("\n✓ All execution tools tests passed!")
|
||||
|
||||
except AssertionError as e:
|
||||
print(f"\n✗ Test failed: {e}")
|
||||
except Exception as e:
|
||||
print(f"\n✗ Error: {e}")
|
||||
import traceback
|
||||
traceback.print_exc()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
asyncio.run(main())
|
||||
@@ -0,0 +1,107 @@
|
||||
"""Test external integration tools."""
|
||||
|
||||
import asyncio
|
||||
from llm_helper import LLMHelper
|
||||
from external_tools import ExternalTools
|
||||
|
||||
|
||||
async def test_google_calendar():
|
||||
"""Test Google Calendar integration."""
|
||||
print("Testing Google Calendar...")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
external_tools = ExternalTools(llm_helper)
|
||||
|
||||
try:
|
||||
result = await external_tools.google_calendar_add(
|
||||
summary="Test Event",
|
||||
start_time="2025-10-01T10:00:00",
|
||||
end_time="2025-10-01T11:00:00",
|
||||
description="This is a test event"
|
||||
)
|
||||
|
||||
if result["success"]:
|
||||
print(f"✓ Calendar event created: {result}")
|
||||
else:
|
||||
print(f"Calendar test skipped or failed: {result['error']}")
|
||||
|
||||
except Exception as e:
|
||||
print(f"Calendar test skipped (likely missing credentials): {e}")
|
||||
|
||||
|
||||
async def test_github_pr():
|
||||
"""Test GitHub PR creation."""
|
||||
print("\nTesting GitHub PR...")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
external_tools = ExternalTools(llm_helper)
|
||||
|
||||
try:
|
||||
# Note: This will fail without a valid repo and token
|
||||
result = await external_tools.github_create_pr(
|
||||
repo_name="test/test-repo",
|
||||
title="Test PR",
|
||||
body="This is a test PR",
|
||||
head_branch="test-branch",
|
||||
base_branch="main"
|
||||
)
|
||||
|
||||
if result["success"]:
|
||||
print(f"✓ PR created: {result}")
|
||||
else:
|
||||
print(f"PR test expected to fail (test repo): {result['error']}")
|
||||
|
||||
except Exception as e:
|
||||
print(f"PR test skipped (likely missing credentials): {e}")
|
||||
|
||||
|
||||
async def test_datetime_parsing():
|
||||
"""Test datetime parsing."""
|
||||
print("\nTesting datetime parsing...")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
external_tools = ExternalTools(llm_helper)
|
||||
|
||||
# Test invalid datetime
|
||||
result = await external_tools.google_calendar_add(
|
||||
summary="Test",
|
||||
start_time="invalid-datetime",
|
||||
end_time="2025-10-01T11:00:00"
|
||||
)
|
||||
|
||||
assert not result["success"], "Should fail with invalid datetime"
|
||||
print(f"✓ Invalid datetime handling works: {result['error']}")
|
||||
|
||||
# Test end before start
|
||||
result = await external_tools.google_calendar_add(
|
||||
summary="Test",
|
||||
start_time="2025-10-01T11:00:00",
|
||||
end_time="2025-10-01T10:00:00"
|
||||
)
|
||||
|
||||
assert not result["success"], "Should fail when end is before start"
|
||||
print(f"✓ Time validation works: {result['error']}")
|
||||
|
||||
|
||||
async def main():
|
||||
"""Run all tests."""
|
||||
print("=== External Tools Tests ===\n")
|
||||
|
||||
try:
|
||||
await test_datetime_parsing()
|
||||
await test_google_calendar()
|
||||
await test_github_pr()
|
||||
|
||||
print("\n✓ External tools tests completed!")
|
||||
print("Note: Some tests may be skipped if credentials are not configured.")
|
||||
|
||||
except AssertionError as e:
|
||||
print(f"\n✗ Test failed: {e}")
|
||||
except Exception as e:
|
||||
print(f"\n✗ Error: {e}")
|
||||
import traceback
|
||||
traceback.print_exc()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
asyncio.run(main())
|
||||
@@ -0,0 +1,106 @@
|
||||
"""Test file system tools."""
|
||||
|
||||
import asyncio
|
||||
import tempfile
|
||||
from pathlib import Path
|
||||
from llm_helper import LLMHelper
|
||||
from file_tools import FileTools
|
||||
from config import Config
|
||||
|
||||
|
||||
async def test_file_write():
|
||||
"""Test file write functionality."""
|
||||
print("Testing file write...")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
file_tools = FileTools(llm_helper)
|
||||
|
||||
# Test writing valid Python code
|
||||
result = await file_tools.write_file(
|
||||
path="test_output.py",
|
||||
content='print("Hello, World!")\n',
|
||||
overwrite=True
|
||||
)
|
||||
|
||||
assert result["success"], f"File write failed: {result.get('error')}"
|
||||
assert result["verification"] in ["passed", "skipped"]
|
||||
print(f"✓ File write successful: {result}")
|
||||
|
||||
# Test syntax error detection
|
||||
result = await file_tools.write_file(
|
||||
path="test_syntax_error.py",
|
||||
content='print("Unclosed string\n',
|
||||
overwrite=True
|
||||
)
|
||||
|
||||
if Config.AUTO_VERIFY_CODE:
|
||||
assert not result["success"], "Should detect syntax error"
|
||||
print(f"✓ Syntax error detected: {result['error']}")
|
||||
else:
|
||||
print("✓ Verification skipped (AUTO_VERIFY_CODE=False)")
|
||||
|
||||
|
||||
async def test_file_edit():
|
||||
"""Test file edit functionality."""
|
||||
print("\nTesting file edit...")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
file_tools = FileTools(llm_helper)
|
||||
|
||||
# Create a test file first
|
||||
await file_tools.write_file(
|
||||
path="test_edit.py",
|
||||
content='message = "Hello"\nprint(message)\n',
|
||||
overwrite=True
|
||||
)
|
||||
|
||||
# Edit the file
|
||||
result = await file_tools.edit_file(
|
||||
path="test_edit.py",
|
||||
search='message = "Hello"',
|
||||
replace='message = "Hi there"'
|
||||
)
|
||||
|
||||
assert result["success"], f"File edit failed: {result.get('error')}"
|
||||
assert "diff_preview" in result
|
||||
print(f"✓ File edit successful: {result}")
|
||||
|
||||
|
||||
async def test_safety_checks():
|
||||
"""Test safety checks for file operations."""
|
||||
print("\nTesting safety checks...")
|
||||
|
||||
llm_helper = LLMHelper()
|
||||
file_tools = FileTools(llm_helper)
|
||||
|
||||
# Test path outside workspace
|
||||
result = await file_tools.write_file(
|
||||
path="/tmp/outside_workspace.txt",
|
||||
content="test"
|
||||
)
|
||||
|
||||
# This should fail unless /tmp is in workspace
|
||||
print(f"Path safety check result: {result}")
|
||||
|
||||
|
||||
async def main():
|
||||
"""Run all tests."""
|
||||
print("=== File Tools Tests ===\n")
|
||||
|
||||
try:
|
||||
await test_file_write()
|
||||
await test_file_edit()
|
||||
await test_safety_checks()
|
||||
|
||||
print("\n✓ All file tools tests passed!")
|
||||
|
||||
except AssertionError as e:
|
||||
print(f"\n✗ Test failed: {e}")
|
||||
except Exception as e:
|
||||
print(f"\n✗ Error: {e}")
|
||||
import traceback
|
||||
traceback.print_exc()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
asyncio.run(main())
|
||||
@@ -0,0 +1,372 @@
|
||||
"""
|
||||
Real tests for enhanced filesystem tools.
|
||||
These tests perform actual file operations to verify functionality.
|
||||
"""
|
||||
import asyncio
|
||||
import json
|
||||
import pytest
|
||||
from pathlib import Path
|
||||
import sys
|
||||
|
||||
# Add current directory to path
|
||||
sys.path.insert(0, str(Path(__file__).parent))
|
||||
|
||||
from config import Config
|
||||
from filesystem_enhanced import FilesystemEnhanced
|
||||
|
||||
|
||||
@pytest.fixture(scope="function")
|
||||
def fs(tmp_path, monkeypatch):
|
||||
"""Create filesystem instance with temp workspace."""
|
||||
monkeypatch.setattr(Config, "WORKSPACE_DIR", tmp_path)
|
||||
return FilesystemEnhanced()
|
||||
|
||||
|
||||
@pytest.fixture(scope="function")
|
||||
def test_files(fs):
|
||||
"""Create test files for testing."""
|
||||
# Create some test files
|
||||
try:
|
||||
(Config.WORKSPACE_DIR / "test1.txt").write_text("Hello World")
|
||||
(Config.WORKSPACE_DIR / "test2.txt").write_text("Python Testing\nLine 2\nLine 3")
|
||||
(Config.WORKSPACE_DIR / "data.json").write_text('{"key": "value"}')
|
||||
(Config.WORKSPACE_DIR / "subdir").mkdir()
|
||||
(Config.WORKSPACE_DIR / "subdir" / "nested.txt").write_text("Nested file")
|
||||
except Exception as e:
|
||||
print(f"Error creating test files: {e}")
|
||||
return fs
|
||||
|
||||
|
||||
class TestReadOperations:
|
||||
"""Tests for file reading operations."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_read_text_file(self, test_files):
|
||||
"""Test reading a text file."""
|
||||
result = await test_files.read_text_file("test1.txt")
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["content"] == "Hello World"
|
||||
assert result["file_size"] > 0
|
||||
assert result["lines"] == 1
|
||||
|
||||
print("✅ Read text file successfully")
|
||||
print(f" Content: {result['content']}")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_read_multiline_file(self, test_files):
|
||||
"""Test reading multiline file."""
|
||||
result = await test_files.read_text_file("test2.txt")
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["lines"] == 3
|
||||
assert "Python Testing" in result["content"]
|
||||
|
||||
print("✅ Read multiline file")
|
||||
print(f" Lines: {result['lines']}")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_read_nonexistent_file(self, fs):
|
||||
"""Test reading non-existent file."""
|
||||
result = await fs.read_text_file("nonexistent.txt")
|
||||
|
||||
assert result["success"] is False
|
||||
assert "does not exist" in result["error"]
|
||||
|
||||
print("✅ Correctly handled nonexistent file")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_read_multiple_files(self, test_files):
|
||||
"""Test reading multiple files at once."""
|
||||
result = await test_files.read_multiple_files([
|
||||
"test1.txt",
|
||||
"test2.txt",
|
||||
"data.json"
|
||||
])
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["files_read"] == 3
|
||||
assert result["files_failed"] == 0
|
||||
assert "test1.txt" in result["results"]
|
||||
assert "test2.txt" in result["results"]
|
||||
|
||||
print(f"✅ Read {result['files_read']} files")
|
||||
print(f" Files: {', '.join(result['results'].keys())}")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_read_multiple_files_with_errors(self, test_files):
|
||||
"""Test reading multiple files with some missing."""
|
||||
result = await test_files.read_multiple_files([
|
||||
"test1.txt",
|
||||
"nonexistent.txt",
|
||||
"test2.txt"
|
||||
])
|
||||
|
||||
assert result["success"] is True # At least some succeeded
|
||||
assert result["files_read"] == 2
|
||||
assert result["files_failed"] == 1
|
||||
assert len(result["errors"]) == 1
|
||||
|
||||
print(f"✅ Read {result['files_read']} files, {result['files_failed']} failed")
|
||||
|
||||
|
||||
class TestListOperations:
|
||||
"""Tests for directory listing operations."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_list_directory_with_sizes(self, test_files):
|
||||
"""Test listing directory with file sizes."""
|
||||
result = await test_files.list_directory_with_sizes(".")
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["total_items"] >= 4 # At least 3 files + 1 dir
|
||||
assert result["total_size"] > 0
|
||||
assert len(result["contents"]) >= 4
|
||||
|
||||
# Check structure
|
||||
item = result["contents"][0]
|
||||
assert "name" in item
|
||||
assert "type" in item
|
||||
assert "size" in item
|
||||
assert "size_human" in item
|
||||
|
||||
print(f"✅ Listed {result['total_items']} items")
|
||||
print(f" Total size: {result['total_size_human']}")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_directory_tree(self, test_files):
|
||||
"""Test generating directory tree."""
|
||||
result = await test_files.directory_tree(".", max_depth=3)
|
||||
|
||||
assert result["success"] is True
|
||||
assert "tree" in result
|
||||
assert result["tree"]["type"] == "directory"
|
||||
assert "children" in result["tree"]
|
||||
assert len(result["tree"]["children"]) >= 4
|
||||
|
||||
print(f"✅ Generated directory tree")
|
||||
print(f" Root: {result['root']}")
|
||||
print(f" Items: {result['tree']['count']}")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_directory_tree_depth_limit(self, test_files):
|
||||
"""Test directory tree with depth limit."""
|
||||
# Create deeper structure
|
||||
(Config.WORKSPACE_DIR / "deep" / "level2" / "level3").mkdir(parents=True)
|
||||
|
||||
result = await test_files.directory_tree(".", max_depth=2)
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["max_depth"] == 2
|
||||
|
||||
print("✅ Directory tree respects depth limit")
|
||||
|
||||
|
||||
class TestSearchOperations:
|
||||
"""Tests for file search operations."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_search_files_pattern(self, test_files):
|
||||
"""Test searching files by pattern."""
|
||||
result = await test_files.search_files("*.txt", ".", recursive=True)
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["matches"] >= 3 # test1.txt, test2.txt, nested.txt
|
||||
assert result["pattern"] == "*.txt"
|
||||
|
||||
# Check results structure
|
||||
if len(result["files"]) > 0:
|
||||
file_info = result["files"][0]
|
||||
assert "path" in file_info
|
||||
assert "size" in file_info
|
||||
assert "size_human" in file_info
|
||||
|
||||
print(f"✅ Found {result['matches']} files matching *.txt")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_search_files_nonrecursive(self, test_files):
|
||||
"""Test non-recursive search."""
|
||||
result = await test_files.search_files("*.txt", ".", recursive=False)
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["recursive"] is False
|
||||
# Should not find nested.txt
|
||||
paths = [f["path"] for f in result["files"]]
|
||||
assert not any("subdir" in p for p in paths)
|
||||
|
||||
print(f"✅ Non-recursive search: {result['matches']} files")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_search_files_json(self, test_files):
|
||||
"""Test searching for specific file type."""
|
||||
result = await test_files.search_files("*.json", ".", recursive=True)
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["matches"] >= 1
|
||||
|
||||
print(f"✅ Found {result['matches']} JSON files")
|
||||
|
||||
|
||||
class TestFileInfo:
|
||||
"""Tests for file information retrieval."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_get_file_info(self, test_files):
|
||||
"""Test getting file information."""
|
||||
result = await test_files.get_file_info("test2.txt")
|
||||
|
||||
assert result["success"] is True
|
||||
info = result["file_info"]
|
||||
|
||||
assert info["name"] == "test2.txt"
|
||||
assert info["extension"] == ".txt"
|
||||
assert info["is_file"] is True
|
||||
assert info["is_directory"] is False
|
||||
assert info["lines"] == 3
|
||||
assert info["size"] > 0
|
||||
|
||||
print("✅ File info retrieved")
|
||||
print(f" Name: {info['name']}")
|
||||
print(f" Size: {info['size_human']}")
|
||||
print(f" Lines: {info['lines']}")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_get_directory_info(self, test_files):
|
||||
"""Test getting directory information."""
|
||||
result = await test_files.get_file_info("subdir")
|
||||
|
||||
assert result["success"] is True
|
||||
info = result["file_info"]
|
||||
|
||||
assert info["is_directory"] is True
|
||||
assert info["is_file"] is False
|
||||
|
||||
print("✅ Directory info retrieved")
|
||||
|
||||
|
||||
class TestMoveOperations:
|
||||
"""Tests for move and copy operations."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_move_file(self, test_files):
|
||||
"""Test moving a file."""
|
||||
result = await test_files.move_file("test1.txt", "moved.txt")
|
||||
|
||||
assert result["success"] is True
|
||||
assert Path(Config.WORKSPACE_DIR / "moved.txt").exists()
|
||||
assert not Path(Config.WORKSPACE_DIR / "test1.txt").exists()
|
||||
|
||||
print("✅ File moved successfully")
|
||||
print(f" From: test1.txt")
|
||||
print(f" To: moved.txt")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_move_file_no_overwrite(self, test_files):
|
||||
"""Test move without overwrite."""
|
||||
result = await test_files.move_file("test1.txt", "test2.txt", overwrite=False)
|
||||
|
||||
assert result["success"] is False
|
||||
assert "already exists" in result["error"]
|
||||
|
||||
print("✅ Correctly prevented overwrite")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_copy_file(self, test_files):
|
||||
"""Test copying a file."""
|
||||
result = await test_files.copy_file("test1.txt", "copied.txt")
|
||||
|
||||
assert result["success"] is True
|
||||
assert Path(Config.WORKSPACE_DIR / "copied.txt").exists()
|
||||
assert Path(Config.WORKSPACE_DIR / "test1.txt").exists() # Original still exists
|
||||
|
||||
print("✅ File copied successfully")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_copy_directory(self, test_files):
|
||||
"""Test copying a directory."""
