Adversarial Review
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
- 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>.md beside 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
1---2name: adversarial-review3description: Use when reviewing the survey, a contribution, or a paper draft as a severe but fair reviewer — claim audit, methodology critique, and venue-reviewer simulation before submission.4license: MIT5---67# Adversarial Review89Find the objections that would block acceptance before external reviewers10do. Review; never fix. Findings go to a report beside the artifact; the11artifact itself is never edited by this skill.1213## Read First1415- the artifact under review: `survey/survey.tex` + PDF, or a contribution16 README + report, or a manuscript + `framing.md` + `venue.md`17- `references/review-lanes.md`, `references/claim-audit.md`1819## Procedure20211. Identify the target and the standard it must meet:22 - survey → coverage vs `sota/index.md`, grouping logic, fairness, and23 the quality of the gaps section24 - contribution → badge-checklist truthfulness, methodology (the25 develop-contribution methodology standard), claim–evidence match26 - paper → venue fit and reviewer simulation per `venue.md`, novelty,27 soundness, clarity282. Read claim-first: contributions/claims, then the evidence offered.293. Run the claim audit on every acceptance-critical claim per30 `references/claim-audit.md`: verify each against the cited synthesis,31 contribution output, or PDF.324. Review through every lane in `references/review-lanes.md`; keep lane33 findings separate until the final synthesis.345. Write the report to `reviews/<YYYY-MM-DD>-<scope>.md` beside the35 artifact (`survey/reviews/`, `contributions/<slug>/reviews/`,36 `papers/<slug>/reviews/` — create the folder on demand) with: executive37 summary and recommendation; strengths with evidence; weaknesses grouped38 by severity; the claim-audit table; questions the authors must answer;39 a prioritized revision roadmap.4041## Criticism Standard4243Every major criticism carries: exact claim or location, why it matters,44evidence, what would fix or weaken it, severity (fatal / major / moderate /45minor), and confidence. No performative harshness; no objection without46evidence; polish never excuses methodological weakness.4748## Done When4950- The report exists with all sections, every fatal/major finding is51 evidence-backed, and no lane finding was dropped during synthesis — if a52 finding is wrong, the report says why
VincenzoImp