ai-agent-book 精选快照（<2MB 代码与文档，来自 github.com/bojieli/ai-agent-book）

chapter10/book-translation/tests/conftest.py

@@ -0,0 +1,26 @@
+"""Shared bootstrap for book-translation regression tests."""
+
+import sys
+from pathlib import Path
+from types import ModuleType
+
+
+PROJECT_ROOT = Path(__file__).resolve().parents[1]
+if str(PROJECT_ROOT) not in sys.path:
+    sys.path.insert(0, str(PROJECT_ROOT))
+
+try:
+    import openai  # noqa: F401
+except ImportError:
+    openai_stub = ModuleType("openai")
+    openai_stub.OpenAI = object
+    sys.modules["openai"] = openai_stub
+
+try:
+    import tiktoken  # noqa: F401
+except ImportError:
+    tiktoken_stub = ModuleType("tiktoken")
+    encoder = type("Enc", (), {"encode": lambda self, text: list(text or "")})
+    tiktoken_stub.encoding_for_model = lambda _model: encoder()
+    tiktoken_stub.get_encoding = lambda _name: encoder()
+    sys.modules["tiktoken"] = tiktoken_stub

chapter10/book-translation/tests/test_glossary_robustness.py

@@ -0,0 +1,61 @@
+"""回归测试：Glossary Agent 返回不合规 JSON 时，run_orchestration 不应崩溃。
+
+覆盖两类模型失误（此前会让整轮管理者模式直接 KeyError/AttributeError）：
+  1) glossary 条目缺 en/zh 键、或值为显式 null / 空串 -> 条目被丢弃；
+  2) 顶层 JSON 是数组而非对象 -> glossary_agent 返回空表。
+不依赖真实 API：llm_chat / get_client 被打桩。
+"""
+
+import json
+
+import agents
+
+# 混合各种坏条目的 glossary：错键名 / null / 空串 都应被丢弃，只有合规条目保留。
+GLOSSARY_JSON = json.dumps({
+    "glossary": [
+        {"term": "token", "translation": "词元"},              # 错键名
+        {"en": None, "zh": "提示词"},                          # 显式 null
+        {"en": "", "zh": "时延"},                              # 空串
+        {"en": "attention", "zh": "注意力", "pos": "名词"},     # 合规
+    ]
+}, ensure_ascii=False)
+
+CHAPTERS = {"Chapter 1: Intro": "# Chapter 1\nSome text about attention."}
+
+
+def _install_fake_llm(glossary_payload=GLOSSARY_JSON):
+    def fake_llm_chat(client, tracker, agent, messages, json_mode=False, note=""):
+        tracker.record(agent, 10, 5, note)
+        if agent == "Glossary":
+            return glossary_payload
+        return "译文"
+    agents.get_client = lambda: object()
+    agents.llm_chat = fake_llm_chat
+
+
+def test_orchestration_skips_malformed_glossary_entries(tmp_path):
+    _install_fake_llm()
+    result = agents.run_orchestration(
+        CHAPTERS, str(tmp_path), enable_glossary=True, enable_proofreading=False)
+    glossary = result["glossary"]
+    # 所有存活条目必须是非空 en/zh 字符串（下游 g["en"]/g["zh"] 索引的前提）
+    for g in glossary:
+        assert isinstance(g["en"], str) and g["en"].strip()
+        assert isinstance(g["zh"], str) and g["zh"].strip()
+    ens = {g["en"] for g in glossary}
+    assert "attention" in ens                       # 合规条目保留
+    assert "term" not in ens                        # 错键名条目已丢弃
+    for en in agents.EDITORIAL_MANDATE:             # 编辑部指定术语仍会补齐
+        assert en in ens
+    assert (tmp_path / "glossary.json").exists()    # 产物正常落盘
+    assert (tmp_path / "chapter1_zh.md").read_text(encoding="utf-8") == "译文"
+
+
+def test_glossary_agent_tolerates_json_array():
+    _install_fake_llm(glossary_payload='["not", "an", "object"]')
+    assert agents.glossary_agent(None, agents.TokenTracker(), "book text") == []
+
+
+def test_glossary_agent_tolerates_missing_glossary_key():
+    _install_fake_llm(glossary_payload='{"terms": []}')
+    assert agents.glossary_agent(None, agents.TokenTracker(), "book text") == []

