{"id":"3dbb22cc-be95-43df-862f-7a2547259e53","slug":"paper-audit-skill","kind":"skill","name":"paper-audit-skill","summary":"Claude Code skill that audits and revises academic papers against paper-type and venue conventions. Six paper types, five venues, tiered voice rules.","owner":"Hiro-Inagawa","repoUrl":"https://github.com/Hiro-Inagawa/paper-audit-skill","license":"MIT","licenseTier":"permissive","trust":"community","stars":1,"match":null,"fewStars":true,"format":"skill","createdAt":"2026-04-17T00:21:17.000Z","addedAt":"2026-10-21T13:14:59.456Z","logoUrl":"https://avatars.githubusercontent.com/u/249512265?u=279f36d2774534a2d2ef4b7cf63fbfc1e9f2ff23&v=4","linkOnly":false,"collection":null,"skillPath":null,"categories":["developer-tools","writing","design-media"],"tags":["academic-writing","arxiv","claude-code","latex","peer-review","research-papers","writing"],"runtime":"binary","local":true,"cloudCapable":false,"cloudCapability":"unknown","cloudReadyOnPlatform":false,"needsApiKey":false,"reviewed":false,"syncedAt":"2026-10-23T05:42:04.016Z","sample":false,"url":"https://conduits.dev/tools/paper-audit-skill","language":"TeX","transports":["stdio"],"install":{},"requirements":{"env":[],"upstreamMayCharge":false},"tools":null,"readme":"```\n                                                  _ _ _\n _ __   __ _ _ __   ___ _ __       __ _ _   _  __| (_) |_\n| '_ \\ / _` | '_ \\ / _ \\ '__|____ / _` | | | |/ _` | | __|\n| |_) | (_| | |_) |  __/ | |_____| (_| | |_| | (_| | | |_\n| .__/ \\__,_| .__/ \\___|_|        \\__,_|\\__,_|\\__,_|_|\\__|\n|_|         |_|\n```\n\n**Draft papers like a journal editor reviews them.**\n\n# paper-audit-skill\n\n**A Claude Code skill for writing academic papers. Six paper types. Five venues. Tiered voice rules.**\n\nPaper-type-aware, venue-aware drafting and revision. The skill asks which paper type you are writing and where you are submitting, then applies the conventions a journal or conference reviewer would expect.\n\n## Why This Exists\n\nGeneric writing tools produce generic prose. Academic papers are judged by different standards per field and venue: a CHI autoethnography and a NeurIPS methods section have almost nothing in common in voice, structure, or reviewer risk.\n\nThis skill encodes those differences. It:\n\n- Identifies the paper type (empirical quantitative, empirical qualitative, autoethnography, theory, systematic review, position).\n- Identifies the venue (arxiv, NeurIPS/ICML/ICLR, CHI/CSCW/DIS, journal, thesis).\n- Loads the references that apply.\n- Applies tiered voice rules (universal hard-rejects and contextual warnings).\n- Flags reviewer risks specific to the paper type.\n- Points at the right LaTeX template and citation style.\n\n## What It Does\n\n```\nYou: \"I'm writing an empirical HCI paper for CHI. Help me draft the introduction.\"\n  |\n  +-- Skill asks clarifying questions if type/venue aren't clear\n  +-- Loads references/00-paper-types.md §empirical-qualitative\n  +-- Loads references/01-venues.md §CHI\n  +-- Loads references/02-section-moves.md §Introduction (CARS model)\n  +-- Loads voice rules (03-voice-positive, 04-voice-negative)\n  |\n  +-- Drafts with:\n        - Authorial \"we\"\n        - CARS Introduction (territory → niche → occupy)\n        - Four contribution bullets at the end\n        - No em dashes, no \"In today's world,\" no \"groundbreaking\"\n        - Hedging where appropriate\n```\n\nThe skill does not:\n\n- Generate citations from nothing. You still provide sources.\n- Check factual accuracy. The skill structures claims that need citations.\n- Run literature searches. It advises on what types of sources to cite.\n- Produce figures or run statistics. It advises on specs and reporting.\n\n## Installation\n\nOne command:\n\n```bash\ncurl -fsSL https://raw.githubusercontent.com/Hiro-Inagawa/paper-audit-skill/main/install.sh | bash\n```\n\nWorks on macOS and Linux. On Windows, run in Git Bash or WSL.\n\n### Alternative: Manual install\n\n```bash\ngit clone https://github.com/Hiro-Inagawa/paper-audit-skill.git /tmp/paper-audit-skill\ncp -r /tmp/paper-audit-skill/skills/paper-audit-skill ~/.claude/skills/\nrm -rf /tmp/paper-audit-skill\n```\n\nNo dependencies to install. The skill is pure markdown; it does not require Node, Playwright, or any other runtime.\n\n## Usage\n\n### In Claude Code\n\nThe skill auto-activates when you mention an academic paper, manuscript, arxiv, thesis, or related terms. It also activates on `.tex` and `.bib` file extensions.\n\nFirst interaction: the skill asks for paper type and venue. Answer honestly; different combinations load different rules.\n\n### Example prompts\n\n```\n\"Help me write the introduction to my NeurIPS paper on attention mechanisms.\"\n\"Review my Related Work section for a CHI submission.\"\n\"I'm writing a systematic review. What sections do I need?\"\n\"My reviewer said n=1 isn't generalizable. How do I respond?\"\n\"Draft the abstract for my arxiv preprint.\"\n```\n\n### Explicit invocation\n\nIf the skill does not auto-activate when you want it, reference it directly:\n\n```\n\"Using the paper-audit-skill skill, review this paragraph.