paper-audit
Skill17 starsChanged 7 months agoMITView on GitHub
Deep-review-first audit for Chinese and English academic papers across LaTeX, Typst, and PDF formats. Use whenever the user wants reviewer-style paper critique, pre-submission readiness checks, pass/fail gate decisions, structured revision roadmaps, or re-audits of revised manuscripts. Trigger even if the user only says "review my paper", "check if this is ready to submit", "audit this PDF", "simulate peer review", "find the biggest problems in this manuscript", or "re-check whether I fixed the review issues". Do not use for direct source editing or compilation-heavy repair; route those to the format-specific writing skills instead.
4 other projects use a version of this, 1 of them edited.
Adopted by
Most recently taken up by
- kinhluan/skillstheir own version, first seen Nov 8, 2026
- wangyi0403/dl-research-workflowtheir own version, first seen Oct 13, 2026
- dongzhigang13305312738-art/paper-skillsedited, first seen Jul 21, 2026
- arnabdeypolimi/claude_code_setuptheir own version, first seen May 5, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 1
- dongzhigang13305312738-art/paper-skills★ 119% same
Wrote their own version · 3
- kinhluan/skills★ 6same name
- arnabdeypolimi/claude_code_setup★ 4same name
- wangyi0403/dl-research-workflow★ 4same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository9 other files it gives its agents.
- .github/copilot-instructions.mdCopilot instructions
- citation-managementSkill
- find-skillsSkill
- latex-paper-enSkill
- math-history-researcherSkill
- openalex-databaseSkill
- skill-creatorSkill
- web-accessSkill
- writing-clearly-and-conciselySkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
18 rules. 0 are common, 13 are distinctive, and 5 are very specific to this project.
0
Common
Most of the 4 edited versions we read keep these.
13
Distinctive
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 17
Use it for audit and review.
- Distinctive: line 43
Never rewrite the paper source unless the user explicitly switches to an editing skill.
- Distinctive: line 44
Never fabricate references, baselines, or reviewer evidence.
- Very specific: line 45
Always distinguish [Script] from [LLM] findings.
- Distinctive: line 46
Always anchor reviewer findings to a quote, section, or exact textual location.
- Distinctive: line 67
Read these references before running reviewer-style work:
- Very specific: line 105
If the user clearly wants reviewer-depth critique after the quick screen, escalate to deep-review.
- Distinctive: line 109
Use this as the default reviewer-style path.
- Distinctive: line 137
Treat this as Phase 0 only.
- Very specific: line 163
If subagents are unavailable, use the built-in deterministic fallback lane pass in scripts/audit.py so the workflow still writes …
- Distinctive: line 178
keep distinct paper-level consequences separate even if they share a root cause
- Distinctive: line 199
Keep advisory items separate from blockers.
- Very specific: line 209
If both old and new final_issues.json bundles are available, also run:
- Distinctive: line 215
use structured prior issue bundles when available, but still accept Markdown previous reports
- Distinctive: line 219
Run the audit precheck:
- Distinctive: line 223
If blockers exist, stop and report them.
- Distinctive: line 224
Only proceed into polishing if the precheck is safe.
- Very specific: line 280
Default deep-review lanes live in agents/:
“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Paper Audit Skill v4.0
paper-audit is now deep-review-first. Its core job is to behave like a serious reviewer: find technical, methodological, claim-level, and cross-section issues; keep script-backed findings separate from reviewer judgment; and return a structured issue bundle plus a revision roadmap.
Use it for audit and review. Do not use it as the first tool for source editing, sentence rewriting, or build fixing.
What This Skill Produces
quick-audit: fast submission-readiness screen with script-backed findingsdeep-review: reviewer-style structured issue bundle with major/moderate/minor findingsgate: PASS/FAIL decision calibrated for submission blockersre-audit: compare current issue bundle against a previous auditpolish: precheck-only handoff into a polishing workflow
The primary product is no longer just a score. For deep-review, the main outputs are:
final_issues.jsonoverall_assessment.txtreview_report.mdrevision_roadmap.md
Do Not Use
- direct source surgery on
.tex/.typ - compilation debugging as the main task
- free-form literature survey writing
- cosmetic grammar cleanup without an audit goal
Critical Rules
- Never rewrite the paper source unless the user explicitly switches to an editing skill.