|
||||
result = await test_files.copy_file("subdir", "subdir_copy")
|
||||
|
||||
assert result["success"] is True
|
||||
assert Path(Config.WORKSPACE_DIR / "subdir_copy").exists()
|
||||
assert Path(Config.WORKSPACE_DIR / "subdir_copy" / "nested.txt").exists()
|
||||
|
||||
print("✅ Directory copied recursively")
|
||||
|
||||
|
||||
class TestDeleteOperations:
|
||||
"""Tests for delete operations."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_delete_file(self, test_files):
|
||||
"""Test deleting a file."""
|
||||
result = await test_files.delete_file("test1.txt")
|
||||
|
||||
assert result["success"] is True
|
||||
assert not Path(Config.WORKSPACE_DIR / "test1.txt").exists()
|
||||
|
||||
print("✅ File deleted")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_delete_directory_recursive(self, test_files):
|
||||
"""Test deleting directory recursively."""
|
||||
result = await test_files.delete_file("subdir", recursive=True)
|
||||
|
||||
assert result["success"] is True
|
||||
assert not Path(Config.WORKSPACE_DIR / "subdir").exists()
|
||||
|
||||
print("✅ Directory deleted recursively")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_delete_directory_without_recursive(self, test_files):
|
||||
"""Test that directory delete requires recursive flag."""
|
||||
result = await test_files.delete_file("subdir", recursive=False)
|
||||
|
||||
assert result["success"] is False
|
||||
assert "recursive" in result["error"].lower()
|
||||
|
||||
print("✅ Correctly required recursive flag")
|
||||
|
||||
|
||||
class TestCreateOperations:
|
||||
"""Tests for create operations."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_create_directory(self, fs):
|
||||
"""Test creating a directory."""
|
||||
result = await fs.create_directory("newdir")
|
||||
|
||||
assert result["success"] is True
|
||||
assert Path(Config.WORKSPACE_DIR / "newdir").exists()
|
||||
|
||||
print("✅ Directory created")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_create_nested_directory(self, fs):
|
||||
"""Test creating nested directories."""
|
||||
result = await fs.create_directory("parent/child/grandchild", parents=True)
|
||||
|
||||
assert result["success"] is True
|
||||
assert Path(Config.WORKSPACE_DIR / "parent" / "child" / "grandchild").exists()
|
||||
|
||||
print("✅ Nested directories created")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_list_allowed_directories(self, fs):
|
||||
"""Test listing allowed directories."""
|
||||
result = await fs.list_allowed_directories()
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["count"] >= 1
|
||||
assert len(result["allowed_directories"]) >= 1
|
||||
|
||||
print(f"✅ Listed {result['count']} allowed directories")
|
||||
|
||||
|
||||
# Run tests
|
||||
if __name__ == "__main__":
|
||||
print("=" * 70)
|
||||
print("Running Enhanced Filesystem Tools Tests")
|
||||
print("=" * 70)
|
||||
print()
|
||||
|
||||
pytest.main([__file__, "-v", "-s"])
|
||||
@@ -0,0 +1,62 @@
|
||||
"""Idempotency coverage for the Experiment 4-3 GitHub execution tool."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import asyncio
|
||||
from datetime import datetime, timezone
|
||||
|
||||
from external_tools import ExternalTools
|
||||
|
||||
|
||||
class _Approval:
|
||||
def request_approval(self, _operation, _details):
|
||||
return True, "bounded test"
|
||||
|
||||
|
||||
class _Ref:
|
||||
def __init__(self, ref: str):
|
||||
self.ref = ref
|
||||
|
||||
|
||||
class _Pull:
|
||||
number = 605
|
||||
html_url = "https://github.com/bojieli/ai-agent-book/pull/605"
|
||||
title = "Experiment 4-3"
|
||||
state = "open"
|
||||
created_at = datetime(2026, 8, 2, tzinfo=timezone.utc)
|
||||
head = _Ref("exp/4-3-gui-environments")
|
||||
base = _Ref("main")
|
||||
|
||||
|
||||
class _Repo:
|
||||
def get_branch(self, name):
|
||||
return _Ref(name)
|
||||
|
||||
def get_pulls(self, *, state, head, base):
|
||||
assert (state, head, base) == (
|
||||
"open", "bojieli:exp/4-3-gui-environments", "main")
|
||||
return [_Pull()]
|
||||
|
||||
def create_pull(self, **_kwargs):
|
||||
raise AssertionError("an existing PR must be reused, not duplicated")
|
||||
|
||||
|
||||
class _GitHub:
|
||||
def get_repo(self, name):
|
||||
assert name == "bojieli/ai-agent-book"
|
||||
return _Repo()
|
||||
|
||||
|
||||
def test_existing_open_pull_request_is_reused() -> None:
|
||||
tool = ExternalTools(_Approval())
|
||||
tool._github_client = _GitHub()
|
||||
result = asyncio.run(tool.github_create_pr(
|
||||
repo_name="bojieli/ai-agent-book",
|
||||
title="Experiment 4-3",
|
||||
body="bounded test",
|
||||
head_branch="exp/4-3-gui-environments",
|
||||
base_branch="main",
|
||||
))
|
||||
assert result["success"] is True
|
||||
assert result["pr_number"] == 605
|
||||
assert result["idempotent_reuse"] is True
|
||||
@@ -0,0 +1,19 @@
|
||||
"""Focused validation for Experiment 4-3 desktop/mobile execution tools."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import asyncio
|
||||
|
||||
from extended_tools import ExtendedTools
|
||||
|
||||
|
||||
def test_virtual_desktop_rejects_non_https() -> None:
|
||||
result = asyncio.run(ExtendedTools().virtual_desktop_execute(
|
||||
"http://example.com", "unused.png"))
|
||||
assert result == {"success": False, "error": "Only HTTPS URLs are allowed"}
|
||||
|
||||
|
||||
def test_virtual_mobile_rejects_unsafe_container_name() -> None:
|
||||
result = asyncio.run(ExtendedTools().virtual_mobile_execute(
|
||||
"container; touch escaped", "unused.png"))
|
||||
assert result == {"success": False, "error": "Invalid Docker container name"}
|
||||
@@ -0,0 +1,171 @@
|
||||
"""Test multi-language code execution."""
|
||||
|
||||
import asyncio
|
||||
import sys
|
||||
from multilang_executor import LanguageExecutor, ExecutionStatus
|
||||
|
||||
|
||||
async def run_language_case(executor: LanguageExecutor, language: str, code: str, description: str):
|
||||
"""Test a specific language."""
|
||||
print(f"\n{'='*60}")
|
||||
print(f"Testing {language}: {description}")
|
||||
print(f"{'='*60}")
|
||||
|
||||
result = await executor.execute_code(code, language, timeout=10.0)
|
||||
|
||||
print(f"Status: {result.get('status')}")
|
||||
if result.get('stdout'):
|
||||
print(f"Output:\n{result['stdout']}")
|
||||
if result.get('stderr'):
|
||||
print(f"Errors:\n{result['stderr']}")
|
||||
if result.get('compile_output'):
|
||||
print(f"Compile output:\n{result['compile_output']}")
|
||||
|
||||
success = result.get('status') == ExecutionStatus.SUCCESS
|
||||
print(f"✅ PASSED" if success else f"❌ FAILED")
|
||||
return success
|
||||
|
||||
|
||||
async def main():
|
||||
"""Run all language tests."""
|
||||
executor = LanguageExecutor()
|
||||
|
||||
tests = [
|
||||
# Python
|
||||
("python", """
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
data = pd.DataFrame({'A': [1, 2, 3], 'B': [4, 5, 6]})
|
||||
print("Data shape:", data.shape)
|
||||
print("Mean of A:", data['A'].mean())
|
||||
""", "NumPy and Pandas"),
|
||||
|
||||
# JavaScript
|
||||
("javascript", """
|
||||
console.log('Hello from Node.js!');
|
||||
const numbers = [1, 2, 3, 4, 5];
|
||||
const sum = numbers.reduce((a, b) => a + b, 0);
|
||||
console.log('Sum:', sum);
|
||||
""", "Array operations"),
|
||||
|
||||
# TypeScript
|
||||
("typescript", """
|
||||
interface Point {
|
||||
x: number;
|
||||
y: number;
|
||||
}
|
||||
|
||||
const point: Point = { x: 10, y: 20 };
|
||||
console.log(`Point: (${point.x}, ${point.y})`);
|
||||
""", "Type-safe interfaces"),
|
||||
|
||||
# Go
|
||||
("go", """
|
||||
package main
|
||||
|
||||
import "fmt"
|
||||
|
||||
func main() {
|
||||
fmt.Println("Hello from Go!")
|
||||
sum := 0
|
||||
for i := 1; i <= 10; i++ {
|
||||
sum += i
|
||||
}
|
||||
fmt.Printf("Sum of 1-10: %d\\n", sum)
|
||||
}
|
||||
""", "Loops and formatting"),
|
||||
|
||||
# Java
|
||||
("java", """
|
||||
public class Main {
|
||||
public static void main(String[] args) {
|
||||
System.out.println("Hello from Java!");
|
||||
int sum = 0;
|
||||
for (int i = 1; i <= 10; i++) {
|
||||
sum += i;
|
||||
}
|
||||
System.out.println("Sum of 1-10: " + sum);
|
||||
}
|
||||
}
|
||||
""", "Basic class and loops"),
|
||||
|
||||
# C++
|
||||
("cpp", """
|
||||
#include <iostream>
|
||||
#include <vector>
|
||||
#include <numeric>
|
||||
|
||||
int main() {
|
||||
std::cout << "Hello from C++!" << std::endl;
|
||||
std::vector<int> numbers = {1, 2, 3, 4, 5};
|
||||
int sum = std::accumulate(numbers.begin(), numbers.end(), 0);
|
||||
std::cout << "Sum: " << sum << std::endl;
|
||||
return 0;
|
||||
}
|
||||
""", "STL vector and accumulate"),
|
||||
|
||||
# Rust
|
||||
("rust", """
|
||||
fn main() {
|
||||
println!("Hello from Rust!");
|
||||
let numbers = vec![1, 2, 3, 4, 5];
|
||||
let sum: i32 = numbers.iter().sum();
|
||||
println!("Sum: {}", sum);
|
||||
}
|
||||
""", "Vector and iterators"),
|
||||
|
||||
# PHP
|
||||
("php", """
|
||||
<?php
|
||||
echo "Hello from PHP!\\n";
|
||||
$numbers = [1, 2, 3, 4, 5];
|
||||
$sum = array_sum($numbers);
|
||||
echo "Sum: $sum\\n";
|
||||
?>
|
||||
""", "Array operations"),
|
||||
|
||||
# Bash
|
||||
("bash", """
|
||||
echo "Hello from Bash!"
|
||||
sum=0
|
||||
for i in {1..10}; do
|
||||
sum=$((sum + i))
|
||||
done
|
||||
echo "Sum of 1-10: $sum"
|
||||
""", "Shell loops"),
|
||||
]
|
||||
|
||||
print(f"\n{'#'*60}")
|
||||
print(f"# Multi-Language Code Execution Test Suite")
|
||||
print(f"{'#'*60}")
|
||||
|
||||
results = []
|
||||
for language, code, description in tests:
|
||||
try:
|
||||
success = await run_language_case(executor, language, code, description)
|
||||
results.append((language, success))
|
||||
except Exception as e:
|
||||
print(f"❌ EXCEPTION: {e}")
|
||||
results.append((language, False))
|
||||
|
||||
# Summary
|
||||
print(f"\n{'='*60}")
|
||||
print(f"Test Summary")
|
||||
print(f"{'='*60}")
|
||||
|
||||
passed = sum(1 for _, success in results if success)
|
||||
total = len(results)
|
||||
|
||||
for language, success in results:
|
||||
status = "✅ PASS" if success else "❌ FAIL"
|
||||
print(f"{status} - {language}")
|
||||
|
||||
print(f"\nTotal: {passed}/{total} passed ({100*passed//total}%)")
|
||||
|
||||
return passed == total
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
success = asyncio.run(main())
|
||||
sys.exit(0 if success else 1)
|
||||
@@ -0,0 +1,29 @@
|
||||
import asyncio
|
||||
import pytest
|
||||
from multilang_executor import get_all_output, LanguageExecutor, ExecutionStatus
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_get_all_output_reads_full_payload():
|
||||
class FakeStream:
|
||||
def __init__(self, data: bytes):
|
||||
self._data = data
|
||||
self._done = False
|
||||
|
||||
async def read(self, n: int = -1):
|
||||
if self._done:
|
||||
return b""
|
||||
self._done = True
|
||||
return self._data
|
||||
|
||||
payload = b"hello world" * 1000
|
||||
out = await get_all_output(FakeStream(payload))
|
||||
assert out == payload.decode()
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_execute_captures_stdout():
|
||||
exe = LanguageExecutor()
|
||||
result = await exe.execute_code("print('ok-from-executor')", "python", timeout=10)
|
||||
assert result["status"] == ExecutionStatus.SUCCESS
|
||||
assert "ok-from-executor" in result["stdout"]
|
||||
@@ -0,0 +1,45 @@
|
||||
"""Null optional language must default to python on public paths."""
|
||||
import asyncio
|
||||
import tempfile
|
||||
from pathlib import Path
|
||||
from unittest.mock import MagicMock
|
||||
|
||||
from multilang_executor import LanguageExecutor
|
||||
from execution_tools import ExecutionTools
|
||||
|
||||
|
||||
def test_null_language_executor_defaults_to_python():
|
||||
le = LanguageExecutor(workspace_dir=Path(tempfile.mkdtemp()))
|
||||
result = asyncio.run(le.execute_code("print(42)", language=None, timeout=10))
|
||||
assert isinstance(result, dict)
|
||||
assert result.get("language") == "python"
|
||||
assert "42" in (result.get("stdout") or "")
|
||||
|
||||
|
||||
def test_null_language_code_interpreter_defaults(monkeypatch):
|
||||
"""Public MCP path: ExecutionTools.code_interpreter(language=None)."""
|
||||
helper = MagicMock()
|
||||
et = ExecutionTools(helper)
|
||||
seen = {}
|
||||
|
||||
async def fake_exec(code, language, timeout=30.0, stdin=None, files=None):
|
||||
seen["language"] = language
|
||||
return {
|
||||
"success": True,
|
||||
"stdout": "ok\n",
|
||||
"stderr": "",
|
||||
"language": language,
|
||||
"returncode": 0,
|
||||
"status": "ok",
|
||||
}
|
||||
|
||||
monkeypatch.setattr(et.lang_executor, "execute_code", fake_exec)
|
||||
import config as cfg
|
||||
monkeypatch.setattr(cfg.Config, "AUTO_VERIFY_CODE", False, raising=False)
|
||||
monkeypatch.setattr(cfg.Config, "REQUIRE_APPROVAL_FOR_DANGEROUS_OPS", False, raising=False)
|
||||
monkeypatch.setattr(cfg.Config, "AUTO_SUMMARIZE_COMPLEX_OUTPUT", False, raising=False)
|
||||
|
||||
out = asyncio.run(et.code_interpreter("print(1)", language=None))
|
||||
assert seen["language"] == "python"
|
||||
assert out["language"] == "python"
|
||||
assert out.get("error") in (None, "")
|
||||
@@ -0,0 +1,5 @@
|
||||
def greet(name):
|
||||
print(f"Hello, {name}!")
|
||||
|
||||
if __name__ == "__main__":
|
||||
greet("MCP Server")
|
||||
@@ -0,0 +1,211 @@
|
||||
"""
|
||||
Tests for Terminal Controller.
|
||||
Tests command execution with directory navigation and file operations.
|
||||
"""
|
||||
import asyncio
|
||||
import json
|
||||
import pytest
|
||||
|
||||
from config import Config
|
||||
from terminal_controller import TerminalController
|
||||
|
||||
|
||||
@pytest.fixture(scope="function")
|
||||
def tc(tmp_path, monkeypatch):
|
||||
"""Create terminal controller with temp workspace."""
|
||||
monkeypatch.setattr(Config, "WORKSPACE_DIR", tmp_path)
|
||||
return TerminalController()
|
||||
|
||||
|
||||
class TestTerminalBasics:
|
||||
"""Tests for basic terminal operations."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_get_current_directory(self, tc):
|
||||
"""Test getting current directory."""
|
||||
result = await tc.get_current_directory()
|
||||
|
||||
assert result["success"] is True
|
||||
assert "current_directory" in result
|
||||
assert "workspace" in result
|
||||
|
||||
print(f"✅ Current directory: {result['current_directory']}")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_execute_command_simple(self, tc):
|
||||
"""Test executing a simple command."""
|
||||
result = await tc.execute_command("echo 'Hello World'")
|
||||
|
||||
assert result["success"] is True
|
||||
assert "Hello World" in result["stdout"]
|
||||
assert result["returncode"] == 0
|
||||
|
||||
print(f"✅ Command executed: {result['command']}")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_execute_command_ls(self, tc):
|
||||
"""Test listing directory."""
|
||||
result = await tc.execute_command("ls")
|
||||
|
||||
assert result["success"] is True
|
||||
print(f"✅ Directory listing completed")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_command_history(self, tc):
|
||||
"""Test command history."""
|
||||
await tc.execute_command("echo 'test1'")
|
||||
await tc.execute_command("echo 'test2'")
|
||||
await tc.execute_command("echo 'test3'")
|
||||
|
||||
result = await tc.get_command_history(count=2)
|
||||
|
||||
assert result["success"] is True
|
||||
assert len(result["history"]) == 2
|
||||
assert result["total"] == 3
|
||||
|
||||
print(f"✅ Command history: {result['count']} recent commands")
|
||||
|
||||
|
||||
class TestDirectoryOperations:
|
||||
"""Tests for directory navigation."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_change_directory(self, tc):
|
||||
"""Test changing directory."""
|
||||
# Create a subdirectory
|
||||
subdir = Config.WORKSPACE_DIR / "subdir"
|
||||
subdir.mkdir()
|
||||
|
||||
result = await tc.change_directory("subdir")
|
||||
|
||||
assert result["success"] is True
|
||||
assert "subdir" in result["current_directory"]
|
||||
|
||||
print(f"✅ Changed to: {result['current_directory']}")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_list_directory(self, tc):
|
||||
"""Test listing directory."""
|
||||
# Create some files
|
||||
(Config.WORKSPACE_DIR / "file1.txt").write_text("test")
|
||||
(Config.WORKSPACE_DIR / "file2.txt").write_text("test")
|
||||
|
||||
result = await tc.list_directory(".")
