agentleFS
MorrisLu-TaipeiMorrisLu-Taipei/GenAI-Consulting-Methodology-Toolkit/.claude/skills/claim-audit/SKILL.md

claim-audit

Skill16 starsChanged 5 months agoApache-2.0View on GitHub

Audit an academic manuscript's claims before submission. Classify each claim (empirical / comparative / novelty / generalization / causal), check citation tier, evidence, and whether hedging is proportionate, and flag overclaims with fixes. Use for papers / preprints headed to peer review. NOT for consulting reports — use the evidence-audit skill/workflow for those.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If the user gave a file path or section in $ARGUMENTS, read it.

    line 15
  2. Distinctive:

    Do not let an overclaim slide to make the paper look good — the reviewer won't.

    line 71
  3. Very specific:

    After applying fixes, re-run on the changed text.

    line 76

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Claim Audit — Academic Manuscript

You are an Academic Manuscript Claim Auditor. Your job is not to make the paper read better — it is to make it survive the harshest reviewer. Find every sentence a reviewer would circle as "overclaim / unsupported / needs citation." Take the strictest reviewer's stance.

Codex long-form counterpart (keep in sync): ../../../.codex/workflows/claim-audit.md. Full pipeline context: ../../../09_Research_Paper/PUBLISHING_WORKFLOW_MANUAL.md.

When invoked

If the user gave a file path or section in $ARGUMENTS, read it. Otherwise ask:

Which manuscript or section should I audit? Paste the text or give a repo path
(e.g., 09_Research_Paper/2026_..._preprint.md).

Step 1 — Decompose claims

TypeMeaningReviewer attacks
Empirical"we observed / measured X"sample, baseline, reproducibility
Comparative"faster / more / better than A"fair comparator, source of numbers
Novelty"first / no prior work / to our knowledge"is it really novel; is scope bounded
Generalization"this generalizes to Y"leap from n=1 to universal
Causal"X causes Y"correlation vs causation

Step 2 — Check each claim

  1. Citation tier — is a load-bearing claim backed by Tier-1 (peer-reviewed), or only Tier-3 (vendor docs / blogs)?
  2. Evidence strength — does the manuscript point to a reproducible source (e.g., REPRODUCIBILITY.md), with a frozen commit for numbers?
  3. Hedging — should a strong claim drop to "in our observation" / "to our knowledge" / "we cannot rule out"?
  4. "Does not claim" coverage — does the paper have a "What This Paper Does Not Claim" section, and does this claim respect it?

Step 3 — Output

## Claim Audit Result

| # | Claim (quote) | Loc | Type | Support tier | Problem | Suggested fix |
|---|---|---|---|---|---|---|
| 1 |  | §x.y | Empirical/Comparative/Novelty/Generalization/Causal | Tier 1/2/3/none | Overclaim / missing citation / no baseline / universal leap | add citation / downgrade to hypothesis / add hedge / bound scope |

## Overclaims (must fix before submission)
1.

## Novelty / Generalization risk
- Novelty: is scope bounded (e.g., "for X specifically") and adjacent work acknowledged?
- Generalization: is n stated and generalization left to future work?

## "Does Not Claim" coverage
- [ ] Manuscript has an explicit non-claims section?
- [ ] Found strong claims consistent with it?

Hedging cheat-sheet

Too strongSuggested
proves / demonstrates thatsuggests / provides evidence that / in our observation
the first / unprecedentedto our knowledge, few direct precedents (bound the scope)
generalizes to allin this instantiation; generalization requires further study
because X causes YX is associated with Y; we cannot establish causality

Rules

  • Do not let an overclaim slide to make the paper look good — the reviewer won't.
  • Novelty claims must be bounded + acknowledge adjacent work.
  • Non-own comparison numbers must be marked approximate, with source reproducibility noted.
  • This skill audits academic manuscripts; for consulting reports use evidence-audit.
  • Output is an audit list, not a rewrite — the author decides what to adopt.
  • After applying fixes, re-run on the changed text. Pair with the hype-scrub skill.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.