CommunityWriting & Editinggithub.com

manuscript-audit

Audit scientific manuscripts for argument structure, evidence, numerical and statistical consistency, terminology, figures and package verification. Use for explicit audits and full or substantive manuscript-revision diagnostics; bounded prose edits use targeted shared checks rather than a full audit.

What is manuscript-audit?

manuscript-audit is a Claude Code agent skill that audit scientific manuscripts for argument structure, evidence, numerical and statistical consistency, terminology, figures and package verification. Use for explicit audits and full or substantive manuscript-revision diagnostics; bounded prose edits use targeted shared checks rather than a full audit.

Works with~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/zachary-zzc/manuscript-writing-skills/tree/HEAD/skills/manuscript-audit

Installed? Explore more Writing & Editing skills: steipete/notion, langchain-ai/langchain, bytedance/podcast-generation · View all 6 →

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Manuscript Audit

Read the shared manuscript contract directly and reuse the established mode, source version, accepted decisions and protected content. An audit-only request remains read-only; it does not authorize manuscript edits, raw-record changes, new experiments or external submission.

Hard rules for manuscript narrative

Apply the Editorial Guideline (mandatory default) in release narrative and defensive-writing rules before drafting, restructuring, polishing, auditing, or reviewing. They are hard defaults, with special priority for manuscript writing, storyline, idea framing, and innovation claims. When the user supplies manuscript text, perform the required point-by-point defensive-writing audit and give a concrete rewrite for every instance. Preserve material limitations, negative findings, uncertainty, ethics, safety, and reproducibility boundaries as precise evidence or scope statements.

For Nature or Nature Portfolio targets, follow the canonical Nature integration map and load only the relevant nature-shared reference; this manuscript skill remains the owner.

Read audit.md for the checks warranted by the scope. Before a full audit or substantive whole-manuscript revision, complete its whole-manuscript diagnostic and coverage ledger. For a new or full manuscript, run the multi-pass audit after H1–H4, after each H7 figure-linked Results unit, after H9S Supplementary Information, after H8–H11 and again at H12. A bounded edit instead uses the shared paragraph review and only necessary evidence/dependency checks. Audit depth is independent of Easy/Normal/Expert supervision.

For statistical analyses and inference in scope, read statistical-reporting.md. For availability statements, data-sharing plans or submission packages whose data/code access is in scope, read data-availability.md. Load only the matching reference; neither introduces new experiments, deposits or review checkpoints.

When a specialised mechanism, measurement proxy, spatial validation or deployment claim is central to the audit, consult the matching examples in domain-evidence.md. Do not apply unrelated domain checklists to the manuscript.

Keep one findings/coverage record. Verify both the original claim and proposed correction against authoritative evidence. Classify confirmed errors, ambiguities, provenance gaps and unassessed material accurately; preserve originals and prior resolutions. Apply numerical, statistical, source, cross-document and visual checks only where their subjects are present.

Report evidence, numerical reproduction, each of the four visual layers, compilation, inspected layout and package completeness separately. Use independent mock peer review only when requested or warranted by scope, with isolated contexts and frozen reports; do not automatically require three reviewers. Finish with concrete repairs or unresolved dependencies, following the selected mode for consequential decisions and final review.

Individual skills in this repo

This repo contains 4 individual skills — each has its own dedicated page.

Related Skills