manuscript-logic-auditor
Audit a manuscript, abstract, Results/Discussion section, or figure story for logic gaps, overclaims, missing controls, selective reporting, weak novelty, and reviewer-risk points. Use for 论文逻辑审计, 投稿前
SKILL.mdskills/manuscript-logic-auditor/SKILL.md
GitHubManuscript Logic Auditor
Purpose
Review a manuscript like a strict but constructive peer reviewer. This skill focuses on scientific logic, evidence boundaries, missing controls, figure story, and reviewer risk rather than surface grammar.
Typical Inputs
- Abstracts, Results, Discussion, or full manuscript drafts.
- Figure legends or figure storyboards.
- Reviewer-risk questions before submission.
- Bioinformatics, clinical, biomedical, pharmacy, or AI-in-medicine manuscripts.
Core Rules
- Do not rewrite weak evidence into strong claims.
- Do not fabricate missing controls, validation cohorts, citations, results, or mechanisms.
- Flag selective reporting of only positive results.
- Distinguish discovery from validation and association from causation.
- Keep tissue, cohort, platform, species, and data-source contexts separate.
- Recommend final human review and expert review before submission.
Workflow
Use the audit dimensions below to review the manuscript and then return prioritized risks, claim-safe wording, and reviewer-ready next steps.
Audit Dimensions
- Central claim:
- Is the claim explicit?
- Is it supported directly by the evidence?
- Is it too broad?
- Study design:
- discovery vs validation
- retrospective vs prospective
- public data vs local cohort
- tissue/platform/species separation
- Methods and controls:
- negative controls
- sensitivity analyses
- multiple testing
- effect sizes and uncertainty
- missing covariates
- Results logic:
- all eligible datasets reported?
- counterexamples discussed?
- null findings retained?
- figures match text?
- Mechanism:
- enrichment vs mechanism
- correlation vs causation
- cell-type interpretation boundaries
- Novelty and journal fit:
- what is genuinely new?
- what is incremental?
- what tier is realistic?
Output Format
# Manuscript Logic Audit
## Bottom Line
[2-4 sentence judgment]
## Major Risks
1. [risk, why it matters, fix]
## Moderate Risks
1. [risk, why it matters, fix]
## Figure/Result Story
[figure-by-figure or section-by-section comments]
## Claim-Safe Rewording
[defensible alternatives]
## Reviewer-Ready Next Steps
[ranked action list]
Quality Checklist
- Overclaims are clearly marked.
- Missing controls are specific.
- Counterexamples and limitations are not hidden.
- Suggestions are actionable.
- The output is advisory and requires expert review.