- Never fabricate references, baselines, or reviewer evidence.
- Always distinguish
[Script]from[LLM]findings. - Always anchor reviewer findings to a quote, section, or exact textual location.
- Be conservative with OCR noise, formatting quirks, and obvious copy-editing trivia.
- Review like a careful reader: understand the author's intended meaning before flagging an issue.
Mode Selection
| Requested intent | Mode |
| "check my paper", "quick audit", "submission readiness" | quick-audit |
| "review my paper", "simulate peer review", "harsh review", "deep review" | deep-review |
| "is this ready to submit", "gate this submission", "blockers only" | gate |
| "did I fix these issues", "re-audit", "compare against old review" | re-audit |
| "polish the writing, but only if safe" | polish |
Legacy aliases still work for one compatibility cycle:
self-check->quick-auditreview->deep-review
Review Standard
Read these references before running reviewer-style work:
references/REVIEW_CRITERIA.mdreferences/DEEP_REVIEW_CRITERIA.mdreferences/CHECKLIST.mdreferences/CONSOLIDATION_RULES.mdreferences/ISSUE_SCHEMA.md
The deep-review workflow uses a 10-part issue taxonomy:
- formula / derivation errors
- notation inconsistency
- prose vs formal object mismatch
- numerical inconsistency
- missing justification
- overclaim or claim inaccuracy
- ambiguity that can mislead a careful reader
- underspecified methods / missing information
- internal contradiction
- self-consistency of standards
Workflow
Common Step 0
Parse $ARGUMENTS and infer the mode if the user did not provide one. State the inferred mode before running commands if you had to infer it.
quick-audit
- Run:
uv run python -B "$SKILL_DIR/scripts/audit.py" <paper> --mode quick-audit ... - Present a concise report:
Submission Blockersfirst- then
Quality Improvements - then checklist items
- mark quick-audit findings with
[Script]provenance
- If the user clearly wants reviewer-depth critique after the quick screen, escalate to
deep-review.
deep-review
Use this as the default reviewer-style path.
Phase 1: Prepare workspace
Run:
uv run python -B "$SKILL_DIR/scripts/prepare_review_workspace.py" <paper> --output-dir ./review_resultsThis creates:
full_text.mdmetadata.jsonsection_index.jsonclaim_map.jsonpaper_summary.mdsections/*.mdcomments/
Phase 2: Phase 0 automated audit
Run:
uv run python -B "$SKILL_DIR/scripts/audit.py" <paper> --mode deep-review ...Treat this as Phase 0 only. It supplies script-backed context and scores, not the final review.
Phase 3: Section and cross-cutting review lanes
Read:
references/SUBAGENT_TEMPLATES.mdreferences/REVIEW_LANE_GUIDE.md
Then dispatch reviewer tasks for:
- section lanes
- introduction / related work
- methods
- results
- discussion / conclusion
- appendix, if present
- cross-cutting lanes
- claims vs evidence
- notation and numeric consistency
- evaluation fairness and reproducibility
- self-standard consistency
- prior-art and novelty grounding
Each lane writes a JSON array into comments/.
If subagents are unavailable, use the built-in deterministic fallback lane pass in scripts/audit.py so the workflow still writes lane-compatible JSON into comments/ before consolidation.