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["count"] >= 2
|
||||
|
||||
print(f"✅ Listed {result['count']} items")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_change_to_nonexistent(self, tc):
|
||||
"""Test changing to nonexistent directory."""
|
||||
result = await tc.change_directory("nonexistent")
|
||||
|
||||
assert result["success"] is False
|
||||
assert "does not exist" in result["error"]
|
||||
|
||||
print("✅ Correctly rejected nonexistent directory")
|
||||
|
||||
|
||||
class TestFileOperations:
|
||||
"""Tests for file operations through terminal controller."""
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_write_file(self, tc):
|
||||
"""Test writing a file."""
|
||||
result = await tc.write_file("test.txt", "Hello Terminal")
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["bytes_written"] > 0
|
||||
|
||||
# Verify file exists
|
||||
file_path = Config.WORKSPACE_DIR / "test.txt"
|
||||
assert file_path.exists()
|
||||
assert file_path.read_text() == "Hello Terminal"
|
||||
|
||||
print(f"✅ Wrote {result['bytes_written']} bytes")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_read_file(self, tc):
|
||||
"""Test reading a file."""
|
||||
# Create a file
|
||||
test_file = Config.WORKSPACE_DIR / "read_test.txt"
|
||||
test_file.write_text("Test content\nLine 2")
|
||||
|
||||
result = await tc.read_file("read_test.txt")
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["content"] == "Test content\nLine 2"
|
||||
assert result["lines"] == 2
|
||||
|
||||
print(f"✅ Read file: {result['lines']} lines")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_insert_file_content(self, tc):
|
||||
"""Test inserting content into file."""
|
||||
# Create a file
|
||||
test_file = Config.WORKSPACE_DIR / "insert_test.txt"
|
||||
test_file.write_text("Line 1\nLine 3")
|
||||
|
||||
result = await tc.insert_file_content("insert_test.txt", "Line 2", 2)
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["line_number"] == 2
|
||||
|
||||
# Verify content
|
||||
content = test_file.read_text()
|
||||
lines = content.splitlines()
|
||||
assert lines[1] == "Line 2"
|
||||
|
||||
print("✅ Inserted content at line 2")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_delete_file_content(self, tc):
|
||||
"""Test deleting lines from file."""
|
||||
# Create a file
|
||||
test_file = Config.WORKSPACE_DIR / "delete_test.txt"
|
||||
test_file.write_text("Line 1\nLine 2\nLine 3\nLine 4")
|
||||
|
||||
result = await tc.delete_file_content("delete_test.txt", 2, 3)
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["deleted_lines"] == 2
|
||||
|
||||
# Verify content
|
||||
content = test_file.read_text()
|
||||
assert "Line 2" not in content
|
||||
assert "Line 3" not in content
|
||||
assert "Line 1" in content
|
||||
assert "Line 4" in content
|
||||
|
||||
print(f"✅ Deleted {result['deleted_lines']} lines")
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_update_file_content(self, tc):
|
||||
"""Test updating a line in file."""
|
||||
# Create a file
|
||||
test_file = Config.WORKSPACE_DIR / "update_test.txt"
|
||||
test_file.write_text("Line 1\nOld Line 2\nLine 3")
|
||||
|
||||
result = await tc.update_file_content("update_test.txt", 2, "New Line 2")
|
||||
|
||||
assert result["success"] is True
|
||||
assert result["old_content"] == "Old Line 2"
|
||||
assert result["new_content"] == "New Line 2"
|
||||
|
||||
# Verify content
|
||||
content = test_file.read_text()
|
||||
assert "New Line 2" in content
|
||||
assert "Old Line 2" not in content
|
||||
|
||||
print("✅ Updated line 2")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
print("=" * 70)
|
||||
print("Running Terminal Controller Tests")
|
||||
print("=" * 70)
|
||||
print()
|
||||
|
||||
pytest.main([__file__, "-v", "-s"])
|
||||
@@ -0,0 +1,36 @@
|
||||
"""tail_lines=0 must keep no tail lines (Python lines[-0:] quirk)."""
|
||||
from execution_tools import truncate_and_persist
|
||||
|
||||
|
||||
def test_tail_lines_zero_keeps_only_head():
|
||||
text = "\n".join(f"line{i}" for i in range(100))
|
||||
out, path = truncate_and_persist(
|
||||
text, head_lines=3, tail_lines=0, max_lines=10, max_chars=50
|
||||
)
|
||||
assert path is not None
|
||||
assert "line0" in out and "line1" in out and "line2" in out
|
||||
assert "line99" not in out
|
||||
assert "line50" not in out
|
||||
assert out.count("line0") == 1
|
||||
|
||||
|
||||
def test_tail_lines_positive_still_keeps_tail():
|
||||
text = "\n".join(f"line{i}" for i in range(100))
|
||||
out, path = truncate_and_persist(
|
||||
text, head_lines=2, tail_lines=2, max_lines=10, max_chars=50
|
||||
)
|
||||
assert path is not None
|
||||
assert "line0" in out and "line1" in out
|
||||
assert "line98" in out and "line99" in out
|
||||
assert "line50" not in out
|
||||
|
||||
|
||||
def test_short_file_char_overflow_does_not_duplicate_lines():
|
||||
text = ("x" * 4000 + "\n") * 3
|
||||
out, path = truncate_and_persist(
|
||||
text, head_lines=50, tail_lines=50, max_lines=200, max_chars=100
|
||||
)
|
||||
assert path is not None
|
||||
# Three content lines plus the guide line — no head/tail duplication.
|
||||
content_lines = [ln for ln in out.split("\n") if ln.startswith("x")]
|
||||
assert len(content_lines) == 3
|
||||
@@ -0,0 +1,8 @@
|
||||
{
|
||||
"experiment": "4-3",
|
||||
"campaign_id": "real_mcp_gui_20260802T093657Z",
|
||||
"status": "blocked",
|
||||
"official_complete": false,
|
||||
"manifest": "validation/experiment_4_3/real_mcp_gui_20260802T093657Z/manifest.json",
|
||||
"manifest_sha256": "fde8976b91b149a61b7d468f4c825c1bdfdc9da3062cbfa66aaa1fd0f3d1966f"
|
||||
}
|
||||
+272
@@ -0,0 +1,272 @@
|
||||
{
|
||||
"transport": "mcp-stdio",
|
||||
"server_name": "execution-tools",
|
||||
"server_version": "1.0.0",
|
||||
"schemas": [
|
||||
{
|
||||
"name": "file_write",
|
||||
"description": "Write content to a file with automatic syntax verification",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"path": {
|
||||
"type": "string",
|
||||
"description": "File path (relative to workspace or absolute)"
|
||||
},
|
||||
"content": {
|
||||
"type": "string",
|
||||
"description": "Content to write"
|
||||
},
|
||||
"overwrite": {
|
||||
"type": "boolean",
|
||||
"description": "Whether to overwrite existing files",
|
||||
"default": false
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"path",
|
||||
"content"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "file_edit",
|
||||
"description": "Edit an existing file by searching and replacing content",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"path": {
|
||||
"type": "string",
|
||||
"description": "File path"
|
||||
},
|
||||
"search": {
|
||||
"type": "string",
|
||||
"description": "Text to search for"
|
||||
},
|
||||
"replace": {
|
||||
"type": "string",
|
||||
"description": "Replacement text"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"path",
|
||||
"search",
|
||||
"replace"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "code_interpreter",
|
||||
"description": "Execute code in multiple programming languages in a sandboxed environment with result analysis. Supports: Python, JavaScript, TypeScript, Go, Java, C++, Rust, PHP, Bash",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"code": {
|
||||
"type": "string",
|
||||
"description": "Code to execute"
|
||||
},
|
||||
"language": {
|
||||
"type": "string",
|
||||
"description": "Programming language (python, javascript, typescript, go, java, cpp, rust, php, bash)",
|
||||
"default": "python"
|
||||
},
|
||||
"timeout": {
|
||||
"type": "number",
|
||||
"description": "Execution timeout in seconds",
|
||||
"default": 30.0
|
||||
},
|
||||
"stdin": {
|
||||
"type": "string",
|
||||
"description": "Optional stdin input for the program"
|
||||
},
|
||||
"files": {
|
||||
"type": "object",
|
||||
"description": "Optional additional files (filename -> content mapping)",
|
||||
"additionalProperties": {
|
||||
"type": "string"
|
||||
}
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"code"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "virtual_terminal",
|
||||
"description": "Execute shell commands with error summarization",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"command": {
|
||||
"type": "string",
|
||||
"description": "Shell command to execute"
|
||||
},
|
||||
"timeout": {
|
||||
"type": "integer",
|
||||
"description": "Timeout in seconds",
|
||||
"default": 30
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"command"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "google_calendar_add",
|
||||
"description": "Add an event to Google Calendar",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"summary": {
|
||||
"type": "string",
|
||||
"description": "Event title"
|
||||
},
|
||||
"start_time": {
|
||||
"type": "string",
|
||||
"description": "Start time (ISO 8601 format, e.g., 2024-01-01T10:00:00)"
|
||||
},
|
||||
"end_time": {
|
||||
"type": "string",
|
||||
"description": "End time (ISO 8601 format)"
|
||||
},
|
||||
"description": {
|
||||
"type": "string",
|
||||
"description": "Event description"
|
||||
},
|
||||
"location": {
|
||||
"type": "string",
|
||||
"description": "Event location"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"summary",
|
||||
"start_time",
|
||||
"end_time"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "github_create_pr",
|
||||
"description": "Create a GitHub Pull Request",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"repo_name": {
|
||||
"type": "string",
|
||||
"description": "Repository name (format: owner/repo)"
|
||||
},
|
||||
"title": {
|
||||
"type": "string",
|
||||
"description": "PR title"
|
||||
},
|
||||
"body": {
|
||||
"type": "string",
|
||||
"description": "PR description"
|
||||
},
|
||||
"head_branch": {
|
||||
"type": "string",
|
||||
"description": "Source branch"
|
||||
},
|
||||
"base_branch": {
|
||||
"type": "string",
|
||||
"description": "Target branch",
|
||||
"default": "main"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"repo_name",
|
||||
"title",
|
||||
"body",
|
||||
"head_branch"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "excel_create_with_formula_and_screenshot",
|
||||
"description": "Create an XLSX workbook, apply formulas, and render a real screenshot with LibreOffice",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"output_path": {
|
||||
"type": "string"
|
||||
},
|
||||
"rows": {
|
||||
"type": "array",
|
||||
"items": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"item": {
|
||||
"type": "string"
|
||||
},
|
||||
"quantity": {
|
||||
"type": "number"
|
||||
},
|
||||
"unit_price": {
|
||||
"type": "number"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"item",
|
||||
"quantity",
|
||||
"unit_price"
|
||||
]
|
||||
}
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"output_path",
|
||||
"rows"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "webhook_post",
|
||||
"description": "POST JSON to a real HTTPS webhook endpoint",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"url": {
|
||||
"type": "string"
|
||||
},
|
||||
"payload": {
|
||||
"type": "object"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"url",
|
||||
"payload"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "browser_navigate",
|
||||
"description": "Navigate with real headless Chromium, extract page content, and save a screenshot",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"url": {
|
||||
"type": "string"
|
||||
},
|
||||
"screenshot_path": {
|
||||
"type": "string"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"url",
|
||||
"screenshot_path"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "environment_capabilities",
|
||||
"description": "Inspect real Computer Use container and Android device availability",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {}
|
||||
}
|
||||
}
|
||||
],
|
||||
"schema_sha256": "d9a5bd49b95a42874d0d588926f4c717df2344c86f22991624c3056720649e74"
|
||||
}
|
||||
+66
@@ -0,0 +1,66 @@
|
||||
[
|
||||
{
|
||||
"purpose": "dangerous_operation_review",
|
||||
"called_at_utc": "2026-07-29T22:20:05.549577+00:00",
|
||||
"provider": "kimi",
|
||||
"request": {
|
||||
"model": "kimi-k3",
|
||||
"messages": [
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are a cautious safety reviewer. Approve operations that are safe and reject risky ones."
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": "You are a safety reviewer for an AI agent execution system.\nReview the following operation and determine if it should be approved.\n\nOperation: terminal_command\nDetails: {\n \"command\": \"rm -rf ./should-never-execute\",\n \"detected_patterns\": [\n \"rm -rf\"\n ]\n}\n\nAnalyze the operation for:\n1. Potential data loss or destructive actions\n2. Security risks\n3. Resource consumption concerns\n4. Compliance with best practices\n\nRespond in JSON format:\n{\n \"approved\": true/false,\n \"reason\": \"Brief explanation of your decision\",\n \"risk_level\": \"low/medium/high\",\n \"recommendations\": [\"List of recommendations if any\"]\n}\n"
|
||||
}
|
||||
],
|
||||
"temperature": 1,
|
||||
"max_tokens": 4096
|
||||
},
|
||||
"response": {
|
||||
"id": "chatcmpl-6a6a7c7da8472cac5b5dc12c",
|
||||
"model": "kimi-k3",
|
||||
"finish_reason": "stop",
|
||||
"content": "```json\n{\n \"approved\": false,\n \"reason\": \"The command uses 'rm -rf', which performs irreversible recursive deletion without confirmation prompts. The target directory name 'should-never-execute' strongly signals this operation must not run. Additionally, the relative path means the deletion scope depends on the unknown current working directory, creating risk of unintended data loss.\",\n \"risk_level\": \"high\",\n \"recommendations\": [\n \"Reject this operation outright; the flagged 'rm -rf' pattern and directory naming indicate it should never execute\",\n \"If deletion of this path is ever legitimately required, first verify the working directory (pwd) and inspect contents (ls -la) before proceeding\",\n \"Replace destructive deletion with safer alternatives: move to a trash/quarantine directory, or use interactive deletion (rm -ri)\",\n \"Require explicit human confirmation for any operation matching recursive force-deletion patterns\",\n \"Implement policy guardrails to automatically block 'rm -rf' commands unless explicitly whitelisted with an absolute, validated path\",\n \"Consider creating a backup or archive of the target before any destructive action\"\n ]\n}\n```"
|
||||
},
|
||||
"usage": {
|
||||
"prompt_tokens": 259,
|
||||
"completion_tokens": 782,
|
||||
"total_tokens": 1041
|
||||
},
|
||||
"latency_seconds": 25.121
|
||||
},
|
||||
{
|
||||
"purpose": "dangerous_operation_review",
|
||||
"called_at_utc": "2026-07-29T22:20:22.117029+00:00",
|
||||
"provider": "kimi",
|
||||
"request": {
|
||||
"model": "kimi-k3",
|
||||
"messages": [
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are a cautious safety reviewer. Approve operations that are safe and reject risky ones."
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": "You are a safety reviewer for an AI agent execution system.\nReview the following operation and determine if it should be approved.\n\nOperation: code_execution\nDetails: {\n \"code\": \"import urllib.request\\ntry:\\n print(urllib.request.urlopen('https://example.com', timeout=3).status)\\nexcept Exception as e:\\n print(type(e).__name__, str(e))\\n\",\n \"language\": \"python\",\n \"detected_patterns\": [\n \"open(\"\n ]\n}\n\nAnalyze the operation for:\n1. Potential data loss or destructive actions\n2. Security risks\n3. Resource consumption concerns\n4. Compliance with best practices\n\nRespond in JSON format:\n{\n \"approved\": true/false,\n \"reason\": \"Brief explanation of your decision\",\n \"risk_level\": \"low/medium/high\",\n \"recommendations\": [\"List of recommendations if any\"]\n}\n"
|
||||
}
|
||||
],
|
||||
"temperature": 1,
|
||||
"max_tokens": 4096
|
||||
},
|
||||
"response": {
|
||||
"id": "chatcmpl-6a6a7c9605fd318b0d416518",
|
||||
"model": "kimi-k3",
|
||||
"finish_reason": "stop",
|
||||
"content": "```json\n{\n \"approved\": true,\n \"reason\": \"The code performs a single, read-only HTTPS GET request to example.com (an IANA-reserved documentation domain) with a 3-second timeout and proper exception handling. It only prints the HTTP status code, sends no data, writes no files, and performs no destructive actions. The detected 'open(' pattern is a false positive matching 'urlopen(', not file access.\",\n \"risk_level\": \"low\",\n \"recommendations\": [\n \"Confirm that outbound network access is permitted in this execution environment's policy\",\n \"Consider restricting network egress to an allowlist of trusted domains as a general safeguard\",\n \"The code follows good practices (timeout, try/except, HTTPS); no changes required\"\n ]\n}\n```"
|
||||
},
|
||||
"usage": {
|
||||
"prompt_tokens": 298,
|
||||
"completion_tokens": 544,
|
||||
"total_tokens": 842
|
||||
},
|
||||
"latency_seconds": 16.405
|
||||
}
|
||||
]
|
||||
+66
@@ -0,0 +1,66 @@
|
||||
[
|
||||
{
|
||||
"purpose": "dangerous_operation_review",
|
||||
"called_at_utc": "2026-07-29T22:20:05.549577+00:00",
|
||||
"provider": "kimi",
|
||||
"request": {
|
||||
"model": "kimi-k3",
|
||||
"messages": [
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are a cautious safety reviewer. Approve operations that are safe and reject risky ones."
