chore: purge references/ research tree; scrub links - #1
Conversation
…js only Co-authored-by: Cursor <cursoragent@cursor.com>
SDK code review (trainer → form-check + review-rigor)Verdict:
Summary
Findings (ranked)F1 · P2 · COR · conf 88% · S1✓ S2✓ S3✓ S4✓ S5✓ S6✓ S7✓Canonical rubric doc is thinner than the deleted markdown
Downstream consequence: Agents or humans following INDEX/SKILL pointers expecting operational rubric detail will find only high-level bullets. Weight percentages still appear in Falsifier: Diff the deleted Mitigation: Acceptable for an intentional dedup if maintainers treat the JS implementation as ground truth. Optional follow-up: expand the inline comment block to include penalty tables (no behavior change). F2 · P3 · COR · conf 82% · S1✓ S2✓ S3✓ S4△ S5✓ S6✓ S7✓Pre-existing comment-vs-implementation drift now labeled canonical The rubric comment at Downstream consequence: Declaring the comment block "canonical source" ( Falsifier: Grep Note: This drift predates the purge; the diff only adds the canonical-source header line. Not a merge blocker for a deletion PR, but worth a comment cleanup in a follow-up. F3 · P3 · ARC · conf 92% · S1✓ S2✓ S3✓ S4✓ S5✓ S6✓ S7✓Internal research/audit trail removed from working tree Seven files under Downstream consequence: Future contributors lose in-repo provenance for pattern decisions unless they consult git history ( Falsifier: Run the verify command; confirm Rollback path: F4 · P4 · TST · conf 90% · S1✓ S2✓ S3✓ S4✓ S5✓ S6✓ S7✓Verify command matches change class; optional breadth gap Supplied verify: test ! -d references && \
! rg -n 'references/quality-scorecard|references/RESEARCH_INDEX|references/ADVERSARIAL' \
--glob '!localonly/**' --glob '!CHANGELOG.md' . && \
node --check tools/score-tests.jsAppropriateness: Yes for a docs-only purge—confirms directory removal, guards against the three highest-risk stale path strings, and syntax-checks the declared canonical tool. Ran locally: exit 0 ( Optional enhancement: A broader Merge recommendationMerge. The branch delivers a focused, intentional purge of duplicate internal reference material with consistent doc retargeting and a passing verify gate. Residual items (F1 rubric doc depth, F2 comment accuracy) are documentation quality nits, not corruption or scope violations. No COR finding reaches ≥90% confidence on active breakage. Optional post-merge follow-ups (non-blocking):
Posted by |
|
Pausing merge to implement review recs then proceeding with delete |
Paraphrase deleted quality-scorecard into tools/score-tests.js comment block (penalty tables, thresholds, limitations). Fix README penalty claims (F2). Split priority vs category tag checks; add .toBeFalsy() penalty. Co-authored-by: Cursor <cursoragent@cursor.com>
|
Codereview repass (after rubric-align commit SDK code review (trainer → form-check + review-rigor)Verdict:
Summary
Findings (ranked)F1 · P2 · COR · conf 92% · S1✓ S2✓ S3✓ S4✓ S5✓ S6✓ S7✓ — RESOLVED (was open on dde1dbd)Canonical rubric depth restored in
Downstream consequence: Agents following INDEX/SKILL get operational rubric without git archaeology. Falsifier: Diff Status: Fixed in F2 · P3 · COR · conf 90% · S1✓ S2✓ S3✓ S4✓ S5✓ S6✓ S7✓ — RESOLVED (was open on dde1dbd)Comment claims aligned with Removed false claims (fixtures, describe/group, test.step scoring). README no longer states the scorer “penalizes getByText over getByRole.” Implementation grep: no Downstream consequence: Declaring the comment block canonical no longer misleads auditors. Falsifier: Status: Fixed in F3 · P3 · COR · conf 78% · S1✓ S2✓ S3△ S4✓ S5✓ S6✓ S7✓README intro still over-claims anti-pattern coverage
Downstream consequence: Skimmers of paragraph 1 may over-trust scorer coverage vs Falsifier: Compare Mitigation: Non-blocking; optional one-sentence soften in intro (“syntactic anti-patterns in the scorer; see validate-suite.sh for the rest”). F4 · P3 · TST · conf 94% · S1✓ S2✓ S3✓ S4✓ S5✓ S6✓ S7✓Scoring change vs dogfood CI Split tag checks and Downstream consequence: External suites using only one tag type may see −5 vs prior single combined −5; northwind-qa unaffected in CI. Falsifier: F5 · P3 · ARC · conf 95% · S1✓ S2✓ S3✓ S4✓ S5✓ S6✓ S7✓References tree removal (unchanged from first pass) Seven files under Rollback: Merge recommendationMerge. Repass confirms prior APPROVE holds after SDK_CODEREVIEW_VERDICT=APPROVE Posted by |
README opening no longer implies every anti-patterns.md row has a score-tests.js penalty; points to validate-suite.sh for the rest. Co-authored-by: Cursor <cursoragent@cursor.com>
1 participant
Summary
Automated SDK weekend worker +
_sdk_verify_and_pr.shverification.SDK code review (trainer → form-check + review-rigor)
Verdict:
APPROVEQueue:
playwrighter-purge-referencesBranch:
chore/purge-references-sdk· Repo:weijia-89/playwrighterlocalonly/sdk-reviews/playwrighter-purge-references-chore-purge-references-sdk.mdcursor-sdk-playground/prompts/reviews/playwrighter-purge-references-chore-purge-references-sdk.mdSummary
references/tree (7 markdown files: research indexes, adversarial reviews, duplicate quality-scorecard) and retargets doc pointers totools/score-tests.js.README.md,INDEX.md, and.claude/skills/playwrighter/SKILL.mdall updated; no stalereferences/path references remain outside intentionalCHANGELOG.mdprose.tools/score-tests.jsscoring logic is untouched; only a header comment declares the inline rubric block as canonical.references/absent, targetedrgclean,node --check tools/score-tests.jsOK).CHANGELOG.mddocuments the removal under[Unreleased].quality-scorecard.md(penalty tables, multi-tier thresholds, limitations section lost).Full ranked findings are in the PR review comment posted by the SDK hook.
Posted by
_sdk_verify_and_pr.shafter_sdk_trainer_codereview.sh. Merge still requires CI green and human review;BLOCK/REQUEST_CHANGES(withSDK_CODEREVIEW_STRICT=1) stop the hook before push.Test plan
test ! -d references && ! rg -n 'references/quality-scorecard|references/RESEARCH_INDEX|references/ADVERSARIAL' --glob '!localonly/**' --glob '!CHANGELOG.md' . && node --check tools/score-tests.jsSDK_CODEREVIEW_SKIP=1)