feat(enrich): thin-page enrichment proposes expansions for stub pages

[claytonlin1110]

summary

implements #309: a maintenance pass that finds thin pages (stub title + one-line body, few or no claims/sources) and drafts an enriched revision synthesized strictly from related approved material already in the kb.

• kb.enrich_page (+ mcp/jsonl/cli registration across all four surfaces, src/vouch/enrich.py) scores one page against enrichment.min_body_chars / enrichment.min_citations (config.yaml, sane defaults, overridable per-invocation). a page is thin when len(page.body) is under the char threshold or len(page.claims) + len(page.sources) is under the citation threshold.
• candidate material is gathered by reusing synthesize.synthesize (feat: kb.synthesize — answer-mode retrieval over the review-gated KB #222) against the page's title+body as query, rather than re-deriving clauses from context.build_context_pack directly — this keeps the "every sentence traces to an approved claim" invariant identical between the two features and only pulls approved, non-retracted claims.
• the synthesized addition is appended to the page's existing body (a stub's hand-written text is never discarded or overwritten), and the cited claim ids / their evidence source ids are unioned into the page's claims / sources.
• the enriched page is filed via propose_page with slug_hint set to the existing page id, landing as a pending proposal — nothing is written durably without a human kb.approve. mirrors compile.py's propose-never-approve shape, including a distinct page-enricher actor so self-approval stays blocked by the default gate.
• vouch enrich-pages batches this over kb.list_pages(), with --dry-run, --min-body-chars, --min-citations, --limit, --json; one page's rejection (e.g. a page kind's required fields tightened since last approval) is recorded as a skip rather than sinking the batch.
• docs/enrich.md + CHANGELOG entry.

test plan

• tests/test_enrich_page.py — thin-page selection (both thresholds), above-threshold / no-related-claim skips, the propose-not-write invariant (proposal then approve revises the existing page, not a duplicate), dry-run, every-sentence-has-a-resolvable-citation, config load (present/default/bad-value fallback), batch dry-run + --limit, audit event, capabilities/HANDLERS parity, jsonl wire (success + clean missing-page error)
• ruff check src tests clean
• mypy src clean (no new errors beyond the pre-existing sandbox.py win32 / missing-jinja2 gaps on this machine)
• targeted pytest (enrich, synthesize, compile, storage, capabilities) green aside from two pre-existing, unrelated failures confirmed present on a pristine upstream/test checkout too (the host_compat/openclaw-manifest tests, and test_jsonl_kb_compile_files_proposals's use of a cat stub llm_cmd, which doesn't exist as a Windows shell built-in)

🤖 Generated with Claude Code

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 7bd3a1de-4d4f-41fa-8b12-4d3f6773fc0f

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}