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": "You are a safety reviewer for an AI agent execution system.\nReview the following operation and determine if it should be approved.\n\nOperation: terminal_command\nDetails: {\n \"command\": \"rm -rf ./should-never-execute\",\n \"detected_patterns\": [\n \"rm -rf\"\n ]\n}\n\nAnalyze the operation for:\n1. Potential data loss or destructive actions\n2. Security risks\n3. Resource consumption concerns\n4. Compliance with best practices\n\nRespond in JSON format:\n{\n \"approved\": true/false,\n \"reason\": \"Brief explanation of your decision\",\n \"risk_level\": \"low/medium/high\",\n \"recommendations\": [\"List of recommendations if any\"]\n}\n"
|
||||
}
|
||||
],
|
||||
"temperature": 1,
|
||||
"max_tokens": 4096
|
||||
},
|
||||
"response": {
|
||||
"id": "chatcmpl-6a6a7c7da8472cac5b5dc12c",
|
||||
"model": "kimi-k3",
|
||||
"finish_reason": "stop",
|
||||
"content": "```json\n{\n \"approved\": false,\n \"reason\": \"The command uses 'rm -rf', which performs irreversible recursive deletion without confirmation prompts. The target directory name 'should-never-execute' strongly signals this operation must not run. Additionally, the relative path means the deletion scope depends on the unknown current working directory, creating risk of unintended data loss.\",\n \"risk_level\": \"high\",\n \"recommendations\": [\n \"Reject this operation outright; the flagged 'rm -rf' pattern and directory naming indicate it should never execute\",\n \"If deletion of this path is ever legitimately required, first verify the working directory (pwd) and inspect contents (ls -la) before proceeding\",\n \"Replace destructive deletion with safer alternatives: move to a trash/quarantine directory, or use interactive deletion (rm -ri)\",\n \"Require explicit human confirmation for any operation matching recursive force-deletion patterns\",\n \"Implement policy guardrails to automatically block 'rm -rf' commands unless explicitly whitelisted with an absolute, validated path\",\n \"Consider creating a backup or archive of the target before any destructive action\"\n ]\n}\n```"
|
||||
},
|
||||
"usage": {
|
||||
"prompt_tokens": 259,
|
||||
"completion_tokens": 782,
|
||||
"total_tokens": 1041
|
||||
},
|
||||
"latency_seconds": 25.121
|
||||
},
|
||||
{
|
||||
"purpose": "dangerous_operation_review",
|
||||
"called_at_utc": "2026-07-29T22:20:22.117029+00:00",
|
||||
"provider": "kimi",
|
||||
"request": {
|
||||
"model": "kimi-k3",
|
||||
"messages": [
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are a cautious safety reviewer. Approve operations that are safe and reject risky ones."
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": "You are a safety reviewer for an AI agent execution system.\nReview the following operation and determine if it should be approved.\n\nOperation: code_execution\nDetails: {\n \"code\": \"import urllib.request\\ntry:\\n print(urllib.request.urlopen('https://example.com', timeout=3).status)\\nexcept Exception as e:\\n print(type(e).__name__, str(e))\\n\",\n \"language\": \"python\",\n \"detected_patterns\": [\n \"open(\"\n ]\n}\n\nAnalyze the operation for:\n1. Potential data loss or destructive actions\n2. Security risks\n3. Resource consumption concerns\n4. Compliance with best practices\n\nRespond in JSON format:\n{\n \"approved\": true/false,\n \"reason\": \"Brief explanation of your decision\",\n \"risk_level\": \"low/medium/high\",\n \"recommendations\": [\"List of recommendations if any\"]\n}\n"
|
||||
}
|
||||
],
|
||||
"temperature": 1,
|
||||
"max_tokens": 4096
|
||||
},
|
||||
"response": {
|
||||
"id": "chatcmpl-6a6a7c9605fd318b0d416518",
|
||||
"model": "kimi-k3",
|
||||
"finish_reason": "stop",
|
||||
"content": "```json\n{\n \"approved\": true,\n \"reason\": \"The code performs a single, read-only HTTPS GET request to example.com (an IANA-reserved documentation domain) with a 3-second timeout and proper exception handling. It only prints the HTTP status code, sends no data, writes no files, and performs no destructive actions. The detected 'open(' pattern is a false positive matching 'urlopen(', not file access.\",\n \"risk_level\": \"low\",\n \"recommendations\": [\n \"Confirm that outbound network access is permitted in this execution environment's policy\",\n \"Consider restricting network egress to an allowlist of trusted domains as a general safeguard\",\n \"The code follows good practices (timeout, try/except, HTTPS); no changes required\"\n ]\n}\n```"
|
||||
},
|
||||
"usage": {
|
||||
"prompt_tokens": 298,
|
||||
"completion_tokens": 544,
|
||||
"total_tokens": 842
|
||||
},
|
||||
"latency_seconds": 16.405
|
||||
}
|
||||
]
|
||||
+158
@@ -0,0 +1,158 @@
|
||||
{
|
||||
"experiment": "4-3",
|
||||
"campaign_id": "real_mcp_20260730T062500Z",
|
||||
"status": "blocked",
|
||||
"official_complete": false,
|
||||
"files": [
|
||||
{
|
||||
"path": "catalog.json",
|
||||
"bytes": 7135,
|
||||
"sha256": "a2459670d25b4cce95b6709846934d6600e1c2eac6090cd77ff6139369ad9c58"
|
||||
},
|
||||
{
|
||||
"path": "llm_receipts.checkpoint.json",
|
||||
"bytes": 5129,
|
||||
"sha256": "c2fc88cedae848981afe8bbb66673dfe8b2dfd7f5fa043fc818661c1030d8fd6"
|
||||
},
|
||||
{
|
||||
"path": "llm_receipts.json",
|
||||
"bytes": 5130,
|
||||
"sha256": "4a004b17ad738a6fc025def17e21752b1e3469ab028ecf755df1c20b0d4bb7ce"
|
||||
},
|
||||
{
|
||||
"path": "outside-witness.txt",
|
||||
"bytes": 16,
|
||||
"sha256": "de6e8ea7f35c8a0261f7fb9eb75022a92311cfd84248567104f5c9e7d3ddf782"
|
||||
},
|
||||
{
|
||||
"path": "protocol.json",
|
||||
"bytes": 1067,
|
||||
"sha256": "f8ca33de720405502a7f0df26a2c9a4bf3988a6eee235500a1355d258ebaebc0"
|
||||
},
|
||||
{
|
||||
"path": "receipts/01_python_valid_write.json",
|
||||
"bytes": 504,
|
||||
"sha256": "37bc5a682d57a9eb27fec152853cfe66e7c749632e58e14dc73e35156a8ca28e"
|
||||
},
|
||||
{
|
||||
"path": "receipts/02_python_invalid_rejected.json",
|
||||
"bytes": 417,
|
||||
"sha256": "b45cf262489b95361bb0b447fe2f8c8c13cba7a85ec667093d26ae0d683eaae7"
|
||||
},
|
||||
{
|
||||
"path": "receipts/03_javascript_valid_write.json",
|
||||
"bytes": 516,
|
||||
"sha256": "c03a9605a65497993e72260f6a9e633bbe8074eb4c479291753d9522411d8b2e"
|
||||
},
|
||||
{
|
||||
"path": "receipts/04_javascript_invalid_rejected.json",
|
||||
"bytes": 897,
|
||||
"sha256": "610a32ec012ec46669b0969b0fc301193fbe5998a068bf7b5dd50b5fc6fbaa2b"
|
||||
},
|
||||
{
|
||||
"path": "receipts/05_verified_edit.json",
|
||||
"bytes": 519,
|
||||
"sha256": "c48383e59f6bce36e6578a19f1a4be26aacd1de7c1de91e525cf354221d33d43"
|
||||
},
|
||||
{
|
||||
"path": "receipts/06_path_escape_rejected.json",
|
||||
"bytes": 369,
|
||||
"sha256": "8f778b3d855f06b85c6d81f23d4e905dc46c438a3669d841324e1578e094b678"
|
||||
},
|
||||
{
|
||||
"path": "receipts/07_terminal_safe.json",
|
||||
"bytes": 494,
|
||||
"sha256": "4637b9a43250b8f65806d9655f970ed0ca362192050ee41ae932e265d1889948"
|
||||
},
|
||||
{
|
||||
"path": "receipts/08_terminal_timeout.json",
|
||||
"bytes": 307,
|
||||
"sha256": "692ae82a1a80008ddabede71d795ec367f95a8ab30ed808f2cb67c72930fae60"
|
||||
},
|
||||
{
|
||||
"path": "receipts/09_terminal_danger_rejected.json",
|
||||
"bytes": 682,
|
||||
"sha256": "dea8813966af4346cd97a7ca55aeb2c9f2d3f73ae5cd91ff65ab4882aa93b84d"
|
||||
},
|
||||
{
|
||||
"path": "receipts/10_python_docker_sandbox.json",
|
||||
"bytes": 1126,
|
||||
"sha256": "6bc08e992ef73a34eb14850e51eb489ce4d7940dca68607ca258a37c1958c0f1"
|
||||
},
|
||||
{
|
||||
"path": "receipts/11_python_network_denied.json",
|
||||
"bytes": 1011,
|
||||
"sha256": "e38c6a1d6423deda44e72057bdf4568e8fa4bfa1b6c5bf44b9654d31e82e09b7"
|
||||
},
|
||||
{
|
||||
"path": "receipts/12_long_output_persisted.json",
|
||||
"bytes": 2176,
|
||||
"sha256": "3a84fc3cf588a45df0d70b2557565ea73d90483bcd7d7bf69f81b12007f79328"
|
||||
},
|
||||
{
|
||||
"path": "receipts/13_excel_formula_screenshot.json",
|
||||
"bytes": 1469,
|
||||
"sha256": "5369cebd287e80969af39f17286bea44cedd16690999eb9f7aa4faa7d7f33384"
|
||||
},
|
||||
{
|
||||
"path": "receipts/14_real_webhook.json",
|
||||
"bytes": 1150,
|
||||
"sha256": "e6a1343693070eeabff7ea6c3c03f527f4ddb52437cea1a4cd8f22179fee9e5f"
|
||||
},
|
||||
{
|
||||
"path": "receipts/15_real_browser.json",
|
||||
"bytes": 884,
|
||||
"sha256": "000f0f374fb5f3738706c7c06d17bce4184c83259b9f92f94a6b5fc250f01abc"
|
||||
},
|
||||
{
|
||||
"path": "receipts/16_calendar_preflight.json",
|
||||
"bytes": 442,
|
||||
"sha256": "67b317de35158904b5b08e891d4a8e31bbe25a94e4d1cfe88ffeebb3d2433556"
|
||||
},
|
||||
{
|
||||
"path": "receipts/17_github_pr_preflight.json",
|
||||
"bytes": 462,
|
||||
"sha256": "a39311ace49b1e2e4db4c614f59f7e7374b83cad38df38a22be0260a041d8e82"
|
||||
},
|
||||
{
|
||||
"path": "receipts/18_desktop_mobile_capabilities.json",
|
||||
"bytes": 489,
|
||||
"sha256": "6b20317fda1651294c714f9f163721e848d637f79a927d48f97d6cf392818579"
|
||||
},
|
||||
{
|
||||
"path": "summary.json",
|
||||
"bytes": 1279,
|
||||
"sha256": "7580ee0e25bcef7e931adc36146bf9590334823e75f38c9a4ec11b57bdeb2699"
|
||||
},
|
||||
{
|
||||
"path": "workspace/browser-example.png",
|
||||
"bytes": 16578,
|
||||
"sha256": "f21d7a2b1f7739641b8838e2ed2a9a907559cece9397568db6e1ccca197cc7b0"
|
||||
},
|
||||
{
|
||||
"path": "workspace/invoice.pdf",
|
||||
"bytes": 19487,
|
||||
"sha256": "a398997cace6b14ee4cd8575124a3be4fd0f1c125958c2e99cbd6ae9f91dbbf5"
|
||||
},
|
||||
{
|
||||
"path": "workspace/invoice.png",
|
||||
"bytes": 14315,
|
||||
"sha256": "b97c76df59cc7972772341ea3a391bb4dba8907b65000f1b360005804d39986d"
|
||||
},
|
||||
{
|
||||
"path": "workspace/invoice.xlsx",
|
||||
"bytes": 5059,
|
||||
"sha256": "621c89e193243ab849d1d20df53c9deaaabe2e8909712edd87d6c137f5551470"
|
||||
},
|
||||
{
|
||||
"path": "workspace/valid.js",
|
||||
"bytes": 40,
|
||||
"sha256": "799574240050acc326491093ba5641e3eb6ec281fc8f7a0a259b80a0a51724cd"
|
||||
},
|
||||
{
|
||||
"path": "workspace/valid.py",
|
||||
"bytes": 32,
|
||||
"sha256": "e1a894022d1a082987b87adecb623438c9e386d86b2b621cff4a5fe7fdf7edc8"
|
||||
}
|
||||
]
|
||||
}
|
||||
+1
@@ -0,0 +1 @@
|
||||
MUST-NOT-CHANGE
|
||||
+45
@@ -0,0 +1,45 @@
|
||||
{
|
||||
"experiment": "4-3",
|
||||
"authority": "book/chapter4.md:274",
|
||||
"required_categories": {
|
||||
"file_write_edit": [
|
||||
"python_linter",
|
||||
"javascript_linter",
|
||||
"structured_errors"
|
||||
],
|
||||
"terminal": [
|
||||
"timeout",
|
||||
"dangerous_command_review",
|
||||
"history_or_receipt"
|
||||
],
|
||||
"code_interpreter": [
|
||||
"real_sandbox",
|
||||
"dangerous_operation_gate",
|
||||
"long_output_persisted"
|
||||
],
|
||||
"data": [
|
||||
"excel_write",
|
||||
"formula",
|
||||
"screenshot"
|
||||
],
|
||||
"external": [
|
||||
"calendar",
|
||||
"github_pr",
|
||||
"email",
|
||||
"webhook"
|
||||
],
|
||||
"gui": [
|
||||
"browser",
|
||||
"virtual_desktop",
|
||||
"virtual_mobile"
|
||||
]
|
||||
},
|
||||
"safety": {
|
||||
"workspace_confinement": true,
|
||||
"automatic_linter": true,
|
||||
"llm_driven_danger_review": true,
|
||||
"long_output_head_tail_and_full_file": true,
|
||||
"credential_free_receipts": true
|
||||
},
|
||||
"completion_rule": "Every named manuscript category must have substantive real execution evidence; missing credentials or active GUI backends produce blocked, never passed."