\"\n```\n\n## Supported Paper Types\n\n| Type | Best for | Structure |\n|------|----------|-----------|\n| Empirical quantitative | Stats, hypothesis testing | IMRaD |\n| Empirical qualitative | Interviews, observation | Methods + Findings + Discussion |\n| Autoethnography | Single-subject, lived experience | Flexible; narrative + reflexivity |\n| Theory | Frameworks, formal proofs | Problem + Thesis + Argument + Consequences |\n| Systematic review | Evidence synthesis | PRISMA 2020 |\n| Position paper | Argument without new data | Scope + Thesis + Argument + Counterargument |\n\n## Supported Venues\n\n| Venue | Length | Voice | Citation style |\n|-------|--------|-------|----------------|\n| arxiv preprint | 8-15 pages | Technical, authorial \"we\" | natbib (author-year) typical |\n| NeurIPS, ICML, ICLR | 8 pages main | Dense, formal | natbib |\n| CHI, CSCW, DIS | 10-14 pages | Narrative OK; reflexive | ACM numeric |\n| Journal (general) | 6,000-10,000 words | Formal, venue-specific | APA, Chicago, Vancouver, etc. |\n| Thesis (Masters, PhD) | 10K-100K words | Formal; full literature chapter | University-specific |\n\n## Tiered Voice Rules\n\nThe skill enforces two tiers of voice rules so it does not over-enforce across disciplines.\n\n**Tier 1 (universal hard-rejects)** scanned on every paragraph:\n\n- Em dashes in formal prose\n- Contractions in formal prose\n- \"As an AI,\" \"As a language model\"\n- Filler openings (\"It's worth noting,\" \"In today's world,\" \"Since the dawn of\")\n- Overclaiming (\"groundbreaking,\" \"revolutionary,\" \"very unique\")\n- \"Clearly shows,\" \"obviously\" (without hedge)\n\n**Tier 2 (contextual warnings)** flagged for human review:\n\n- Passive voice outside Methods\n- Nominalizations where verb form is clearer\n- Hedging where boosters are warranted (and vice versa)\n- First-person \"I\" in a \"we\" venue\n\nThe user can override any Tier 2 flag based on disciplinary convention.\n\n## Verification\n\nThis skill is auditable. See `verify.md` in the repository root for 20 ablation test scenarios.\n\nEach test provides a deliberately flawed input and lists the issues the skill should catch, plus false-positive watch items the skill should NOT flag. Run the tests in a fresh Claude Code session to confirm the skill works as documented.\n\n### Quick audit\n\n```bash\ncat verify.md\n```\n\nRead through the tests. Pick three or four. Try them in a fresh Claude Code session. If the skill catches the expected issues and does not over-flag, it passes.\n\n## Reference Files\n\nThe skill body is in `skills/paper-audit-skill/`. It contains:\n\n```\nSKILL.md                           # Top-level rules and routing\nreferences/\n  00-paper-types.md                # Six types, structure per type\n  01-venues.md                     # Five venues, conventions per venue\n  02-section-moves.md              # Swales moves per section\n  03-voice-positive.md             # What to do: character-subject, old-new, hedging\n  04-voice-negative.md             # What to avoid: tiered forbidden phrases\n  05-argument-structure.md         # Booth: claim, reason, evidence, warrant, acknowledgment\n  06-citation-triangle.md          # Every claim has evidence, evidence has citation\n  07-citation-styles.md            # APA, Chicago, IEEE, ACM, Vancouver, natbib, biblatex\n  08-latex-conventions.md          # Multi-venue LaTeX, .bbl gotcha\n  09-defense-playbooks.md          # Reviewer risks per paper type, preemption\n  10-abstract-craft.md             # 150-250 word formula, four moves\n  11-revision-discipline.md        # Williams Style six-pass revision\n  12-ai-ethics.md                  # Disclosure rules per venue (2024-2026)\nassets/\n  templates/\n    arxiv-generic.tex              # Open-license arxiv scaffold\n    journal-generic.tex            # Generic journal scaffold\n    POINTERS.md                    # Where to get venue-specific templates\n  bibliography-templates.bib       # BibTeX entry templates per source type\n```\n\n## Requirements\n\n- Claude Code installed.\n- No additional runtime dependencies.\n- Target: academic writing in English, Latin-script. Non-Latin-script conventions are out of scope for v1.\n\n## Philosophy\n\nAcademic writing is a craft. The skill does not replace the writer; it enforces the conventions the writer needs to respect to be taken seriously. Every rule in the skill is derived from:\n\n- Swales & Feak, *Academic Writing for Graduate Students*.\n- Booth, Colomb, & Williams, *The Craft of Research*.\n- Williams & Bizup, *Style: Lessons in Clarity and Grace*.\n- PRISMA 2020 reporting standards.\n- Ellis, Adams, & Bochner, autoethnography tradition.\n- Yin, *Case Study Research*.\n- APA, Chicago, IEEE, ACM, Vancouver citation standards.\n- Current (2024-2026) AI-assisted writing disclosure policies at ACM, ICML, CHI, and major journals.\n\n## License\n\nMIT. See [LICENSE](LICENSE).\n\n## Contributing\n\nIssues and pull requests welcome. Particularly: domain-specific additions (e.g., biomedical reporting standards beyond Vancouver, humanities conventions beyond Chicago), venue-specific templates, and translations of the voice rules for non-English academic contexts.\n\n## Related\n\n- [web-reader](https://github.com/Hiro-Inagawa/web-reader) — Read any URL from Claude Code. Same repo pattern; reads websites instead of writing papers.\n"}