Phase 4: Consolidation
Run:
uv run python -B "$SKILL_DIR/scripts/consolidate_review_findings.py" <review_dir>
uv run python -B "$SKILL_DIR/scripts/verify_quotes.py" <review_dir> --write-back
uv run python -B "$SKILL_DIR/scripts/render_deep_review_report.py" <review_dir>Consolidation rules:
- merge exact duplicates
- keep distinct paper-level consequences separate even if they share a root cause
- preserve singleton findings unless clearly false positive
- assign
comment_type,severity,confidence, androot_cause_key
Phase 5: Present result
Summarize:
- 1 short paragraph overall assessment
- counts of major / moderate / minor issues
- 3 highest-priority revision items
- path to
review_report.mdandfinal_issues.json
gate
- Run:
uv run python -B "$SKILL_DIR/scripts/audit.py" <paper> --mode gate ... - Report PASS/FAIL.
- List blockers first.
- Keep advisory items separate from blockers.
- For IEEE pseudocode checks, make it explicit which issues are mandatory and which are only IEEE-safe recommendations.
re-audit
- Requires
--previous-report PATH. - Run:
uv run python -B "$SKILL_DIR/scripts/audit.py" <paper> --mode re-audit --previous-report <path> ... - If both old and new
final_issues.jsonbundles are available, also run:uv run python -B "$SKILL_DIR/scripts/diff_review_issues.py" <old_final_issues.json> <new_final_issues.json> - Present:
- root-cause-aware status labels:
FULLY_ADDRESSED,PARTIALLY_ADDRESSED,NOT_ADDRESSED,NEW - use structured prior issue bundles when available, but still accept Markdown previous reports
- root-cause-aware status labels:
polish
- Run the audit precheck:
uv run python -B "$SKILL_DIR/scripts/audit.py" <paper> --mode polish ... - If blockers exist, stop and report them.
- Only proceed into polishing if the precheck is safe.
Output Contract
For deep-review, the final issue schema is:
{
"title": "short issue title",
"quote": "exact quote from paper",
"explanation": "why this matters and what remains problematic",
"comment_type": "methodology|claim_accuracy|presentation|missing_information",
"severity": "major|moderate|minor",
"confidence": "high|medium|low",
"source_kind": "script|llm",
"source_section": "methods",
"related_sections": ["results", "appendix"],
"root_cause_key": "shared-normalized-key",
"review_lane": "claims_vs_evidence",
"gate_blocker": false,
"quote_verified": true
}Always prefer:
- exact quotes over vague paraphrase
- evidence-backed findings over style commentary
- issue bundle + roadmap over raw script dumps
References
| File | Purpose |
references/REVIEW_CRITERIA.md | top-level audit scoring and mapping |
references/DEEP_REVIEW_CRITERIA.md | deep-review-specific issue taxonomy and leniency rules |
references/CONSOLIDATION_RULES.md | deduplication and root-cause merge policy |
references/ISSUE_SCHEMA.md | canonical JSON schema |
references/REVIEW_LANE_GUIDE.md | section lanes and cross-cutting lanes |
references/SUBAGENT_TEMPLATES.md | reviewer task templates |
references/QUICK_REFERENCE.md | CLI and mode cheat sheet |
Scripts
| Script | Purpose |
scripts/audit.py | Phase 0 audit and mode entrypoint |
scripts/prepare_review_workspace.py | create deep-review workspace |
scripts/build_claim_map.py | extract headline claims and closure targets |
scripts/consolidate_review_findings.py | deduplicate comment JSONs |
scripts/verify_quotes.py | verify exact quote presence |
scripts/render_deep_review_report.py | render final Markdown report |
scripts/diff_review_issues.py | compare old vs new issue bundles |
Reviewer Lanes
Default deep-review lanes live in agents/:
section_reviewer_agent.mdclaims_evidence_reviewer_agent.mdnotation_consistency_reviewer_agent.mdevaluation_fairness_reviewer_agent.mdself_consistency_reviewer_agent.mdprior_art_reviewer_agent.mdsynthesis_agent.md
Legacy persona agents remain for compatibility but are no longer the default backbone of deep-review.
Examples
- “Review this manuscript like a serious conference reviewer and tell me the biggest validity risks.”
- “Run a quick audit on
paper.texand tell me what blocks submission.” - “Gate this IEEE submission and separate blockers from recommendations.”
- “Re-audit this revision against my previous report.”
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
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.