|
||||
}
|
||||
+18
@@ -0,0 +1,18 @@
|
||||
{
|
||||
"case": "python_valid_write",
|
||||
"tool": "file_write",
|
||||
"arguments": {
|
||||
"path": "valid.py",
|
||||
"content": "def add(a, b):\n return a + b\n",
|
||||
"overwrite": true
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T062500Z/workspace/valid.py",
|
||||
"bytes_written": 32,
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 0.002
|
||||
}
|
||||
+17
@@ -0,0 +1,17 @@
|
||||
{
|
||||
"case": "python_invalid_rejected",
|
||||
"tool": "file_write",
|
||||
"arguments": {
|
||||
"path": "invalid.py",
|
||||
"content": "def broken(:\n pass\n",
|
||||
"overwrite": true
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Syntax validation failed: Syntax error at line 1: invalid syntax",
|
||||
"verification": "failed"
|
||||
},
|
||||
"latency_seconds": 0.001
|
||||
}
|
||||
+18
@@ -0,0 +1,18 @@
|
||||
{
|
||||
"case": "javascript_valid_write",
|
||||
"tool": "file_write",
|
||||
"arguments": {
|
||||
"path": "valid.js",
|
||||
"content": "const answer = 42;\nconsole.log(answer);\n",
|
||||
"overwrite": true
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T062500Z/workspace/valid.js",
|
||||
"bytes_written": 40,
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 0.066
|
||||
}
|
||||
+17
@@ -0,0 +1,17 @@
|
||||
{
|
||||
"case": "javascript_invalid_rejected",
|
||||
"tool": "file_write",
|
||||
"arguments": {
|
||||
"path": "invalid.js",
|
||||
"content": "const broken = ;\n",
|
||||
"overwrite": true
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Syntax validation failed: [stdin]:1\nconst broken = ;\n ^\n\nSyntaxError: Unexpected token ';'\n at wrapSafe (node:internal/modules/cjs/loader:1740:18)\n at checkSyntax (node:internal/main/check_syntax:76:3)\n at node:internal/main/check_syntax:45:5\n at Socket.<anonymous> (node:internal/process/execution:205:5)\n at Socket.emit (node:events:520:22)\n at endReadableNT (node:internal/streams/readable:1729:12)\n at process.processTicksAndRejections (node:internal/process/task_queues:90:21)\n\nNode.js v25.6.0",
|
||||
"verification": "failed"
|
||||
},
|
||||
"latency_seconds": 0.066
|
||||
}
|
||||
+18
@@ -0,0 +1,18 @@
|
||||
{
|
||||
"case": "verified_edit",
|
||||
"tool": "file_edit",
|
||||
"arguments": {
|
||||
"path": "valid.py",
|
||||
"search": "a + b",
|
||||
"replace": "a - b"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T062500Z/workspace/valid.py",
|
||||
"diff_preview": "Line 2:\n - return a + b\n + return a - b",
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 0.002
|
||||
}
|
||||
+16
@@ -0,0 +1,16 @@
|
||||
{
|
||||
"case": "path_escape_rejected",
|
||||
"tool": "file_write",
|
||||
"arguments": {
|
||||
"path": "../../escape.py",
|
||||
"content": "print('escape')\n",
|
||||
"overwrite": true
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Path ../../escape.py is outside workspace directory"
|
||||
},
|
||||
"latency_seconds": 0.002
|
||||
}
|
||||
+19
@@ -0,0 +1,19 @@
|
||||
{
|
||||
"case": "terminal_safe",
|
||||
"tool": "virtual_terminal",
|
||||
"arguments": {
|
||||
"command": "pwd && printf SAFE",
|
||||
"timeout": 10
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"returncode": 0,
|
||||
"stdout": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T062500Z/workspace\nSAFE",
|
||||
"stderr": "",
|
||||
"stdout_file": null,
|
||||
"stderr_file": null
|
||||
},
|
||||
"latency_seconds": 0.007
|
||||
}
|
||||
+15
@@ -0,0 +1,15 @@
|
||||
{
|
||||
"case": "terminal_timeout",
|
||||
"tool": "virtual_terminal",
|
||||
"arguments": {
|
||||
"command": "sleep 2",
|
||||
"timeout": 1
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Command timed out after 1 seconds"
|
||||
},
|
||||
"latency_seconds": 1.008
|
||||
}
|
||||
+15
@@ -0,0 +1,15 @@
|
||||
{
|
||||
"case": "terminal_danger_rejected",
|
||||
"tool": "virtual_terminal",
|
||||
"arguments": {
|
||||
"command": "rm -rf ./should-never-execute",
|
||||
"timeout": 10
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Command execution not approved: The command uses 'rm -rf', which performs irreversible recursive deletion without confirmation prompts. The target directory name 'should-never-execute' strongly signals this operation must not run. Additionally, the relative path means the deletion scope depends on the unknown current working directory, creating risk of unintended data loss."
|
||||
},
|
||||
"latency_seconds": 25.182
|
||||
}
|
||||
+36
@@ -0,0 +1,36 @@
|
||||
{
|
||||
"case": "python_docker_sandbox",
|
||||
"tool": "code_interpreter",
|
||||
"arguments": {
|
||||
"language": "python",
|
||||
"timeout": 30,
|
||||
"code": "import os, json\nprint(json.dumps({'root': os.listdir('/'), 'network_proxy': os.environ.get('HTTPS_PROXY')}))\n"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"status": "success",
|
||||
"language": "python",
|
||||
"stdout": "{\"root\": [\"sys\", \"root\", \"boot\", \"opt\", \"sbin\", \"srv\", \"dev\", \"media\", \"proc\", \"var\", \"tmp\", \"etc\", \"home\", \"lib\", \"usr\", \"run\", \"mnt\", \"bin\", \"workspace\", \".dockerenv\"], \"network_proxy\": null}\n",
|
||||
"stderr": "",
|
||||
"stdout_file": null,
|
||||
"stderr_file": null,
|
||||
"returncode": 0,
|
||||
"error": null,
|
||||
"compile_output": null,
|
||||
"phase": null,
|
||||
"execution_time": 0.15053701400756836,
|
||||
"sandbox": {
|
||||
"kind": "docker",
|
||||
"image": "python:3.11-slim",
|
||||
"network": "none",
|
||||
"rootfs": "read-only",
|
||||
"memory": "256m",
|
||||
"cpus": 1,
|
||||
"pids_limit": 64
|
||||
},
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 0.158
|
||||
}
|
||||
+36
@@ -0,0 +1,36 @@
|
||||
{
|
||||
"case": "python_network_denied",
|
||||
"tool": "code_interpreter",
|
||||
"arguments": {
|
||||
"language": "python",
|
||||
"timeout": 30,
|
||||
"code": "import urllib.request\ntry:\n print(urllib.request.urlopen('https://example.com', timeout=3).status)\nexcept Exception as e:\n print(type(e).__name__, str(e))\n"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"status": "success",
|
||||
"language": "python",
|
||||
"stdout": "URLError <urlopen error [Errno -3] Temporary failure in name resolution>\n",
|
||||
"stderr": "",
|
||||
"stdout_file": null,
|
||||
"stderr_file": null,
|
||||
"returncode": 0,
|
||||
"error": null,
|
||||
"compile_output": null,
|
||||
"phase": null,
|
||||
"execution_time": 0.27611231803894043,
|
||||
"sandbox": {
|
||||
"kind": "docker",
|
||||
"image": "python:3.11-slim",
|
||||
"network": "none",
|
||||
"rootfs": "read-only",
|
||||
"memory": "256m",
|
||||
"cpus": 1,
|
||||
"pids_limit": 64
|
||||
},
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 16.694
|
||||
}
|
||||
+36
@@ -0,0 +1,36 @@
|
||||
{
|
||||
"case": "long_output_persisted",
|
||||
"tool": "code_interpreter",
|
||||
"arguments": {
|
||||
"language": "python",
|
||||
"timeout": 30,
|
||||
"code": "for i in range(260): print(f'LINE-{i:03d}')\n"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"status": "success",
|
||||
"language": "python",
|
||||
"stdout": "LINE-000\nLINE-001\nLINE-002\nLINE-003\nLINE-004\nLINE-005\nLINE-006\nLINE-007\nLINE-008\nLINE-009\nLINE-010\nLINE-011\nLINE-012\nLINE-013\nLINE-014\nLINE-015\nLINE-016\nLINE-017\nLINE-018\nLINE-019\nLINE-020\nLINE-021\nLINE-022\nLINE-023\nLINE-024\nLINE-025\nLINE-026\nLINE-027\nLINE-028\nLINE-029\nLINE-030\nLINE-031\nLINE-032\nLINE-033\nLINE-034\nLINE-035\nLINE-036\nLINE-037\nLINE-038\nLINE-039\nLINE-040\nLINE-041\nLINE-042\nLINE-043\nLINE-044\nLINE-045\nLINE-046\nLINE-047\nLINE-048\nLINE-049\n... [省略 161 行，完整输出已保存至 /var/folders/0l/vk1w1b5n2fxfwdlz3f_w25_w0000gp/T/code_interpreter_output_psx202sw.txt] ...\nLINE-211\nLINE-212\nLINE-213\nLINE-214\nLINE-215\nLINE-216\nLINE-217\nLINE-218\nLINE-219\nLINE-220\nLINE-221\nLINE-222\nLINE-223\nLINE-224\nLINE-225\nLINE-226\nLINE-227\nLINE-228\nLINE-229\nLINE-230\nLINE-231\nLINE-232\nLINE-233\nLINE-234\nLINE-235\nLINE-236\nLINE-237\nLINE-238\nLINE-239\nLINE-240\nLINE-241\nLINE-242\nLINE-243\nLINE-244\nLINE-245\nLINE-246\nLINE-247\nLINE-248\nLINE-249\nLINE-250\nLINE-251\nLINE-252\nLINE-253\nLINE-254\nLINE-255\nLINE-256\nLINE-257\nLINE-258\nLINE-259\n\n[如需完整输出，请使用 read_file 工具读取 /var/folders/0l/vk1w1b5n2fxfwdlz3f_w25_w0000gp/T/code_interpreter_output_psx202sw.txt]",
|
||||
"stderr": "",
|
||||
"stdout_file": "/var/folders/0l/vk1w1b5n2fxfwdlz3f_w25_w0000gp/T/code_interpreter_output_psx202sw.txt",
|
||||
"stderr_file": null,
|
||||
"returncode": 0,
|
||||
"error": null,
|
||||
"compile_output": null,
|
||||
"phase": null,
|
||||
"execution_time": 0.12428689002990723,
|
||||
"sandbox": {
|
||||
"kind": "docker",
|
||||
"image": "python:3.11-slim",
|
||||
"network": "none",
|
||||
"rootfs": "read-only",
|
||||
"memory": "256m",
|
||||
"cpus": 1,
|
||||
"pids_limit": 64
|
||||
},
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 0.135
|
||||
}
|
||||
+48
@@ -0,0 +1,48 @@
|
||||
{
|
||||
"case": "excel_formula_screenshot",
|
||||
"tool": "excel_create_with_formula_and_screenshot",
|
||||
"arguments": {
|
||||
"output_path": "invoice.xlsx",
|
||||
"rows": [
|
||||
{
|
||||
"item": "Compute",
|
||||
"quantity": 2,
|
||||
"unit_price": 12.5
|
||||
},
|
||||
{
|
||||
"item": "Storage",
|
||||
"quantity": 3,
|
||||
"unit_price": 7.0
|
||||
}
|
||||
]
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"xlsx": {
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T062500Z/workspace/invoice.xlsx",
|
||||
"bytes": 5059,
|
||||
"sha256": "621c89e193243ab849d1d20df53c9deaaabe2e8909712edd87d6c137f5551470"
|
||||
},
|
||||
"pdf": {
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T062500Z/workspace/invoice.pdf",
|
||||
"bytes": 19487,
|
||||
"sha256": "a398997cace6b14ee4cd8575124a3be4fd0f1c125958c2e99cbd6ae9f91dbbf5"
|
||||
},
|
||||
"screenshot": {
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T062500Z/workspace/invoice.png",
|
||||
"bytes": 14315,
|
||||
"sha256": "b97c76df59cc7972772341ea3a391bb4dba8907b65000f1b360005804d39986d"
|
||||
},
|
||||
"formula_cells": [
|
||||
"D2",
|
||||
"D3",
|
||||
"D4"
|
||||
],
|
||||
"rows": 2,
|
||||
"renderer": "LibreOffice headless + PyMuPDF",
|
||||
"latency_seconds": 10.029
|
||||
},
|
||||
"latency_seconds": 10.136
|
||||
}
|
||||
+45
@@ -0,0 +1,45 @@
|
||||
{
|
||||
"case": "real_webhook",
|
||||
"tool": "webhook_post",
|
||||
"arguments": {
|
||||
"url": "https://postman-echo.com/post",
|
||||
"payload": {
|
||||
"experiment": "4-3",
|
||||
"marker": "REAL-WEBHOOK-RECEIPT"
|
||||
}
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"status": 200,
|
||||
"url": "https://postman-echo.com/post",
|
||||
"response": {
|
||||
"args": {},
|
||||
"data": {
|
||||
"experiment": "4-3",
|
||||
"marker": "REAL-WEBHOOK-RECEIPT"
|
||||
},
|
||||
"files": {},
|
||||
"form": {},
|
||||
"headers": {
|
||||
"host": "postman-echo.com",
|
||||
"content-length": "52",
|
||||
"accept": "*/*",
|
||||
"content-type": "application/json",
|
||||
"user-agent": "python-httpx/0.28.1",
|
||||
"x-forwarded-proto": "https",
|
||||
"accept-encoding": "gzip, br"
|
||||
},
|
||||
"json": {
|
||||
"experiment": "4-3",
|
||||
"marker": "REAL-WEBHOOK-RECEIPT"
|
||||
},
|
||||
"url": "https://postman-echo.com/post"
|
||||
},
|
||||
"response_sha256": "727438d7ad2e56ef124d7b14904d546de199bf2bda40bb56989885625ecc45f2",
|
||||
"response_bytes": 391,
|
||||
"latency_seconds": 1.728
|
||||
},
|
||||
"latency_seconds": 1.731
|
||||
}
|
||||
+25
@@ -0,0 +1,25 @@
|
||||
{
|
||||
"case": "real_browser",
|
||||
"tool": "browser_navigate",
|
||||
"arguments": {
|
||||
"url": "https://example.com",
|
||||
"screenshot_path": "browser-example.png"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"url": "https://example.com",
|
||||
"status": 200,
|
||||
"title": "Example Domain",
|
||||
"body_text": "Example Domain\n\nThis domain is for use in documentation examples without needing permission. Avoid use in operations.\n\nLearn more",
|
||||
"screenshot": {
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T062500Z/workspace/browser-example.png",
|
||||
"bytes": 16578,
|
||||
"sha256": "f21d7a2b1f7739641b8838e2ed2a9a907559cece9397568db6e1ccca197cc7b0"
|
||||
},
|
||||
"browser": "Chromium via Playwright",
|
||||
"latency_seconds": 2.984
|
||||
},
|
||||
"latency_seconds": 2.986
|
||||
}
|
||||
+16
@@ -0,0 +1,16 @@
|
||||
{
|
||||
"case": "calendar_preflight",
|
||||
"tool": "google_calendar_add",
|
||||
"arguments": {
|
||||
"summary": "Experiment 4-3",
|
||||
"start_time": "2026-08-01T10:00:00+00:00",
|
||||
"end_time": "2026-08-01T10:30:00+00:00"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Failed to initialize Google Calendar: Credentials file not found: credentials.json"
|
||||
},
|
||||
"latency_seconds": 0.002
|
||||
}
|
||||
+17
@@ -0,0 +1,17 @@
|
||||
{
|
||||
"case": "github_pr_preflight",
|
||||
"tool": "github_create_pr",
|
||||
"arguments": {
|
||||
"repo_name": "bojieli/ai-agent-book",
|
||||
"title": "Experiment 4-3 preflight",
|
||||
"body": "Credential-gated preflight",
|
||||
"head_branch": "nonexistent-exp4-3"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Failed to initialize GitHub client: GitHub token not configured"
|
||||
},
|
||||
"latency_seconds": 0.002
|
||||
}
|
||||
+16
@@ -0,0 +1,16 @@
|
||||
{
|
||||
"case": "desktop_mobile_capabilities",
|
||||
"tool": "environment_capabilities",
|
||||
"arguments": {},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"computer_use_container_image_present": true,
|
||||
"computer_use_active_session": false,
|
||||
"android_world_adb_present": true,
|
||||
"android_active_devices": [],
|
||||
"note": "Availability probe only; absent active sessions cannot satisfy execution gates."
|
||||
},
|
||||
"latency_seconds": 0.036
|
||||
}
|
||||
+38
@@ -0,0 +1,38 @@
|
||||
{
|
||||
"experiment": "4-3",
|
||||
"campaign_id": "real_mcp_20260730T062500Z",
|
||||
"generated_at": "2026-07-29T22:20:37.628241+00:00",
|
||||
"status": "blocked",
|
||||
"official_complete": false,
|
||||
"gates": {
|
||||
"real_mcp_catalog_and_calls": true,
|
||||
"python_and_javascript_linter": true,
|
||||
"file_edit_verified_and_escape_rejected": true,
|
||||
"terminal_timeout_and_llm_danger_review": true,
|
||||
"real_python_container_sandbox": true,
|
||||
"long_output_truncated_and_persisted": true,
|
||||
"real_excel_formula_and_screenshot": true,
|
||||
"real_webhook": true,
|
||||
"real_browser": true,
|
||||
"real_calendar_mutation": false,
|
||||
"real_github_pr_mutation": false,
|
||||
"real_email_mutation": false,
|
||||
"real_virtual_desktop_session": false,
|
||||
"real_virtual_mobile_session": false,
|
||||
"credential_free_usage_latency_receipts": true
|
||||
},
|
||||
"long_output_full_file": {
|
||||
"path": "/var/folders/0l/vk1w1b5n2fxfwdlz3f_w25_w0000gp/T/code_interpreter_output_psx202sw.txt",
|
||||
"bytes": 2340,
|
||||
"sha256": "86b815da715192ef997d8a8d0c6adcaa8b1fdf9aa349889a4fc46462e4c02cb1"
|
||||
},
|
||||
"blockers": [
|
||||
"real_calendar_mutation",
|
||||
"real_github_pr_mutation",
|
||||
"real_email_mutation",
|
||||
"real_virtual_desktop_session",
|
||||
"real_virtual_mobile_session"
|
||||
],
|
||||
"receipt_count": 18,
|
||||
"llm_call_count": 2
|
||||
}
|
||||
BIN
Binary file not shown.
|
After Width: | Height: | Size: 16 KiB |
BIN
Binary file not shown.
BIN
Binary file not shown.
|
After Width: | Height: | Size: 14 KiB |
BIN
Binary file not shown.
