fix(space): artifact shape reader polish — evolution evidence + list_artifacts kind filter

[lsm]

Two reader fixes deferred from #2313 (the generic artifact-shapes PR), now that it has merged to dev.

extractArtifactDetail (evolution evidence) read only legacy fields, so fresh shape writes never surfaced — it now checks the canonical shape fields url/text/recommendation first (link/note/decision), with legacy fields as fallback so migrated rows still resolve.

list_artifacts mapped legacy type:'pr'→link and type:'review'→decision but didn't filter on data.kind, so {type:'pr'} over-returned issue/preview/doc links and {type:'review'} over-returned gate decisions. It now post-filters those two by their legacy semantic kind; result (overloaded), progress, and shape-name queries stay unfiltered.

Covered by new unit tests in both files; the list_artifacts tests fail without the kind filter (verified).

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[greptile-apps]

Greptile Summary

This PR delivers two targeted reader fixes deferred from the generic artifact-shapes PR (#2313): extractArtifactDetail now checks canonical shape fields (url/text/recommendation) before legacy fields, so fresh shape writes surface in evolution evidence; and list_artifacts adds a post-filter on data.kind when the legacy type:'pr' or type:'review' is used, eliminating the over-return of unrelated link/decision kinds.

• extractArtifactDetail (evolution-scope-service.ts): exported for testability; prepends url, text, recommendation to the key scan order, with all original legacy keys retained as fallback — migrated rows stay resolved.
• list_artifacts (node-agent-tools.ts): kindFilter is only set for the two overloaded legacy types (pr/review); direct shape-name queries and the result/progress aliases are intentionally left unfiltered.
• New unit tests cover both changes with realistic mixed-kind seed data; the kind-less-row edge case (pre-migration rows excluded by the filter) is explicitly documented and tested.

Confidence Score: 5/5

Safe to merge — both fixes are narrowly scoped reader-only changes with no writes or schema mutations, and the migration-boundary edge case is explicitly documented and covered by a dedicated test.

Both changes are read-path only: extractArtifactDetail adds three key lookups to an existing ordered scan, and list_artifacts post-filters on an already-loaded in-memory slice. Neither mutates stored data nor changes the write path. The new test suite covers happy paths, priority ordering, legacy fallback, the overloaded result alias, and the intentional migration-gap narrowing.

Files Needing Attention: No files require special attention.

Important Files Changed

Filename	Overview
packages/daemon/src/lib/space/evolution-scope-service.ts	Prepends canonical shape fields to extractArtifactDetail's key scan; function exported for unit testing. Logic and field ordering are correct.
packages/daemon/src/lib/space/tools/node-agent-tools.ts	Adds kindFilter post-filter for pr/review legacy type queries; narrow, correct, and well-commented.
packages/daemon/tests/unit/5-space/agent/node-agent-tools.test.ts	Adds comprehensive list_artifacts test suite: mixed-kind seeding, per-legacy-type assertions, kind-less migration boundary case explicitly documented and tested.
packages/daemon/tests/unit/5-space/evolution-scope-service.test.ts	Adds extractArtifactDetail unit tests covering new canonical fields, priority ordering, legacy fallback, and null/empty guard cases.

_{Reviews (2): Last reviewed commit: "test(space): cover kind-less link/decisi..." | Re-trigger Greptile}

[lsm]

🤖 Review by glm-5.1 (z.ai)

Model: glm-5.1 | Client: NeoKai | Provider: z.ai

Recommendation: REQUEST_CHANGES — one P2 (test gap) + 2 open Greptile threads to resolve before merge.

Both fixes are correct, well-scoped, and well-tested. Verified independently.

extractArtifactDetail — canonical fields (url/text/recommendation) now probed before legacy fields, so fresh link/note/decision writes surface in evolution evidence. Migrated rows still resolve (normalizeLinkData carries url; result→decision keeps summary). The one behavior change — a decision carrying both recommendation and summary now returns the verdict — is intentional and correct (the verdict is the primary signal). Clean. New export passes knip (used by the test).

list_artifacts kind filter — correct and tight: kindFilter applies only to pr/review; result, progress, direct shape-name queries, and no-type stay unfiltered. I confirmed the new tests genuinely catch the bug by running them against the pre-fix source (origin/dev): without the filter {type:'pr'} returns 3 links (not 1) and {type:'review'} returns 2 decisions (not 1). Restored after; tree clean, tests green (213 pass), oxlint/tsc/knip all clean.

Greptile P1 (kind filter "silently drops" kind-less rows) — reviewed and dismissed as unfounded. The backfill migration at migrations.ts:11290-11299 sets kind:'pr' for pr→link, kind:'review' for review→decision, and infers kind from URL fields for result→link. So every semantically-PR/review row carries the right kind; the only kind-less link/decision rows are legitimately-not-PR/not-review direct shape writes, which should be excluded. No silent regression (Greptile itself rates this 4/5 "safe to merge with awareness").

P2 — add a kindless-link exclusion test (endorsing Greptile's second thread; anchored comment posted). The suite only covers the kind-always-written path. Add a test asserting a link written without kind is excluded from { type:'pr' } (and the kind-less decision analog for { type:'review' }). Documents the intentional trade-off at the migration boundary and guards against a future change that wrongly includes kindless rows.

Before merge: resolve the 2 open Greptile threads — dismiss P1 with the migration-backfill note above; P2 is addressed by the new test.

Note: the local bun run check check:test-quality blip in provider-registry.test.ts is pre-existing on origin/dev, outside this diff, and not in PR CI — not a concern for this PR.

[lsm]

🤖 Review by glm-5.1 (z.ai)

Model: glm-5.1 | Client: NeoKai | Provider: z.ai

Recommendation: APPROVE. Zero findings remain — P2 addressed, all threads resolved.

The kindless-row exclusion test (commit 3a855bc) is correct and is a genuine guard: it writes a kind-less link and decision, asserts {type:'pr'}/{type:'review'} exclude them while direct {type:'link'}/{type:'decision'} still return them. Without the kind filter the {type:'pr'} assertion would fail (the 1 kindless link would be returned), so this locks the behavior in. Passes locally (179 tests).

Verified on the new head: 0 unresolved review threads; CI green (Lint/Knip/Typecheck, Web, Greptile, all unit shards, and most online shards pass — rpc-4 and space-2 still in-flight, neither touched by this test-only change). PR is OPEN/MERGEABLE, targets dev.

Both core fixes from the prior round stand correct and well-tested: extractArtifactDetail now probes canonical shape fields first, and list_artifacts post-filters pr/review by data.kind. Ready to squash-merge (do not delete branch).