research-writing
Skill141 starsChanged 5 months agoMITView on GitHub
Draft and maintain academic papers from verified corpus evidence, manage bibliography, and produce output in any target format. Use when outlining chapters, mapping evidence to arguments, drafting or revising sections, managing citations, placing figures, or running integrity checks before building output.
7 other projects use a version of this.
Adopted by
Most recently taken up by
- DJDeborah/research-mechanics-skillstheir own version, first seen Oct 2, 2026
- Rifa-G/skillstheir own version, first seen Aug 9, 2026
- Odinary-AI/research-writing-skilltheir own version, first seen Jul 29, 2026
- lirabenjamin/claude-code-for-researcherstheir own version, first seen Apr 15, 2026
- myhsieh1002/claude-skillstheir own version, first seen Apr 11, 2026
- TabithaFanny/claude-skills-collectiontheir own version, first seen Mar 31, 2026
Other versions7 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 7
- TabithaFanny/claude-skills-collection★ 4same name
- lirabenjamin/claude-code-for-researchers★ 2same name
- Odinary-AI/research-writing-skill★ 1same name
- qike7777/skill-library★ 1same name
- Rifa-G/skills★ 1same name
- DJDeborah/research-mechanics-skills★ 0same name
- myhsieh1002/claude-skills★ 0same 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 repository4 other files it gives its agents.
- corpus-buildingSkill
- evidence-reviewSkill
- literature-discoverySkill
- source-intakeSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
29 rules. 0 are common, 22 are distinctive, and 7 are very specific to this project.
0
Common
Most of the 6 edited versions we read keep these.
22
Distinctive
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 19
Write from verified evidence, not from memory.
- Distinctive: line 33
Do not start drafting paragraphs before this mapping is clear.
- Distinctive: line 40
Write synthesis paragraphs in your own analytical voice
- Distinctive: line 47
Keep one master bibliography file.
- Distinctive: line 54
Add to the bibliography as you add citations to the draft, not at the end.
- Distinctive: line 56
Verify bibliography integrity regularly:
- Distinctive: line 57
Every inline citation resolves to a bibliography entry
- Distinctive: line 58
Every bibliography entry is actually cited somewhere (no phantom references)
- Distinctive: line 59
Every cited slice can be traced to a source object
- Distinctive: line 66
Use the canonical source page image or a crop of it, not a screenshot
- Distinctive: line 69
Keep a record of: source image path, crop coordinates, output filename, and why this is a single figure element
- Distinctive: line 73
Before compiling or exporting:
- Distinctive: line 74
Every major claim has a citation
- Distinctive: line 75
Every citation resolves in the bibliography
- Distinctive: line 76
Every figure has a documented source trail
- Distinctive: line 77
No internal process terms appear in the body text
- Distinctive: line 83
If you generate a manuscript snapshot for app-layer reading or collaboration, treat it as read-only — edits belong in the canonical source, …
- Distinctive: line 89
Page locators: keep them in footnotes or endnotes, not inline in body text
- Distinctive: line 91
Term lists: use a table for more than 3–4 term correspondences, followed by a prose paragraph that interprets the table
- Distinctive: line 92
Alarm vocabulary: avoid "濒危" (endangered), "亟待抢救" (urgently needs rescue) and similar terms unless directly supported by a cited source
- Very specific: line 96
Write only from evidence at reviewed level or above.
- Very specific: line 97
Do not introduce a claim without a citation trace.
- Very specific: line 98
Do not expose internal workflow terms in the body text: corpus, slice, SLC-, SRC-, source card, OCR, agent.
- Distinctive: line 99
Do not place a whole source page as a figure.
- Distinctive: line 100
Do not change a citekey once it appears in any draft file.
- Distinctive: line 102
Run bibliography integrity check before every build.
- Very specific: line 106
Read references/bibliography-management.md for citekey rules, deduplication procedure, and bibliography sync workflow.
- Very specific: line 107
Read references/figure-workflow.md for the full figure extraction, cropping, and captioning procedure.
- Very specific: line 108
Read references/proposal-conventions.md for doctoral proposal rhetorical structure and common mistakes to avoid.
