[zengchuang-hw]

Governing issue and delivery slice

Refs #371, step 4. Expose the Apps Python entry for an explicitly authorized Knowledge candidate store. It opens existing stores read-only by default for history. When the caller expressly enables writes, the returned Knowledge port provides create and expected-revision update. Existing Apps reader composition uses the same read-only entry. Knowledge retains its candidate, adapter authority, transaction and conflict rules.

The regression covers no initialization on a read-only miss, rejection of a nonboolean write flag, candidate create/update, stale-write rejection, parent lineage and reopening the original revision. The public entry is Python, with write validation and publication owned by Knowledge.

Validation

• 60 tests passed on Windows across Apps Knowledge, the store, Knowledge contracts and selection.
• The archive-protection test initially hit the default tape device in Windows tar. The same failure reproduced at the base. The full rerun used TAPE=- so tar reads the archive from stdin.
• Repository Ruff and staged diff checks passed.

[amy-why-3459]

LGTM

PR review verdict — PASS
Reviewed base c2f463f066921a15120a3c86506c4f97b1909acb and head 6b7f5d3c1f9d90124c4c1894c7f78ce0f177144a.

What I checked:

• src/inferrsi/apps/knowledge.py: open_knowledge_store is a thin wrapper over KnowledgeStore(store_root, skill_root=..., read_only=not writable), and it defaults to read-only. A non-boolean writable is rejected before the store is touched. knowledge_reader now goes through the same read-only entry, and its behavior is unchanged.
• Candidate authority, transactions, expected-revision conflicts and the read-only write guard all stay in KnowledgeStore (knowledge_store.py L44-62, L118-143). Apps adds no write logic of its own.
• tests/test_apps_knowledge.py: a read-only open of an empty directory fails with "requires an existing database" and leaves the directory empty. The test also covers create/update, stale-write RevisionConflict, parent lineage in history, rejection of writes through a read-only viewer, and reading back both revisions through knowledge_reader.
• The docs/operations.md row and section match the entry's actual read-only default and the explicit writable=True opt-in.
• All 8 checks on this head are green (full-check 3.10/3.11, installed-package 3.10/3.11, declarations x2, pr-references, dashboard-browser).

[cursor]

LGTM

PR review verdict — PASS
Reviewed base c2f463f066921a15120a3c86506c4f97b1909acb and head 6b7f5d3c1f9d90124c4c1894c7f78ce0f177144a.

Governing: Refs #371 step 4 (explicit candidate create/update and read-only history entry). Independent first-pass as cursor[bot] (steward/Dong-bot standing delegation; not author or accountable parent for this slice).

Verified:

• src/inferrsi/apps/knowledge.py: open_knowledge_store defaults writable=False → KnowledgeStore(..., read_only=True); rejects non-boolean writable before touching the store; knowledge_reader composes the same read-only entry.
• Write/CAS/history authority stays in Knowledge (KnowledgeStore.create/update/history); Apps adds no write logic.
• tests/test_apps_knowledge.py: RO miss does not initialize; non-boolean write flag rejected; create/update; stale-write RevisionConflict; parent lineage via history; RO viewer rejects writes; pinned and current revisions readable through knowledge_reader.
• Local: pytest tests/test_apps_knowledge.py → 3 passed; ruff check on changed Python → pass.
• Ready CI on this head: declarations, pr-references, full-check 3.10/3.11, installed-package 3.10/3.11, dashboard-browser — all success.
• docs/operations.md documents the entry, RO default, and explicit writable=True opt-in.

Non-blockers: PR is one commit behind current main (36440109, #425) with a clean auto-merge of docs/operations.md; Amy already APPROVED this same head (5909004107). Not merged by this review.