+2
@@ -0,0 +1,2 @@
|
||||
const answer = 42;
|
||||
console.log(answer);
|
||||
+2
@@ -0,0 +1,2 @@
|
||||
def add(a, b):
|
||||
return a - b
|
||||
+260
@@ -0,0 +1,260 @@
|
||||
LINE-000
|
||||
LINE-001
|
||||
LINE-002
|
||||
LINE-003
|
||||
LINE-004
|
||||
LINE-005
|
||||
LINE-006
|
||||
LINE-007
|
||||
LINE-008
|
||||
LINE-009
|
||||
LINE-010
|
||||
LINE-011
|
||||
LINE-012
|
||||
LINE-013
|
||||
LINE-014
|
||||
LINE-015
|
||||
LINE-016
|
||||
LINE-017
|
||||
LINE-018
|
||||
LINE-019
|
||||
LINE-020
|
||||
LINE-021
|
||||
LINE-022
|
||||
LINE-023
|
||||
LINE-024
|
||||
LINE-025
|
||||
LINE-026
|
||||
LINE-027
|
||||
LINE-028
|
||||
LINE-029
|
||||
LINE-030
|
||||
LINE-031
|
||||
LINE-032
|
||||
LINE-033
|
||||
LINE-034
|
||||
LINE-035
|
||||
LINE-036
|
||||
LINE-037
|
||||
LINE-038
|
||||
LINE-039
|
||||
LINE-040
|
||||
LINE-041
|
||||
LINE-042
|
||||
LINE-043
|
||||
LINE-044
|
||||
LINE-045
|
||||
LINE-046
|
||||
LINE-047
|
||||
LINE-048
|
||||
LINE-049
|
||||
LINE-050
|
||||
LINE-051
|
||||
LINE-052
|
||||
LINE-053
|
||||
LINE-054
|
||||
LINE-055
|
||||
LINE-056
|
||||
LINE-057
|
||||
LINE-058
|
||||
LINE-059
|
||||
LINE-060
|
||||
LINE-061
|
||||
LINE-062
|
||||
LINE-063
|
||||
LINE-064
|
||||
LINE-065
|
||||
LINE-066
|
||||
LINE-067
|
||||
LINE-068
|
||||
LINE-069
|
||||
LINE-070
|
||||
LINE-071
|
||||
LINE-072
|
||||
LINE-073
|
||||
LINE-074
|
||||
LINE-075
|
||||
LINE-076
|
||||
LINE-077
|
||||
LINE-078
|
||||
LINE-079
|
||||
LINE-080
|
||||
LINE-081
|
||||
LINE-082
|
||||
LINE-083
|
||||
LINE-084
|
||||
LINE-085
|
||||
LINE-086
|
||||
LINE-087
|
||||
LINE-088
|
||||
LINE-089
|
||||
LINE-090
|
||||
LINE-091
|
||||
LINE-092
|
||||
LINE-093
|
||||
LINE-094
|
||||
LINE-095
|
||||
LINE-096
|
||||
LINE-097
|
||||
LINE-098
|
||||
LINE-099
|
||||
LINE-100
|
||||
LINE-101
|
||||
LINE-102
|
||||
LINE-103
|
||||
LINE-104
|
||||
LINE-105
|
||||
LINE-106
|
||||
LINE-107
|
||||
LINE-108
|
||||
LINE-109
|
||||
LINE-110
|
||||
LINE-111
|
||||
LINE-112
|
||||
LINE-113
|
||||
LINE-114
|
||||
LINE-115
|
||||
LINE-116
|
||||
LINE-117
|
||||
LINE-118
|
||||
LINE-119
|
||||
LINE-120
|
||||
LINE-121
|
||||
LINE-122
|
||||
LINE-123
|
||||
LINE-124
|
||||
LINE-125
|
||||
LINE-126
|
||||
LINE-127
|
||||
LINE-128
|
||||
LINE-129
|
||||
LINE-130
|
||||
LINE-131
|
||||
LINE-132
|
||||
LINE-133
|
||||
LINE-134
|
||||
LINE-135
|
||||
LINE-136
|
||||
LINE-137
|
||||
LINE-138
|
||||
LINE-139
|
||||
LINE-140
|
||||
LINE-141
|
||||
LINE-142
|
||||
LINE-143
|
||||
LINE-144
|
||||
LINE-145
|
||||
LINE-146
|
||||
LINE-147
|
||||
LINE-148
|
||||
LINE-149
|
||||
LINE-150
|
||||
LINE-151
|
||||
LINE-152
|
||||
LINE-153
|
||||
LINE-154
|
||||
LINE-155
|
||||
LINE-156
|
||||
LINE-157
|
||||
LINE-158
|
||||
LINE-159
|
||||
LINE-160
|
||||
LINE-161
|
||||
LINE-162
|
||||
LINE-163
|
||||
LINE-164
|
||||
LINE-165
|
||||
LINE-166
|
||||
LINE-167
|
||||
LINE-168
|
||||
LINE-169
|
||||
LINE-170
|
||||
LINE-171
|
||||
LINE-172
|
||||
LINE-173
|
||||
LINE-174
|
||||
LINE-175
|
||||
LINE-176
|
||||
LINE-177
|
||||
LINE-178
|
||||
LINE-179
|
||||
LINE-180
|
||||
LINE-181
|
||||
LINE-182
|
||||
LINE-183
|
||||
LINE-184
|
||||
LINE-185
|
||||
LINE-186
|
||||
LINE-187
|
||||
LINE-188
|
||||
LINE-189
|
||||
LINE-190
|
||||
LINE-191
|
||||
LINE-192
|
||||
LINE-193
|
||||
LINE-194
|
||||
LINE-195
|
||||
LINE-196
|
||||
LINE-197
|
||||
LINE-198
|
||||
LINE-199
|
||||
LINE-200
|
||||
LINE-201
|
||||
LINE-202
|
||||
LINE-203
|
||||
LINE-204
|
||||
LINE-205
|
||||
LINE-206
|
||||
LINE-207
|
||||
LINE-208
|
||||
LINE-209
|
||||
LINE-210
|
||||
LINE-211
|
||||
LINE-212
|
||||
LINE-213
|
||||
LINE-214
|
||||
LINE-215
|
||||
LINE-216
|
||||
LINE-217
|
||||
LINE-218
|
||||
LINE-219
|
||||
LINE-220
|
||||
LINE-221
|
||||
LINE-222
|
||||
LINE-223
|
||||
LINE-224
|
||||
LINE-225
|
||||
LINE-226
|
||||
LINE-227
|
||||
LINE-228
|
||||
LINE-229
|
||||
LINE-230
|
||||
LINE-231
|
||||
LINE-232
|
||||
LINE-233
|
||||
LINE-234
|
||||
LINE-235
|
||||
LINE-236
|
||||
LINE-237
|
||||
LINE-238
|
||||
LINE-239
|
||||
LINE-240
|
||||
LINE-241
|
||||
LINE-242
|
||||
LINE-243
|
||||
LINE-244
|
||||
LINE-245
|
||||
LINE-246
|
||||
LINE-247
|
||||
LINE-248
|
||||
LINE-249
|
||||
LINE-250
|
||||
LINE-251
|
||||
LINE-252
|
||||
LINE-253
|
||||
LINE-254
|
||||
LINE-255
|
||||
LINE-256
|
||||
LINE-257
|
||||
LINE-258
|
||||
LINE-259
|
||||
+272
@@ -0,0 +1,272 @@
|
||||
{
|
||||
"transport": "mcp-stdio",
|
||||
"server_name": "execution-tools",
|
||||
"server_version": "1.0.0",
|
||||
"schemas": [
|
||||
{
|
||||
"name": "file_write",
|
||||
"description": "Write content to a file with automatic syntax verification",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"path": {
|
||||
"type": "string",
|
||||
"description": "File path (relative to workspace or absolute)"
|
||||
},
|
||||
"content": {
|
||||
"type": "string",
|
||||
"description": "Content to write"
|
||||
},
|
||||
"overwrite": {
|
||||
"type": "boolean",
|
||||
"description": "Whether to overwrite existing files",
|
||||
"default": false
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"path",
|
||||
"content"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "file_edit",
|
||||
"description": "Edit an existing file by searching and replacing content",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"path": {
|
||||
"type": "string",
|
||||
"description": "File path"
|
||||
},
|
||||
"search": {
|
||||
"type": "string",
|
||||
"description": "Text to search for"
|
||||
},
|
||||
"replace": {
|
||||
"type": "string",
|
||||
"description": "Replacement text"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"path",
|
||||
"search",
|
||||
"replace"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "code_interpreter",
|
||||
"description": "Execute code in multiple programming languages in a sandboxed environment with result analysis. Supports: Python, JavaScript, TypeScript, Go, Java, C++, Rust, PHP, Bash",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"code": {
|
||||
"type": "string",
|
||||
"description": "Code to execute"
|
||||
},
|
||||
"language": {
|
||||
"type": "string",
|
||||
"description": "Programming language (python, javascript, typescript, go, java, cpp, rust, php, bash)",
|
||||
"default": "python"
|
||||
},
|
||||
"timeout": {
|
||||
"type": "number",
|
||||
"description": "Execution timeout in seconds",
|
||||
"default": 30.0
|
||||
},
|
||||
"stdin": {
|
||||
"type": "string",
|
||||
"description": "Optional stdin input for the program"
|
||||
},
|
||||
"files": {
|
||||
"type": "object",
|
||||
"description": "Optional additional files (filename -> content mapping)",
|
||||
"additionalProperties": {
|
||||
"type": "string"
|
||||
}
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"code"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "virtual_terminal",
|
||||
"description": "Execute shell commands with error summarization",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"command": {
|
||||
"type": "string",
|
||||
"description": "Shell command to execute"
|
||||
},
|
||||
"timeout": {
|
||||
"type": "integer",
|
||||
"description": "Timeout in seconds",
|
||||
"default": 30
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"command"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "google_calendar_add",
|
||||
"description": "Add an event to Google Calendar",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"summary": {
|
||||
"type": "string",
|
||||
"description": "Event title"
|
||||
},
|
||||
"start_time": {
|
||||
"type": "string",
|
||||
"description": "Start time (ISO 8601 format, e.g., 2024-01-01T10:00:00)"
|
||||
},
|
||||
"end_time": {
|
||||
"type": "string",
|
||||
"description": "End time (ISO 8601 format)"
|
||||
},
|
||||
"description": {
|
||||
"type": "string",
|
||||
"description": "Event description"
|
||||
},
|
||||
"location": {
|
||||
"type": "string",
|
||||
"description": "Event location"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"summary",
|
||||
"start_time",
|
||||
"end_time"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "github_create_pr",
|
||||
"description": "Create a GitHub Pull Request",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"repo_name": {
|
||||
"type": "string",
|
||||
"description": "Repository name (format: owner/repo)"
|
||||
},
|
||||
"title": {
|
||||
"type": "string",
|
||||
"description": "PR title"
|
||||
},
|
||||
"body": {
|
||||
"type": "string",
|
||||
"description": "PR description"
|
||||
},
|
||||
"head_branch": {
|
||||
"type": "string",
|
||||
"description": "Source branch"
|
||||
},
|
||||
"base_branch": {
|
||||
"type": "string",
|
||||
"description": "Target branch",
|
||||
"default": "main"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"repo_name",
|
||||
"title",
|
||||
"body",
|
||||
"head_branch"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "excel_create_with_formula_and_screenshot",
|
||||
"description": "Create an XLSX workbook, apply formulas, and render a real screenshot with LibreOffice",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"output_path": {
|
||||
"type": "string"
|
||||
},
|
||||
"rows": {
|
||||
"type": "array",
|
||||
"items": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"item": {
|
||||
"type": "string"
|
||||
},
|
||||
"quantity": {
|
||||
"type": "number"
|
||||
},
|
||||
"unit_price": {
|
||||
"type": "number"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"item",
|
||||
"quantity",
|
||||
"unit_price"
|
||||
]
|
||||
}
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"output_path",
|
||||
"rows"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "webhook_post",
|
||||
"description": "POST JSON to a real HTTPS webhook endpoint",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"url": {
|
||||
"type": "string"
|
||||
},
|
||||
"payload": {
|
||||
"type": "object"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"url",
|
||||
"payload"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "browser_navigate",
|
||||
"description": "Navigate with real headless Chromium, extract page content, and save a screenshot",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"url": {
|
||||
"type": "string"
|
||||
},
|
||||
"screenshot_path": {
|
||||
"type": "string"
|
||||
}
|
||||
},
|
||||
"required": [
|
||||
"url",
|
||||
"screenshot_path"
|
||||
]
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "environment_capabilities",
|
||||
"description": "Inspect real Computer Use container and Android device availability",
|
||||
"inputSchema": {
|
||||
"type": "object",
|
||||
"properties": {}
|
||||
}
|
||||
}
|
||||
],
|
||||
"schema_sha256": "d9a5bd49b95a42874d0d588926f4c717df2344c86f22991624c3056720649e74"
|
||||
}
|
||||
+66
@@ -0,0 +1,66 @@
|
||||
[
|
||||
{
|
||||
"purpose": "dangerous_operation_review",
|
||||
"called_at_utc": "2026-07-29T22:26:40.931107+00:00",
|
||||
"provider": "kimi",
|
||||
"request": {
|
||||
"model": "kimi-k3",
|
||||
"messages": [
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are a cautious safety reviewer. Approve operations that are safe and reject risky ones."
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": "You are a safety reviewer for an AI agent execution system.\nReview the following operation and determine if it should be approved.\n\nOperation: terminal_command\nDetails: {\n \"command\": \"rm -rf ./should-never-execute\",\n \"detected_patterns\": [\n \"rm -rf\"\n ]\n}\n\nAnalyze the operation for:\n1. Potential data loss or destructive actions\n2. Security risks\n3. Resource consumption concerns\n4. Compliance with best practices\n\nRespond in JSON format:\n{\n \"approved\": true/false,\n \"reason\": \"Brief explanation of your decision\",\n \"risk_level\": \"low/medium/high\",\n \"recommendations\": [\"List of recommendations if any\"]\n}\n"
|
||||
}
|
||||
],
|
||||
"temperature": 1,
|
||||
"max_tokens": 4096
|
||||
},
|
||||
"response": {
|
||||
"id": "chatcmpl-6a6a7e10ec6f19570af6af7a",
|
||||
"model": "kimi-k3",
|
||||
"finish_reason": "stop",
|
||||
"content": "```json\n{\n \"approved\": false,\n \"reason\": \"The command 'rm -rf' performs recursive, forced deletion without confirmation, which is inherently destructive and irreversible. The target directory name 'should-never-execute' further signals this operation should not proceed. There is no verification of the path's contents, no backup mechanism, and no explicit user confirmation for this destructive action.\",\n \"risk_level\": \"high\",\n \"recommendations\": [\n \"Avoid using 'rm -rf' with force flags; require explicit user confirmation before any recursive deletion\",\n \"Verify the target path contents with 'ls -la ./should-never-execute' before any deletion attempt\",\n \"Use safer alternatives such as moving the directory to a trash location (e.g., 'mv ./should-never-execute ~/.trash/') for recoverability\",\n \"Create a backup of the directory before deletion if the data has any potential value\",\n \"Implement path validation to prevent deletion of directories with suspicious or protected names\",\n \"Use 'rm -ri' for interactive deletion if removal is truly intended and confirmed by the user\"\n ]\n}\n```"
|
||||
},
|
||||
"usage": {
|
||||
"prompt_tokens": 259,
|
||||
"completion_tokens": 505,
|
||||
"total_tokens": 764
|
||||
},
|
||||
"latency_seconds": 18.213
|
||||
},
|
||||
{
|
||||
"purpose": "dangerous_operation_review",
|
||||
"called_at_utc": "2026-07-29T22:26:58.129910+00:00",
|
||||
"provider": "kimi",
|
||||
"request": {
|
||||
"model": "kimi-k3",
|
||||
"messages": [
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are a cautious safety reviewer. Approve operations that are safe and reject risky ones."
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": "You are a safety reviewer for an AI agent execution system.\nReview the following operation and determine if it should be approved.\n\nOperation: code_execution\nDetails: {\n \"code\": \"import urllib.request\\ntry:\\n print(urllib.request.urlopen('https://example.com', timeout=3).status)\\nexcept Exception as e:\\n print(type(e).__name__, str(e))\\n\",\n \"language\": \"python\",\n \"detected_patterns\": [\n \"open(\"\n ]\n}\n\nAnalyze the operation for:\n1. Potential data loss or destructive actions\n2. Security risks\n3. Resource consumption concerns\n4. Compliance with best practices\n\nRespond in JSON format:\n{\n \"approved\": true/false,\n \"reason\": \"Brief explanation of your decision\",\n \"risk_level\": \"low/medium/high\",\n \"recommendations\": [\"List of recommendations if any\"]\n}\n"
|
||||
}
|
||||
],
|
||||
"temperature": 1,
|
||||
"max_tokens": 4096
|
||||
},
|
||||
"response": {
|
||||
"id": "chatcmpl-6a6a7e229eab2b3298a1be04",
|
||||
"model": "kimi-k3",
|
||||
"finish_reason": "stop",
|
||||
"content": "```json\n{\n \"approved\": true,\n \"reason\": \"The operation is a benign read-only HTTPS GET request to example.com (an IANA-reserved documentation/test domain) that simply prints the HTTP status code. It includes a timeout and proper exception handling, involves no destructive actions, no data writes, no credential handling, and minimal resource use. The detected 'open(' pattern is a false positive from urllib.request.urlopen(), not a local file operation.\",\n \"risk_level\": \"low\",\n \"recommendations\": [\n \"If the execution environment has an egress policy, confirm that outbound network access to arbitrary hosts is permitted, even though this request targets a harmless reserved domain.\",\n \"Consider logging outbound network requests for audit purposes in production agent environments.\"\n ]\n}\n```"
|
||||
},
|
||||
"usage": {
|
||||
"prompt_tokens": 298,
|
||||
"completion_tokens": 504,
|
||||
"total_tokens": 802
|
||||
},
|
||||
"latency_seconds": 17.028
|
||||
}
|
||||
]
|
||||
+66
@@ -0,0 +1,66 @@
|
||||
[
|
||||
{
|
||||
"purpose": "dangerous_operation_review",
|
||||
"called_at_utc": "2026-07-29T22:26:40.931107+00:00",
|
||||
"provider": "kimi",
|
||||
"request": {
|
||||
"model": "kimi-k3",
|
||||
"messages": [
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are a cautious safety reviewer. Approve operations that are safe and reject risky ones."