“Common” is measured: kept by at least 6 in 10 of the 6 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
Research Writing
The core problem
Two failure modes are common in AI-assisted academic writing:
- Writing from memory: using AI to draft paragraphs without grounding claims in verified source evidence. Produces fluent but unchecked text.
- Leaking the workflow: letting internal process terms (
corpus,slice,source card,OCR pipeline) appear in the final document. The paper should read as the researcher's scholarly analysis, not as a description of how it was produced.
This skill addresses both.
Core principle
Write from verified evidence, not from memory. Every claim in the paper should trace back to at least one evidence slice at reviewed or quote_ready level. Every figure should trace back to a reviewed image slice.
The paper is the output of the research system. It should not expose how the system works.
Workflow
1. Build the chapter outline before drafting
For each chapter or section, define:
- The central question or argument this section answers
- The evidence slices that support the argument (from your evidence matrix)
- The figures or tables that illustrate it
- The citekeys required
Do not start drafting paragraphs before this mapping is clear.
2. Draft from grouped evidence
For each subsection:
- Gather the evidence slices assigned to it
- Identify what they collectively show
- Write synthesis paragraphs in your own analytical voice
- Insert citations inline as you draft — do not leave "[CITE]" placeholders
On citations: insert a citation only when the supporting slice is at least verification_status: checked. For direct quotations, quote_ready: true is required.
3. Manage bibliography in parallel
Keep one master bibliography file. One source gets one citekey.
Citekey rules:
- ASCII only
- Format:
AuthorYYYYKeyword(e.g.,Smith2019vernacular) - Once a citekey appears in any draft, treat it as frozen — changing it later breaks citations silently
Add to the bibliography as you add citations to the draft, not at the end.
Verify bibliography integrity regularly:
- Every inline citation resolves to a bibliography entry
- Every bibliography entry is actually cited somewhere (no phantom references)
- Every cited slice can be traced to a source object
4. Place figures purposefully
A figure earns its place by serving a specific argumentative role. Place it after the paragraph that establishes why it matters, not before.
For each figure:
- Use the canonical source page image or a crop of it, not a screenshot
- Crop exactly one figure element — not a full page, not a region containing multiple independent figures
- The caption states the scholarly source and page
- Keep a record of: source image path, crop coordinates, output filename, and why this is a single figure element
5. Run an integrity check before every output build
Before compiling or exporting:
- Every major claim has a citation
- Every citation resolves in the bibliography
- Every figure has a documented source trail
- No internal process terms appear in the body text
6. Produce output in your target format
The manuscript source (LaTeX, Markdown, Word, etc.) is canonical. All exports (PDF, DOCX, HTML) are derivatives.
If you generate a manuscript snapshot for app-layer reading or collaboration, treat it as read-only — edits belong in the canonical source, not in the snapshot.
Formatting rules for Chinese academic prose
(Skip this section if not writing in Chinese)
- Page locators: keep them in footnotes or endnotes, not inline in body text
- Abbreviations: spell out every abbreviation on first use:
中文含义(Full Term, ABBR) - Term lists: use a table for more than 3–4 term correspondences, followed by a prose paragraph that interprets the table
- Alarm vocabulary: avoid "濒危" (endangered), "亟待抢救" (urgently needs rescue) and similar terms unless directly supported by a cited source
Rules
- Write only from evidence at
reviewedlevel or above. Do not draft claims you cannot immediately trace to a slice. - Do not introduce a claim without a citation trace. Unsupported claims belong in
notes/until evidence is found. - Do not expose internal workflow terms in the body text:
corpus,slice,SLC-,SRC-,source card,OCR,agent. These belong in comments, notes, and working documents. - Do not place a whole source page as a figure. Crop one meaningful figure element.
- Do not change a citekey once it appears in any draft file.
- The manuscript snapshot is read-only. Canonical edits go into the source file.
- Run bibliography integrity check before every build.
References
- Read
references/bibliography-management.mdfor citekey rules, deduplication procedure, and bibliography sync workflow. - Read
references/figure-workflow.mdfor the full figure extraction, cropping, and captioning procedure. - Read
references/proposal-conventions.mdfor doctoral proposal rhetorical structure and common mistakes to avoid.
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.

