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 +1,34 @@
|
||||
# Fifth independent review: realistic tool-call placement
|
||||
|
||||
The second terminal acceptance review rejected the corrected candidate after
|
||||
running all requested checks. Continue the same self-update and address these
|
||||
exact findings:
|
||||
|
||||
1. In the normal tool loop, the newest durable message is commonly an assistant
|
||||
tool-call message whose `content` is `None` or empty, followed by a tool
|
||||
result. The current backward search selects that assistant message, then
|
||||
fails closed because its content is unsupported. The model-visible status is
|
||||
therefore absent during common tool-loop requests.
|
||||
2. The current test uses assistant tool-call messages with ordinary string
|
||||
content, so it does not cover the production shape above.
|
||||
3. If the assistant tool-call message does have text, the current code appends
|
||||
status before the following tool result instead of at the request context
|
||||
end. This violates the book/report placement requirement that current
|
||||
model-visible state be closest to generation.
|
||||
|
||||
Implement a durable solution at the actual string-only persistence/replay
|
||||
boundary. It may extend string `api_content` sidecar replay to tool messages if
|
||||
that is supported by the state and flush contracts, but must not widen the
|
||||
database type, mutate clean transcript content, disturb role/tool-call ordering,
|
||||
or attach status to unsupported list/multimodal content. Inspect and test the
|
||||
real production contract rather than relying on a permissive test helper.
|
||||
|
||||
Add coverage for an assistant tool call with `content=None` followed by a
|
||||
string tool result, across at least three successive requests. Assert both
|
||||
byte-identical historical replay and placement after the newest tool evidence.
|
||||
Retain the adversarial output-bound test.
|
||||
|
||||
Run the focused tests, existing sidecar/cache/turn-context regressions,
|
||||
compilation, and `git diff --check`. Update `BOOK_SELF_EVOLUTION_REPORT.md` with
|
||||
this fifth review and exact results. Do not edit the book, commit, push, or
|
||||
claim downstream task improvement.
|
||||
Reference in New Issue
Block a user