manuscript-review-council

name	manuscript-review-council
description	Run a multi-agent manuscript critique with specialist reports, disagreement checks, and editor synthesis. Use for scientific review, revision assessment, or judging whether an author response resolves prior concerns.

Manuscript Review Council

Run a journal-style review council instead of a single blended opinion. This skill is self-contained and works in both Codex and Claude Code.

The council definition lives in local files:

• role roster: references/reviewer-roles.md
• workflow and artifact bundle: references/council-workflow.md
• per-reviewer handoff template: templates/reviewer-brief.md
• editor synthesis template: templates/editor-meta-review.md

Instructions

1. Confirm the review context: manuscript or revision stage, target venue if known, desired decision scale, and whether the user wants a full review, triage, version comparison, or rebuttal assessment.
2. Gather the review packet before delegating:
   • manuscript text or accessible PDF/DOCX
   • title, abstract, main claims, methods snapshot, and figure or table list
   • user priorities such as novelty, rigor, statistics, reproducibility, or journal fit
   • prior reviews or author response if this is a revision
3. Normalize the manuscript into a shared packet for every reviewer. At minimum, preserve section boundaries, figure references, and the claims being evaluated. If the source is a PDF or DOCX, extract a sectioned text view before review. Use references/council-workflow.md for the packet contents.
4. Load the local council definition before delegating:
   • reviewer roles and activation rules: references/reviewer-roles.md
   • stage flow and artifact bundle: references/council-workflow.md
   • reviewer prompt skeleton: templates/reviewer-brief.md
   • editor synthesis skeleton: templates/editor-meta-review.md
5. Use the platform's native delegation primitive when available:
   • Codex: spawn specialist sub-agents or workers
   • Claude Code: spawn Task agents or equivalent delegated runs
   • If delegation is unavailable, emulate the same reviewer roles sequentially and keep outputs role-separated
6. Launch these three default reviewers in parallel:
   • Domain reviewer: novelty, significance, positioning against prior work, and overclaiming
   • Methods and statistics reviewer: design, controls, benchmarks, sample size, figure interpretation, and analysis validity
   • Skeptical reviewer: weakest links, alternate explanations, unsupported causal claims, and missing controls
7. Add support reviewers only when triggered by the manuscript:
   • Reproducibility reviewer for computational papers, code or data availability, workflow clarity, and parameter transparency
   • Ethics or compliance reviewer for human subjects, animal work, privacy, conflicts, dual-use, or citation bias concerns
   • Translational reviewer when the manuscript makes clinical, ecological, or deployment claims that need reality checks
8. Give every reviewer the same structured contract:
   • one-paragraph summary
   • top strengths
   • major concerns
   • minor concerns
   • must-fix experiments or analyses
   • confidence level
   • provisional recommendation: accept, minor_revision, major_revision, or reject
   • direct grounding in manuscript sections, figures, tables, or explicit missing information
9. Require reviewers to ground criticisms in the manuscript text or in clearly missing information. Do not invent citations, datasets, reviewer expectations, or unstated experiments.
10. Run a cross-review pass after the first round:

• compare disagreements
• merge duplicate concerns
• identify the few issues that actually drive the decision
• separate fatal flaws from fixable revision items

11. Capture a lightweight artifact bundle even if the user only asked for prose:
    • shared review packet
    • reviewer reports
    • disagreement or conflict notes
    • editor meta-review
    • validate deterministic paths and machine-readable issues against schemas/review-bundle.schema.json with scripts/validate_review_bundle.py
12. Write an editor meta-review that includes:

• headline recommendation
• rationale across novelty, rigor, evidence strength, clarity, reproducibility, and significance
• ranked major revisions
• ranked minor revisions
• questions for the authors
• a short decision letter or reviewer summary

13. If outside validation is needed, do targeted spot-checks instead of broad literature review:
    • use /polars-dovmed for claim-specific literature context
    • use /bio-logic for evidence-strength and causal-claim stress tests
    • never fabricate supporting papers
14. If the user wants polished review prose, a rebuttal outline, or a cleaned-up decision letter, use /scientific-writing after the council establishes the factual review skeleton.
    • Requests to rewrite manuscript prose or draft an author rebuttal/response letter route directly to /scientific-writing; do not start a council unless the user also asks for scientific critique or resolution assessment.
15. Preserve provenance in the final deliverable:
    • keep per-reviewer notes separate from the editor synthesis
    • point to sections, figures, or tables when possible
    • clearly mark inference versus explicit manuscript statement

Quick Reference

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 reviewer plus skeptic, then write a short editor recommendation
Rebuttal check	Judge whether the author response closes the decision-driving issues
Reviewer definitions	Read references/reviewer-roles.md
Stage flow and artifacts	Read references/council-workflow.md

Input Requirements

• Manuscript text or an accessible PDF/DOCX
• Optional journal rubric, scorecard, or decision scale
• Optional prior reviews, decision letter, rebuttal, or previous manuscript version
• Optional focus areas such as novelty, methods, statistics, reproducibility, or clarity

Output

• Role-separated reviewer notes
• A short disagreement or adjudication log
• An editor meta-review with a clear recommendation
• A prioritized major and minor revision list
• Specific questions for the authors
• Optional decision letter, rebuttal assessment, or version-delta summary

Quality Gates

• At least three distinct reviewer roles are used, or the reduced scope is justified explicitly
• Reviewer outputs stay role-separated until synthesis
• Major concerns are grounded in manuscript text or explicit missing information
• Reviewer disagreements are adjudicated explicitly instead of silently averaged
• The final recommendation matches the ranked issues
• No citations, datasets, experiments, or venue rules are invented
• Reproducibility and ethics checks are added when the manuscript warrants them
• Artifacts use reviews/<manuscript-slug>/<UTC-run-id>/..., issue records validate against the bundled schema, and prose editing/rebuttal drafting remains outside the council route.

Examples

Example 1: Full journal-style review

Review 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.

Example 2: Revision comparison

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.

Example 3: Fast triage

Give me a fast desk-review style assessment of this preprint. Use only a domain
reviewer and a skeptic, then summarize whether it is promising, immature, or
fatally flawed.

Troubleshooting

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.

Related Skills

• /scientific-writing — apply the reviewer feedback to the manuscript
• /bio-logic — deeper reasoning about evidence strength
• /ai-scientist-evaluator — equivalent pipeline for AI scientist outputs