[chuiwenwei]

Summary

The reverse case→skill bridge keyed on Candidate.id, the index row id
(<owner>_ac_<date>_<seq>). SKILL.md's source_case_ids stores the
md-side entry id (ac_<date>_<seq>), so the two never matched and the
bridge could not return a skill.

Carry the entry id from the case shaper in a private DTO attribute and
build the internal bridge candidates on it. The public response keeps
the index row id; only the internal bridge key changes.

Closes #478.

Area

• Architecture method
• Benchmark
• Use case
• Documentation
• Developer experience
• CI, build, or release

Verification

make lint        passed
make test        2702 passed, 4 skipped
make integration 186 passed, 7 skipped, 7 deselected

Additional tests:

• tests/unit/test_memory/test_search/test_shaper.py — the shaper carries
  the md-side entry_id into the private DTO attribute; the public id
  remains the index row id.
• tests/unit/test_memory/test_search/test_manager.py — the bridge
  receives ac_... (not <owner>_ac_...) and max-pools the matched case
  score; the existing end-to-end bridge test now uses the real ID shape.

Checklist

• I kept the change scoped to the relevant area.
• I am opening this from a separate branch, not pushing directly to main.
• I updated docs, examples, or setup notes when behavior changed.
• I added or updated tests when the change affects behavior.
• I did not commit secrets, .env files, dependency folders, or generated output.
• Active relative links in Markdown files resolve.

Notes for Reviewers

• No docs change: the reverse bridge is internal, and the public response
  schema is unchanged.
• SearchAgentCaseItem._entry_id is a Pydantic PrivateAttr: it is never
  serialized, so the public schema and response are unchanged.
• Cases without an entry id are skipped from the bridge rather than
  falling back to the row id (which can never match source_case_ids).
• The alternative is to store index row ids in source_case_ids; I chose
  the internal entry-id path because it does not rewrite existing skills.

By submitting this pull request, I agree that my contribution is licensed under
the Apache License 2.0.