Find the objections that would block acceptance before external reviewers do. Review; never fix. Findings go to a report beside the artifact; the artifact itself is never edited by this skill.
Read First
- the artifact under review:
survey/survey.tex+ PDF, or a contribution README + report, or a manuscript +framing.md+venue.md references/review-lanes.md,references/claim-audit.md
Procedure
- Identify the target and the standard it must meet:
- survey → coverage vs
sota/index.md, grouping logic, fairness, and the quality of the gaps section - contribution → badge-checklist truthfulness, methodology (the develop-contribution methodology standard), claim–evidence match
- paper → venue fit and reviewer simulation per
venue.md, novelty, soundness, clarity
- survey → coverage vs
- Read claim-first: contributions/claims, then the evidence offered.
- Run the claim audit on every acceptance-critical claim per
references/claim-audit.md: verify each against the cited synthesis, contribution output, or PDF. - Review through every lane in
references/review-lanes.md; keep lane findings separate until the final synthesis. - Write the report to
reviews/<YYYY-MM-DD>-<scope>.mdbeside the artifact (survey/reviews/,contributions/<slug>/reviews/,papers/<slug>/reviews/— create the folder on demand) with: executive summary and recommendation; strengths with evidence; weaknesses grouped by severity; the claim-audit table; questions the authors must answer; a prioritized revision roadmap.
Criticism Standard
Every major criticism carries: exact claim or location, why it matters, evidence, what would fix or weaken it, severity (fatal / major / moderate / minor), and confidence. No performative harshness; no objection without evidence; polish never excuses methodological weakness.
Done When
- The report exists with all sections, every fatal/major finding is evidence-backed, and no lane finding was dropped during synthesis — if a finding is wrong, the report says why