Files
ai-agent-book/chapter9/hermes-self-evolution/review_task_3.md
T
liqiang b119135836
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
ai-agent-book 精选快照(<2MB 代码与文档,来自 github.com/bojieli/ai-agent-book)
2026-08-20 13:12:50 +00:00

1.3 KiB
Raw Blame History

Third independent review: persistence boundary

The second correction still overclaims multimodal support. Inspection of the actual persistence boundary found:

  • hermes_state.py types api_content as Optional[str] and writes non-string values as None (around lines 56325750 and 60786106);
  • run_agent.py discards non-string _row_api_content during flush (around lines 21042110);
  • agent/turn_context.py exposes string-only sidecar helpers.

Thus a list sidecar is replayed within the in-memory loop but disappears after persistence/resume, invalidating the claimed durable byte stability. Make the smallest compatible correction: fail closed for non-string content and do not attach the model status on unsupported multimodal/list messages. Do not widen the database schema or persistence contract in this experiment.

Update tests to assert list, empty, mapping, and numeric values are rejected by the same production helper, and that string sidecars remain stable across three requests. Remove any multimodal-support claim from code comments and the report. Run the focused tests, the existing replay/cache regression set, compilation, and git diff --check; update the report with exact results.

Do not edit the book, commit, push, or claim downstream task improvement.