manuscript-review-council — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited manuscript-review-council (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Run a journal-style review council instead of a single blended opinion. The skill is self-contained and works in Codex and Claude Code.
The council is defined in local files:
/scientific-writing; start a council only when the user also wants scientific critique or a resolution assessment./pdf-to-md), title, abstract, main claims, methods snapshot, figure and table list, user priorities, and prior reviews or author response for a revision. Every reviewer gets the same packet; keep section boundaries, figure references, and the claims under test.domain: novelty, significance, positioning against prior work, and overclaimingmethods-statistics: design, controls, benchmarks, sample size, figure interpretation, and analysis validityskeptic: weakest links, alternative explanations, unsupported causal claims, results explained by unchecked causes, and missing controlsreproducibility: computational papers, code or data availability, workflow clarity, and parameter transparencyethics-compliance: human subjects, animal work, privacy, conflicts, dual use, or citation biastranslational: clinical, ecological, or deployment claims that need a reality checktemplates/reviewer-brief.md: summary, strengths, major concerns, minor concerns, must-fix analyses, confidence (0 to 1), and a provisional recommendation (accept, minor_revision, major_revision, or reject), each concern anchored to a section, figure, table, or explicitly missing information. Reviewers do not invent citations, datasets, reviewer expectations, or experiments the manuscript does not motivate.reviews/<manuscript-slug>/<YYYYMMDDTHHMMSSZ>/ as described in references/council-workflow.md, and validate the bundle index: uv run --script "$HOME/.agents/skills/manuscript-review-council/scripts/validate_review_bundle.py" bundle.jsonThe validator rejects non-deterministic paths, duplicate roles or issue IDs, an issue whose source_roles names a reviewer absent from reviewer_reports, and a council without all three default reviewers unless reduced_council_reason explains why.
templates/editor-meta-review.md: recommendation, rationale across novelty, rigor, evidence strength, clarity, reproducibility, and significance, ranked major and minor revisions, questions for the authors, and a short decision letter./polars-dovmed for claim-specific literature and /bio-logic for evidence strength and causal claims. Never fabricate supporting papers./scientific-writing after the council has fixed the factual review content.| Task | Action |
|---|---|
| Full manuscript review | Run the 3-reviewer council plus editor synthesis |
| Computational manuscript | Add a reproducibility reviewer |
| Human, animal, or clinical manuscript | Add an ethics or compliance reviewer |
| Revision assessment | Compare prior critiques to the new draft and label each issue resolved, partial, or unresolved |
| Fast triage | Use domain plus skeptic, set reduced_council_reason, then write a short editor recommendation |
| Rebuttal check | Judge whether the author response closes the decision-driving issues |
reviews/<manuscript-slug>/<run-id>/reviews/<manuscript-slug>/<UTC-run-id>/... and the bundle passes validate_review_bundle.py/scientific-writingReview this microbiome manuscript with a multi-agent council. Use domain,
methods/statistics, skeptic, and reproducibility reviewers. End with an editor
meta-review, a recommendation, and ranked major/minor revisions.Compare this revised manuscript against the prior decision letter. Tell me
which major issues are resolved, partially resolved, or still open, then write
an editor recommendation.Give me a fast desk-review style assessment of this preprint. Use only a domain
reviewer and a skeptic, record the reduced council on the bundle, then summarize
whether it is promising, immature, or fatally flawed.Issue: The manuscript is too long for every reviewer to read in full. Solution: Build a shared review packet first, then route only the relevant sections, claims, and figures to each reviewer while keeping a common abstract and methods snapshot.
Issue: Reviewers disagree sharply. Solution: Make the disagreement explicit, identify the evidence each reviewer is using, and let the editor resolve the conflict instead of averaging positions.
Issue: No journal rubric or venue is provided. Solution: State the default criteria being applied: novelty, rigor, evidence strength, clarity, reproducibility, and significance.
Issue: Only one agent can run. Solution: Emulate the council sequentially with role-scoped passes and keep the notes separated until the editor synthesis step.
/scientific-writing: apply the reviewer feedback to the manuscript/bio-logic: deeper reasoning about evidence strength/ai-scientist-evaluator: equivalent review for AI scientist outputs~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.