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": "You are a safety reviewer for an AI agent execution system.\nReview the following operation and determine if it should be approved.\n\nOperation: terminal_command\nDetails: {\n \"command\": \"rm -rf ./should-never-execute\",\n \"detected_patterns\": [\n \"rm -rf\"\n ]\n}\n\nAnalyze the operation for:\n1. Potential data loss or destructive actions\n2. Security risks\n3. Resource consumption concerns\n4. Compliance with best practices\n\nRespond in JSON format:\n{\n \"approved\": true/false,\n \"reason\": \"Brief explanation of your decision\",\n \"risk_level\": \"low/medium/high\",\n \"recommendations\": [\"List of recommendations if any\"]\n}\n"
|
||||
}
|
||||
],
|
||||
"temperature": 1,
|
||||
"max_tokens": 4096
|
||||
},
|
||||
"response": {
|
||||
"id": "chatcmpl-6a6a7e10ec6f19570af6af7a",
|
||||
"model": "kimi-k3",
|
||||
"finish_reason": "stop",
|
||||
"content": "```json\n{\n \"approved\": false,\n \"reason\": \"The command 'rm -rf' performs recursive, forced deletion without confirmation, which is inherently destructive and irreversible. The target directory name 'should-never-execute' further signals this operation should not proceed. There is no verification of the path's contents, no backup mechanism, and no explicit user confirmation for this destructive action.\",\n \"risk_level\": \"high\",\n \"recommendations\": [\n \"Avoid using 'rm -rf' with force flags; require explicit user confirmation before any recursive deletion\",\n \"Verify the target path contents with 'ls -la ./should-never-execute' before any deletion attempt\",\n \"Use safer alternatives such as moving the directory to a trash location (e.g., 'mv ./should-never-execute ~/.trash/') for recoverability\",\n \"Create a backup of the directory before deletion if the data has any potential value\",\n \"Implement path validation to prevent deletion of directories with suspicious or protected names\",\n \"Use 'rm -ri' for interactive deletion if removal is truly intended and confirmed by the user\"\n ]\n}\n```"
|
||||
},
|
||||
"usage": {
|
||||
"prompt_tokens": 259,
|
||||
"completion_tokens": 505,
|
||||
"total_tokens": 764
|
||||
},
|
||||
"latency_seconds": 18.213
|
||||
},
|
||||
{
|
||||
"purpose": "dangerous_operation_review",
|
||||
"called_at_utc": "2026-07-29T22:26:58.129910+00:00",
|
||||
"provider": "kimi",
|
||||
"request": {
|
||||
"model": "kimi-k3",
|
||||
"messages": [
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are a cautious safety reviewer. Approve operations that are safe and reject risky ones."
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": "You are a safety reviewer for an AI agent execution system.\nReview the following operation and determine if it should be approved.\n\nOperation: code_execution\nDetails: {\n \"code\": \"import urllib.request\\ntry:\\n print(urllib.request.urlopen('https://example.com', timeout=3).status)\\nexcept Exception as e:\\n print(type(e).__name__, str(e))\\n\",\n \"language\": \"python\",\n \"detected_patterns\": [\n \"open(\"\n ]\n}\n\nAnalyze the operation for:\n1. Potential data loss or destructive actions\n2. Security risks\n3. Resource consumption concerns\n4. Compliance with best practices\n\nRespond in JSON format:\n{\n \"approved\": true/false,\n \"reason\": \"Brief explanation of your decision\",\n \"risk_level\": \"low/medium/high\",\n \"recommendations\": [\"List of recommendations if any\"]\n}\n"
|
||||
}
|
||||
],
|
||||
"temperature": 1,
|
||||
"max_tokens": 4096
|
||||
},
|
||||
"response": {
|
||||
"id": "chatcmpl-6a6a7e229eab2b3298a1be04",
|
||||
"model": "kimi-k3",
|
||||
"finish_reason": "stop",
|
||||
"content": "```json\n{\n \"approved\": true,\n \"reason\": \"The operation is a benign read-only HTTPS GET request to example.com (an IANA-reserved documentation/test domain) that simply prints the HTTP status code. It includes a timeout and proper exception handling, involves no destructive actions, no data writes, no credential handling, and minimal resource use. The detected 'open(' pattern is a false positive from urllib.request.urlopen(), not a local file operation.\",\n \"risk_level\": \"low\",\n \"recommendations\": [\n \"If the execution environment has an egress policy, confirm that outbound network access to arbitrary hosts is permitted, even though this request targets a harmless reserved domain.\",\n \"Consider logging outbound network requests for audit purposes in production agent environments.\"\n ]\n}\n```"
|
||||
},
|
||||
"usage": {
|
||||
"prompt_tokens": 298,
|
||||
"completion_tokens": 504,
|
||||
"total_tokens": 802
|
||||
},
|
||||
"latency_seconds": 17.028
|
||||
}
|
||||
]
|
||||
+163
@@ -0,0 +1,163 @@
|
||||
{
|
||||
"experiment": "4-3",
|
||||
"campaign_id": "real_mcp_20260730T070500Z",
|
||||
"status": "blocked",
|
||||
"official_complete": false,
|
||||
"files": [
|
||||
{
|
||||
"path": "artifacts/long_output.full.txt",
|
||||
"bytes": 2340,
|
||||
"sha256": "86b815da715192ef997d8a8d0c6adcaa8b1fdf9aa349889a4fc46462e4c02cb1"
|
||||
},
|
||||
{
|
||||
"path": "catalog.json",
|
||||
"bytes": 7135,
|
||||
"sha256": "a2459670d25b4cce95b6709846934d6600e1c2eac6090cd77ff6139369ad9c58"
|
||||
},
|
||||
{
|
||||
"path": "llm_receipts.checkpoint.json",
|
||||
"bytes": 5112,
|
||||
"sha256": "9bfcd8ee68e880c1d9c3141963d97eb7d1743d71647200d34807465498cfed1b"
|
||||
},
|
||||
{
|
||||
"path": "llm_receipts.json",
|
||||
"bytes": 5113,
|
||||
"sha256": "dabaf8a14e145ebc88d93a4a7032afce049fae98cbf17db957e5f82126f07619"
|
||||
},
|
||||
{
|
||||
"path": "outside-witness.txt",
|
||||
"bytes": 16,
|
||||
"sha256": "de6e8ea7f35c8a0261f7fb9eb75022a92311cfd84248567104f5c9e7d3ddf782"
|
||||
},
|
||||
{
|
||||
"path": "protocol.json",
|
||||
"bytes": 1067,
|
||||
"sha256": "f8ca33de720405502a7f0df26a2c9a4bf3988a6eee235500a1355d258ebaebc0"
|
||||
},
|
||||
{
|
||||
"path": "receipts/01_python_valid_write.json",
|
||||
"bytes": 504,
|
||||
"sha256": "5237151b2fc0efb59da8e02619c6190aef5a7a0400e98b44b859474eabf3e686"
|
||||
},
|
||||
{
|
||||
"path": "receipts/02_python_invalid_rejected.json",
|
||||
"bytes": 417,
|
||||
"sha256": "b42060e04825d24bb20bb646ab0c8a5b11663ff4d25409b1d784d12881831590"
|
||||
},
|
||||
{
|
||||
"path": "receipts/03_javascript_valid_write.json",
|
||||
"bytes": 516,
|
||||
"sha256": "a73b1750929371562748fd73e7f3ca32ed8aa51cb68b98e59dfe5b9dbc23f6d6"
|
||||
},
|
||||
{
|
||||
"path": "receipts/04_javascript_invalid_rejected.json",
|
||||
"bytes": 897,
|
||||
"sha256": "509808a4d105f7e38b22dee9ed24b0d1756e5060c2a92a6143bedf040f902d91"
|
||||
},
|
||||
{
|
||||
"path": "receipts/05_verified_edit.json",
|
||||
"bytes": 519,
|
||||
"sha256": "9bc914d9862f1a7d550c4572335ba789232e45051772edd4525d612d9e93bb9b"
|
||||
},
|
||||
{
|
||||
"path": "receipts/06_path_escape_rejected.json",
|
||||
"bytes": 369,
|
||||
"sha256": "8f778b3d855f06b85c6d81f23d4e905dc46c438a3669d841324e1578e094b678"
|
||||
},
|
||||
{
|
||||
"path": "receipts/07_terminal_safe.json",
|
||||
"bytes": 494,
|
||||
"sha256": "29704f32819b8cb6358b67f93f525aa2098e4f0ba5874aec9f5e43733b14380d"
|
||||
},
|
||||
{
|
||||
"path": "receipts/08_terminal_timeout.json",
|
||||
"bytes": 307,
|
||||
"sha256": "692ae82a1a80008ddabede71d795ec367f95a8ab30ed808f2cb67c72930fae60"
|
||||
},
|
||||
{
|
||||
"path": "receipts/09_terminal_danger_rejected.json",
|
||||
"bytes": 699,
|
||||
"sha256": "0f4d4d9d6f053e7fd4d08801646e3cec3485b5610ad5ebd2287a2077aafb596c"
|
||||
},
|
||||
{
|
||||
"path": "receipts/10_python_docker_sandbox.json",
|
||||
"bytes": 1126,
|
||||
"sha256": "20a912310d8fc364dd2a9fbac4ddcf2ac5fe579b105e76e9864009a0a5be7985"
|
||||
},
|
||||
{
|
||||
"path": "receipts/11_python_network_denied.json",
|
||||
"bytes": 1011,
|
||||
"sha256": "e211efef5aa93802c96816cb2369bc78205c9611f4204ce266c87add833e3c2f"
|
||||
},
|
||||
{
|
||||
"path": "receipts/12_long_output_persisted.json",
|
||||
"bytes": 2176,
|
||||
"sha256": "f673d9820e49343127f133385ac8c49684e40cda5aaca25d586ac951e03f4126"
|
||||
},
|
||||
{
|
||||
"path": "receipts/13_excel_formula_screenshot.json",
|
||||
"bytes": 1466,
|
||||
"sha256": "1e276f77ec6f402aa841e213e59c11685f854ad5ed90aff83751e05fc28d5a8d"
|
||||
},
|
||||
{
|
||||
"path": "receipts/14_real_webhook.json",
|
||||
"bytes": 1149,
|
||||
"sha256": "2a18e4e400e5b5ce094289492be430ae0bd35d2ffafdb526e4f95aa7fd7eb9ce"
|
||||
},
|
||||
{
|
||||
"path": "receipts/15_real_browser.json",
|
||||
"bytes": 884,
|
||||
"sha256": "685348ff21b92983b01678c16dc227e1345e9578c998d28ecdc373e408dc5d50"
|
||||
},
|
||||
{
|
||||
"path": "receipts/16_calendar_preflight.json",
|
||||
"bytes": 442,
|
||||
"sha256": "67b317de35158904b5b08e891d4a8e31bbe25a94e4d1cfe88ffeebb3d2433556"
|
||||
},
|
||||
{
|
||||
"path": "receipts/17_github_pr_preflight.json",
|
||||
"bytes": 462,
|
||||
"sha256": "a39311ace49b1e2e4db4c614f59f7e7374b83cad38df38a22be0260a041d8e82"
|
||||
},
|
||||
{
|
||||
"path": "receipts/18_desktop_mobile_capabilities.json",
|
||||
"bytes": 489,
|
||||
"sha256": "2fc4d52f7a7c499e74184572af210e77e6fe0e9633a60f95a69b4cb31e6fa74a"
|
||||
},
|
||||
{
|
||||
"path": "summary.json",
|
||||
"bytes": 1323,
|
||||
"sha256": "1e49d2131e8c694465142c2d7e844f7601026cab8b76592b8ed156233b7fc22c"
|
||||
},
|
||||
{
|
||||
"path": "workspace/browser-example.png",
|
||||
"bytes": 16578,
|
||||
"sha256": "f21d7a2b1f7739641b8838e2ed2a9a907559cece9397568db6e1ccca197cc7b0"
|
||||
},
|
||||
{
|
||||
"path": "workspace/invoice.pdf",
|
||||
"bytes": 19487,
|
||||
"sha256": "d0b2ab6e0bff2e0e77730bb22d5066782618b164a43b76c79a334440f7f3a493"
|
||||
},
|
||||
{
|
||||
"path": "workspace/invoice.png",
|
||||
"bytes": 14315,
|
||||
"sha256": "b97c76df59cc7972772341ea3a391bb4dba8907b65000f1b360005804d39986d"
|
||||
},
|
||||
{
|
||||
"path": "workspace/invoice.xlsx",
|
||||
"bytes": 5060,
|
||||
"sha256": "59aac1249a4b3b9472113eab025d6bb5312341c63cb443a001166b4ff7ab1c8b"
|
||||
},
|
||||
{
|
||||
"path": "workspace/valid.js",
|
||||
"bytes": 40,
|
||||
"sha256": "799574240050acc326491093ba5641e3eb6ec281fc8f7a0a259b80a0a51724cd"
|
||||
},
|
||||
{
|
||||
"path": "workspace/valid.py",
|
||||
"bytes": 32,
|
||||
"sha256": "e1a894022d1a082987b87adecb623438c9e386d86b2b621cff4a5fe7fdf7edc8"
|
||||
}
|
||||
]
|
||||
}
|
||||
+1
@@ -0,0 +1 @@
|
||||
MUST-NOT-CHANGE
|
||||
+45
@@ -0,0 +1,45 @@
|
||||
{
|
||||
"experiment": "4-3",
|
||||
"authority": "book/chapter4.md:274",
|
||||
"required_categories": {
|
||||
"file_write_edit": [
|
||||
"python_linter",
|
||||
"javascript_linter",
|
||||
"structured_errors"
|
||||
],
|
||||
"terminal": [
|
||||
"timeout",
|
||||
"dangerous_command_review",
|
||||
"history_or_receipt"
|
||||
],
|
||||
"code_interpreter": [
|
||||
"real_sandbox",
|
||||
"dangerous_operation_gate",
|
||||
"long_output_persisted"
|
||||
],
|
||||
"data": [
|
||||
"excel_write",
|
||||
"formula",
|
||||
"screenshot"
|
||||
],
|
||||
"external": [
|
||||
"calendar",
|
||||
"github_pr",
|
||||
"email",
|
||||
"webhook"
|
||||
],
|
||||
"gui": [
|
||||
"browser",
|
||||
"virtual_desktop",
|
||||
"virtual_mobile"
|
||||
]
|
||||
},
|
||||
"safety": {
|
||||
"workspace_confinement": true,
|
||||
"automatic_linter": true,
|
||||
"llm_driven_danger_review": true,
|
||||
"long_output_head_tail_and_full_file": true,
|
||||
"credential_free_receipts": true
|
||||
},
|
||||
"completion_rule": "Every named manuscript category must have substantive real execution evidence; missing credentials or active GUI backends produce blocked, never passed."