chapter10/book-translation/tests/test_non_dict_issue_item.py

@@ -0,0 +1,28 @@
+"""Non-dict proofread issue entries must not AttributeError on .get."""
+from agents import _report_issues
+
+
+def test_string_issue_items_dropped():
+    report = {
+        "issues": [
+            "术语不一致：token",
+            {"chapter": "Ch1", "type": "术语不一致", "detail": "用了标记"},
+        ],
+    }
+    issues = _report_issues(report)
+    assert issues == [{"chapter": "Ch1", "type": "术语不一致", "detail": "用了标记"}]
+    details = [
+        i.get("detail", "") for i in issues if i.get("chapter") == "Ch1"
+    ]
+    assert details == ["用了标记"]
+
+
+def test_null_and_dict_issues_still_work():
+    assert _report_issues({"issues": None}) == []
+    assert _report_issues({"issues": [{"chapter": "a", "detail": "x"}]}) == [
+        {"chapter": "a", "detail": "x"}
+    ]
+
+
+def test_issues_scalar_like_empty():
+    assert _report_issues({"issues": "not a list"}) == []

chapter10/book-translation/tests/test_null_glossary.py

@@ -0,0 +1,33 @@
+"""Null glossary from Glossary Agent must behave like empty list."""
+
+import agents
+
+
+def test_glossary_agent_null_glossary_like_empty():
+    def fake_llm_chat(client, tracker, agent, messages, json_mode=False, note=""):
+        tracker.record(agent, 10, 5, note)
+        return '{"glossary": null}'
+
+    agents.llm_chat = fake_llm_chat
+    assert agents.glossary_agent(None, agents.TokenTracker(), "book") == []
+
+
+def test_orchestration_tolerates_null_glossary(tmp_path):
+    def fake_llm_chat(client, tracker, agent, messages, json_mode=False, note=""):
+        tracker.record(agent, 10, 5, note)
+        if agent == "Glossary":
+            return '{"glossary": null}'
+        return "译文"
+
+    agents.get_client = lambda: object()
+    agents.llm_chat = fake_llm_chat
+    result = agents.run_orchestration(
+        {"Chapter 1": "token embedding"},
+        str(tmp_path),
+        enable_glossary=True,
+        enable_proofreading=False,
+    )
+    assert isinstance(result["glossary"], list)
+    for g in result["glossary"]:
+        assert isinstance(g["en"], str) and g["en"].strip()
+    assert result["translations"]["Chapter 1"] == "译文"

chapter10/book-translation/tests/test_null_issues.py

@@ -0,0 +1,15 @@
+"""Null proofread issues must not TypeError when building report summaries."""
+from agents import _report_issues
+
+
+def test_null_issues_like_empty():
+    assert _report_issues({"issues": None}) == []
+    summary_issues = _report_issues({"issues": None})[:5]
+    assert summary_issues == []
+    details = [i.get("detail", "") for i in _report_issues({"issues": None})]
+    assert details == []
+
+
+def test_issues_preserved():
+    issues = [{"chapter": "a", "detail": "fix me"}]
+    assert _report_issues({"issues": issues}) == issues

chapter10/book-translation/tests/test_proofread_non_dict.py

@@ -0,0 +1,54 @@
+"""Proofreading must return a dict when the model emits a JSON array or junk."""
+from agents import _loads_lenient, _report_issues, proofreading_agent
+
+
+def test_loads_lenient_empty_and_junk_return_none():
+    assert _loads_lenient("") is None
+    assert _loads_lenient("not json") is None
+    assert _loads_lenient('{"a": 1}') == {"a": 1}
+
+
+def test_report_issues_non_dict():
+    assert _report_issues([]) == []
+    assert _report_issues(None) == []
+
+
+def test_proofreading_agent_json_array_returns_empty_dict(monkeypatch):
+    calls = []
+
+    def fake_llm_chat(client, tracker, agent, messages, json_mode=False, note=""):
+        calls.append(note)
+        return "[]"
+
+    monkeypatch.setattr("agents.llm_chat", fake_llm_chat)
+    report = proofreading_agent(
+        client=object(),
+        tracker=type("T", (), {"record": lambda *a, **k: None})(),
+        translations={"ch1": "hello"},
+        glossary=[],
+    )
+    assert report == {}
+    assert _report_issues(report) == []
+    assert report.get("chapters_need_revision", []) == []
+    assert calls == ["一致性审校"]
+
+
+def test_proofreading_agent_valid_object(monkeypatch):
+    payload = {
+        "issues": [{"chapter": "ch1", "detail": "x"}],
+        "chapters_need_revision": ["ch1"],
+        "summary": "ok",
+    }
+
+    def fake_llm_chat(client, tracker, agent, messages, json_mode=False, note=""):
+        import json
+        return json.dumps(payload)
+
+    monkeypatch.setattr("agents.llm_chat", fake_llm_chat)
+    report = proofreading_agent(
+        client=object(),
+        tracker=type("T", (), {"record": lambda *a, **k: None})(),
+        translations={"ch1": "hello"},
+        glossary=[],
+    )
+    assert report == payload