academic-writing-skills

academic-writing-skills

A shareable Claude/Codex skill for rigorous academic paper writing, revision, reviewer response, figure-text consistency, and pre-submission audits.

Traditional Chinese README

📚 Part of the agentic AI learning roadmap — a 7-stage curated path for building agentic AI, multilingual (zh-TW · zh-CN · English). This skill is referenced in §13 (research workflow skills).

Purpose

This skill helps researchers turn manuscript work into a repeatable workflow:

• Draft sections with findings-first structure.
• Rewrite results and discussion paragraphs with explicit mechanisms.
• Detect GPT-style prose, vague intensifiers, and overclaim language.
• Check whether claims are backed by figures, tables, statistics, code output, or literature.
• Keep numbers, figure panels, captions, and prose consistent.
• Build point-by-point reviewer response tables.
• Prepare journal submission checklists and declaration items.
• Create .paper/ context files so future AI sessions use fewer tokens.

The skill is field-agnostic. Journal-specific rules and paper-specific terminology live inside each paper repository.

Install

Install via the ai-research-skills Claude Code marketplace:

claude plugin marketplace add WenyuChiou/ai-research-skills
claude plugin install academic-writing-skills@ai-research-skills --scope user

Migrating from a previous git clone? The skill repo was restructured to a skills/<name>/SKILL.md layout that the marketplace requires. The old git clone https://github.com/WenyuChiou/academic-writing-skills ~/.claude/skills/academic-writing-skills path no longer loads (Claude Code's user-skills loader only scans one level deep). Either uninstall the cloned copy (rm -rf ~/.claude/skills/academic-writing-skills) and run the marketplace install above, or clone the inner subdirectory: git clone -n https://github.com/WenyuChiou/academic-writing-skills /tmp/aws && mv /tmp/aws/skills/academic-writing-skills ~/.claude/skills/academic-writing-skills.

Per-Paper Setup

For each manuscript, create a .paper/ folder in the paper repository:

<paper-repo>/
  .paper/
    journal_format.md
    style_overrides.md
    context.md
    figure_inventory.md
    claim_evidence_ledger.md
    reviewer_comments.md
    submissions_log.md

Minimum setup:

1. Copy references/journal_format_template.md to <paper-repo>/.paper/journal_format.md.
2. Fill in target journal rules from current author guidelines.
3. Optionally copy references/style_overrides_example.md to <paper-repo>/.paper/style_overrides.md.
4. For long projects, create the context packet described in references/paper_context_packet.md.

Skill Layout

academic-writing-skills/
  SKILL.md
  references/
    banned_words.md
    claim_evidence_audit.md
    figure_conventions.md
    journal_format_template.md
    paper_context_packet.md
    reviewer_response_workflow.md
    section_checklists.md
    style_overrides_example.md
    submission_checklist.md
    writing_principles.md
  evals/
    evals.json
  tests/
    test_skill_integrity.py

Typical Workflows

Draft or revise a section

The skill loads journal format, paper overrides, writing principles, banned words, and the relevant section checklist. It then rewrites with:

• finding before figure citation,
• mechanism after each result,
• precise terminology,
• no unsupported overclaim,
• panel-specific figure references when needed.

Audit claims and evidence

The skill builds a claim-evidence ledger:

claim -> evidence source -> allowed certainty -> missing check -> revision

This is useful before editing Abstract, Discussion, Conclusion, cover letter, or reviewer response.

Respond to reviewers

The skill converts comments into a response table:

comment -> anchor text -> response -> manuscript change -> evidence

It avoids the common failure mode of saying "clarified" without a visible manuscript change.

Save tokens across AI sessions

For long manuscripts, the skill can maintain .paper/context.md, .paper/figure_inventory.md, and .paper/claim_evidence_ledger.md. Future sessions can read those concise files before opening the full manuscript.

What This Skill Is Not

• It is not a Zotero manager.
• It is not an Obsidian or NotebookLM workspace skill.
• It is not a coding-agent delegation skill.
• It is not a generic grammar checker.
• It does not invent scientific assumptions, results, or citations.

Testing

Run integrity tests with:

python -m pytest -q

The tests validate frontmatter, bundled references, eval prompts, and common mojibake/encoding corruption markers.

License

MIT