|
||||
}
|
||||
+18
@@ -0,0 +1,18 @@
|
||||
{
|
||||
"case": "python_valid_write",
|
||||
"tool": "file_write",
|
||||
"arguments": {
|
||||
"path": "valid.py",
|
||||
"content": "def add(a, b):\n return a + b\n",
|
||||
"overwrite": true
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T070500Z/workspace/valid.py",
|
||||
"bytes_written": 32,
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 0.002
|
||||
}
|
||||
+17
@@ -0,0 +1,17 @@
|
||||
{
|
||||
"case": "python_invalid_rejected",
|
||||
"tool": "file_write",
|
||||
"arguments": {
|
||||
"path": "invalid.py",
|
||||
"content": "def broken(:\n pass\n",
|
||||
"overwrite": true
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Syntax validation failed: Syntax error at line 1: invalid syntax",
|
||||
"verification": "failed"
|
||||
},
|
||||
"latency_seconds": 0.002
|
||||
}
|
||||
+18
@@ -0,0 +1,18 @@
|
||||
{
|
||||
"case": "javascript_valid_write",
|
||||
"tool": "file_write",
|
||||
"arguments": {
|
||||
"path": "valid.js",
|
||||
"content": "const answer = 42;\nconsole.log(answer);\n",
|
||||
"overwrite": true
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T070500Z/workspace/valid.js",
|
||||
"bytes_written": 40,
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 0.068
|
||||
}
|
||||
+17
@@ -0,0 +1,17 @@
|
||||
{
|
||||
"case": "javascript_invalid_rejected",
|
||||
"tool": "file_write",
|
||||
"arguments": {
|
||||
"path": "invalid.js",
|
||||
"content": "const broken = ;\n",
|
||||
"overwrite": true
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Syntax validation failed: [stdin]:1\nconst broken = ;\n ^\n\nSyntaxError: Unexpected token ';'\n at wrapSafe (node:internal/modules/cjs/loader:1740:18)\n at checkSyntax (node:internal/main/check_syntax:76:3)\n at node:internal/main/check_syntax:45:5\n at Socket.<anonymous> (node:internal/process/execution:205:5)\n at Socket.emit (node:events:520:22)\n at endReadableNT (node:internal/streams/readable:1729:12)\n at process.processTicksAndRejections (node:internal/process/task_queues:90:21)\n\nNode.js v25.6.0",
|
||||
"verification": "failed"
|
||||
},
|
||||
"latency_seconds": 0.067
|
||||
}
|
||||
+18
@@ -0,0 +1,18 @@
|
||||
{
|
||||
"case": "verified_edit",
|
||||
"tool": "file_edit",
|
||||
"arguments": {
|
||||
"path": "valid.py",
|
||||
"search": "a + b",
|
||||
"replace": "a - b"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T070500Z/workspace/valid.py",
|
||||
"diff_preview": "Line 2:\n - return a + b\n + return a - b",
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 0.002
|
||||
}
|
||||
+16
@@ -0,0 +1,16 @@
|
||||
{
|
||||
"case": "path_escape_rejected",
|
||||
"tool": "file_write",
|
||||
"arguments": {
|
||||
"path": "../../escape.py",
|
||||
"content": "print('escape')\n",
|
||||
"overwrite": true
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Path ../../escape.py is outside workspace directory"
|
||||
},
|
||||
"latency_seconds": 0.002
|
||||
}
|
||||
+19
@@ -0,0 +1,19 @@
|
||||
{
|
||||
"case": "terminal_safe",
|
||||
"tool": "virtual_terminal",
|
||||
"arguments": {
|
||||
"command": "pwd && printf SAFE",
|
||||
"timeout": 10
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"returncode": 0,
|
||||
"stdout": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T070500Z/workspace\nSAFE",
|
||||
"stderr": "",
|
||||
"stdout_file": null,
|
||||
"stderr_file": null
|
||||
},
|
||||
"latency_seconds": 0.008
|
||||
}
|
||||
+15
@@ -0,0 +1,15 @@
|
||||
{
|
||||
"case": "terminal_timeout",
|
||||
"tool": "virtual_terminal",
|
||||
"arguments": {
|
||||
"command": "sleep 2",
|
||||
"timeout": 1
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Command timed out after 1 seconds"
|
||||
},
|
||||
"latency_seconds": 1.008
|
||||
}
|
||||
+15
@@ -0,0 +1,15 @@
|
||||
{
|
||||
"case": "terminal_danger_rejected",
|
||||
"tool": "virtual_terminal",
|
||||
"arguments": {
|
||||
"command": "rm -rf ./should-never-execute",
|
||||
"timeout": 10
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Command execution not approved: The command 'rm -rf' performs recursive, forced deletion without confirmation, which is inherently destructive and irreversible. The target directory name 'should-never-execute' further signals this operation should not proceed. There is no verification of the path's contents, no backup mechanism, and no explicit user confirmation for this destructive action."
|
||||
},
|
||||
"latency_seconds": 18.275
|
||||
}
|
||||
+36
@@ -0,0 +1,36 @@
|
||||
{
|
||||
"case": "python_docker_sandbox",
|
||||
"tool": "code_interpreter",
|
||||
"arguments": {
|
||||
"language": "python",
|
||||
"timeout": 30,
|
||||
"code": "import os, json\nprint(json.dumps({'root': os.listdir('/'), 'network_proxy': os.environ.get('HTTPS_PROXY')}))\n"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"status": "success",
|
||||
"language": "python",
|
||||
"stdout": "{\"root\": [\"sys\", \"root\", \"boot\", \"opt\", \"sbin\", \"srv\", \"dev\", \"media\", \"proc\", \"var\", \"tmp\", \"etc\", \"home\", \"lib\", \"usr\", \"run\", \"mnt\", \"bin\", \"workspace\", \".dockerenv\"], \"network_proxy\": null}\n",
|
||||
"stderr": "",
|
||||
"stdout_file": null,
|
||||
"stderr_file": null,
|
||||
"returncode": 0,
|
||||
"error": null,
|
||||
"compile_output": null,
|
||||
"phase": null,
|
||||
"execution_time": 0.15543341636657715,
|
||||
"sandbox": {
|
||||
"kind": "docker",
|
||||
"image": "python:3.11-slim",
|
||||
"network": "none",
|
||||
"rootfs": "read-only",
|
||||
"memory": "256m",
|
||||
"cpus": 1,
|
||||
"pids_limit": 64
|
||||
},
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 0.164
|
||||
}
|
||||
+36
@@ -0,0 +1,36 @@
|
||||
{
|
||||
"case": "python_network_denied",
|
||||
"tool": "code_interpreter",
|
||||
"arguments": {
|
||||
"language": "python",
|
||||
"timeout": 30,
|
||||
"code": "import urllib.request\ntry:\n print(urllib.request.urlopen('https://example.com', timeout=3).status)\nexcept Exception as e:\n print(type(e).__name__, str(e))\n"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"status": "success",
|
||||
"language": "python",
|
||||
"stdout": "URLError <urlopen error [Errno -3] Temporary failure in name resolution>\n",
|
||||
"stderr": "",
|
||||
"stdout_file": null,
|
||||
"stderr_file": null,
|
||||
"returncode": 0,
|
||||
"error": null,
|
||||
"compile_output": null,
|
||||
"phase": null,
|
||||
"execution_time": 0.25653815269470215,
|
||||
"sandbox": {
|
||||
"kind": "docker",
|
||||
"image": "python:3.11-slim",
|
||||
"network": "none",
|
||||
"rootfs": "read-only",
|
||||
"memory": "256m",
|
||||
"cpus": 1,
|
||||
"pids_limit": 64
|
||||
},
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 17.298
|
||||
}
|
||||
+36
@@ -0,0 +1,36 @@
|
||||
{
|
||||
"case": "long_output_persisted",
|
||||
"tool": "code_interpreter",
|
||||
"arguments": {
|
||||
"language": "python",
|
||||
"timeout": 30,
|
||||
"code": "for i in range(260): print(f'LINE-{i:03d}')\n"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"status": "success",
|
||||
"language": "python",
|
||||
"stdout": "LINE-000\nLINE-001\nLINE-002\nLINE-003\nLINE-004\nLINE-005\nLINE-006\nLINE-007\nLINE-008\nLINE-009\nLINE-010\nLINE-011\nLINE-012\nLINE-013\nLINE-014\nLINE-015\nLINE-016\nLINE-017\nLINE-018\nLINE-019\nLINE-020\nLINE-021\nLINE-022\nLINE-023\nLINE-024\nLINE-025\nLINE-026\nLINE-027\nLINE-028\nLINE-029\nLINE-030\nLINE-031\nLINE-032\nLINE-033\nLINE-034\nLINE-035\nLINE-036\nLINE-037\nLINE-038\nLINE-039\nLINE-040\nLINE-041\nLINE-042\nLINE-043\nLINE-044\nLINE-045\nLINE-046\nLINE-047\nLINE-048\nLINE-049\n... [省略 161 行，完整输出已保存至 /var/folders/0l/vk1w1b5n2fxfwdlz3f_w25_w0000gp/T/code_interpreter_output_v37osy_b.txt] ...\nLINE-211\nLINE-212\nLINE-213\nLINE-214\nLINE-215\nLINE-216\nLINE-217\nLINE-218\nLINE-219\nLINE-220\nLINE-221\nLINE-222\nLINE-223\nLINE-224\nLINE-225\nLINE-226\nLINE-227\nLINE-228\nLINE-229\nLINE-230\nLINE-231\nLINE-232\nLINE-233\nLINE-234\nLINE-235\nLINE-236\nLINE-237\nLINE-238\nLINE-239\nLINE-240\nLINE-241\nLINE-242\nLINE-243\nLINE-244\nLINE-245\nLINE-246\nLINE-247\nLINE-248\nLINE-249\nLINE-250\nLINE-251\nLINE-252\nLINE-253\nLINE-254\nLINE-255\nLINE-256\nLINE-257\nLINE-258\nLINE-259\n\n[如需完整输出，请使用 read_file 工具读取 /var/folders/0l/vk1w1b5n2fxfwdlz3f_w25_w0000gp/T/code_interpreter_output_v37osy_b.txt]",
|
||||
"stderr": "",
|
||||
"stdout_file": "/var/folders/0l/vk1w1b5n2fxfwdlz3f_w25_w0000gp/T/code_interpreter_output_v37osy_b.txt",
|
||||
"stderr_file": null,
|
||||
"returncode": 0,
|
||||
"error": null,
|
||||
"compile_output": null,
|
||||
"phase": null,
|
||||
"execution_time": 0.12924504280090332,
|
||||
"sandbox": {
|
||||
"kind": "docker",
|
||||
"image": "python:3.11-slim",
|
||||
"network": "none",
|
||||
"rootfs": "read-only",
|
||||
"memory": "256m",
|
||||
"cpus": 1,
|
||||
"pids_limit": 64
|
||||
},
|
||||
"verification": "passed"
|
||||
},
|
||||
"latency_seconds": 0.141
|
||||
}
|
||||
+48
@@ -0,0 +1,48 @@
|
||||
{
|
||||
"case": "excel_formula_screenshot",
|
||||
"tool": "excel_create_with_formula_and_screenshot",
|
||||
"arguments": {
|
||||
"output_path": "invoice.xlsx",
|
||||
"rows": [
|
||||
{
|
||||
"item": "Compute",
|
||||
"quantity": 2,
|
||||
"unit_price": 12.5
|
||||
},
|
||||
{
|
||||
"item": "Storage",
|
||||
"quantity": 3,
|
||||
"unit_price": 7.0
|
||||
}
|
||||
]
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"xlsx": {
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T070500Z/workspace/invoice.xlsx",
|
||||
"bytes": 5060,
|
||||
"sha256": "59aac1249a4b3b9472113eab025d6bb5312341c63cb443a001166b4ff7ab1c8b"
|
||||
},
|
||||
"pdf": {
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T070500Z/workspace/invoice.pdf",
|
||||
"bytes": 19487,
|
||||
"sha256": "d0b2ab6e0bff2e0e77730bb22d5066782618b164a43b76c79a334440f7f3a493"
|
||||
},
|
||||
"screenshot": {
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T070500Z/workspace/invoice.png",
|
||||
"bytes": 14315,
|
||||
"sha256": "b97c76df59cc7972772341ea3a391bb4dba8907b65000f1b360005804d39986d"
|
||||
},
|
||||
"formula_cells": [
|
||||
"D2",
|
||||
"D3",
|
||||
"D4"
|
||||
],
|
||||
"rows": 2,
|
||||
"renderer": "LibreOffice headless + PyMuPDF",
|
||||
"latency_seconds": 1.13
|
||||
},
|
||||
"latency_seconds": 1.222
|
||||
}
|
||||
+45
@@ -0,0 +1,45 @@
|
||||
{
|
||||
"case": "real_webhook",
|
||||
"tool": "webhook_post",
|
||||
"arguments": {
|
||||
"url": "https://postman-echo.com/post",
|
||||
"payload": {
|
||||
"experiment": "4-3",
|
||||
"marker": "REAL-WEBHOOK-RECEIPT"
|
||||
}
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"status": 200,
|
||||
"url": "https://postman-echo.com/post",
|
||||
"response": {
|
||||
"args": {},
|
||||
"data": {
|
||||
"experiment": "4-3",
|
||||
"marker": "REAL-WEBHOOK-RECEIPT"
|
||||
},
|
||||
"files": {},
|
||||
"form": {},
|
||||
"headers": {
|
||||
"host": "postman-echo.com",
|
||||
"content-length": "52",
|
||||
"accept": "*/*",
|
||||
"content-type": "application/json",
|
||||
"user-agent": "python-httpx/0.28.1",
|
||||
"x-forwarded-proto": "https",
|
||||
"accept-encoding": "gzip, br"
|
||||
},
|
||||
"json": {
|
||||
"experiment": "4-3",
|
||||
"marker": "REAL-WEBHOOK-RECEIPT"
|
||||
},
|
||||
"url": "https://postman-echo.com/post"
|
||||
},
|
||||
"response_sha256": "727438d7ad2e56ef124d7b14904d546de199bf2bda40bb56989885625ecc45f2",
|
||||
"response_bytes": 391,
|
||||
"latency_seconds": 1.06
|
||||
},
|
||||
"latency_seconds": 1.063
|
||||
}
|
||||
+25
@@ -0,0 +1,25 @@
|
||||
{
|
||||
"case": "real_browser",
|
||||
"tool": "browser_navigate",
|
||||
"arguments": {
|
||||
"url": "https://example.com",
|
||||
"screenshot_path": "browser-example.png"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"url": "https://example.com",
|
||||
"status": 200,
|
||||
"title": "Example Domain",
|
||||
"body_text": "Example Domain\n\nThis domain is for use in documentation examples without needing permission. Avoid use in operations.\n\nLearn more",
|
||||
"screenshot": {
|
||||
"path": "/Users/boj/book/ai-agent-book/chapter4/execution-tools/validation/experiment_4_3/real_mcp_20260730T070500Z/workspace/browser-example.png",
|
||||
"bytes": 16578,
|
||||
"sha256": "f21d7a2b1f7739641b8838e2ed2a9a907559cece9397568db6e1ccca197cc7b0"
|
||||
},
|
||||
"browser": "Chromium via Playwright",
|
||||
"latency_seconds": 1.902
|
||||
},
|
||||
"latency_seconds": 1.906
|
||||
}
|
||||
+16
@@ -0,0 +1,16 @@
|
||||
{
|
||||
"case": "calendar_preflight",
|
||||
"tool": "google_calendar_add",
|
||||
"arguments": {
|
||||
"summary": "Experiment 4-3",
|
||||
"start_time": "2026-08-01T10:00:00+00:00",
|
||||
"end_time": "2026-08-01T10:30:00+00:00"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Failed to initialize Google Calendar: Credentials file not found: credentials.json"
|
||||
},
|
||||
"latency_seconds": 0.002
|
||||
}
|
||||
+17
@@ -0,0 +1,17 @@
|
||||
{
|
||||
"case": "github_pr_preflight",
|
||||
"tool": "github_create_pr",
|
||||
"arguments": {
|
||||
"repo_name": "bojieli/ai-agent-book",
|
||||
"title": "Experiment 4-3 preflight",
|
||||
"body": "Credential-gated preflight",
|
||||
"head_branch": "nonexistent-exp4-3"
|
||||
},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": false,
|
||||
"error": "Failed to initialize GitHub client: GitHub token not configured"
|
||||
},
|
||||
"latency_seconds": 0.002
|
||||
}
|
||||
+16
@@ -0,0 +1,16 @@
|
||||
{
|
||||
"case": "desktop_mobile_capabilities",
|
||||
"tool": "environment_capabilities",
|
||||
"arguments": {},
|
||||
"transport": "mcp-stdio",
|
||||
"mcp_result_is_error": false,
|
||||
"payload": {
|
||||
"success": true,
|
||||
"computer_use_container_image_present": true,
|
||||
"computer_use_active_session": false,
|
||||
"android_world_adb_present": true,
|
||||
"android_active_devices": [],
|
||||
"note": "Availability probe only; absent active sessions cannot satisfy execution gates."
|
||||
},
|
||||
"latency_seconds": 0.035
|
||||
}
|
||||
+39
@@ -0,0 +1,39 @@
|
||||
{
|
||||
"experiment": "4-3",
|
||||
"campaign_id": "real_mcp_20260730T070500Z",
|
||||
"generated_at": "2026-07-29T22:27:02.959073+00:00",
|
||||
"status": "blocked",
|
||||
"official_complete": false,
|
||||
"gates": {
|
||||
"real_mcp_catalog_and_calls": true,
|
||||
"python_and_javascript_linter": true,
|
||||
"file_edit_verified_and_escape_rejected": true,
|
||||
"terminal_timeout_and_llm_danger_review": true,
|
||||
"real_python_container_sandbox": true,
|
||||
"long_output_truncated_and_persisted": true,
|
||||
"real_excel_formula_and_screenshot": true,
|
||||
"real_webhook": true,
|
||||
"real_browser": true,
|
||||
"real_calendar_mutation": false,
|
||||
"real_github_pr_mutation": false,
|
||||
"real_email_mutation": false,
|
||||
"real_virtual_desktop_session": false,
|
||||
"real_virtual_mobile_session": false,
|
||||
"credential_free_usage_latency_receipts": true
|
||||
},
|
||||
"long_output_full_file": {
|
||||
"path": "artifacts/long_output.full.txt",
|
||||
"bytes": 2340,
|
||||
"sha256": "86b815da715192ef997d8a8d0c6adcaa8b1fdf9aa349889a4fc46462e4c02cb1",
|
||||
"source_temp_path_sha256": "db74bcaabc997ed39ea3e943101c35b51e395946dccb8ba84b2da2d20e72a8ce"
|
||||
},
|
||||
"blockers": [
|
||||
"real_calendar_mutation",
|
||||
"real_github_pr_mutation",
|
||||
"real_email_mutation",
|
||||
"real_virtual_desktop_session",
|
||||
"real_virtual_mobile_session"
|
||||
],
|
||||
"receipt_count": 18,
|
||||
"llm_call_count": 2
|
||||
}
|
||||
BIN
Binary file not shown.
|
After Width: | Height: | Size: 16 KiB |
BIN
Binary file not shown.
BIN
Binary file not shown.
|
After Width: | Height: | Size: 14 KiB |
BIN
Binary file not shown.
+2
@@ -0,0 +1,2 @@
|
||||
const answer = 42;
|
||||
console.log(answer);
|
||||
+2
@@ -0,0 +1,2 @@
|
||||
def add(a, b):
|
||||
return a - b
|
||||
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user