writing-skills
AGENTS.md54 starsChanged 4 months agoMITView on GitHub
These instructions apply to the whole repository. - Immediate child directories under for-agents/ containing SKILL.md are installable agent skills. - Everything under for-humans/ is for people and must not contain SKILL.md. Do not route ordinary writing tasks there. paper-writing may point to the guide or consult the relevant chapter for explicit template-use, teaching, or human-instruction requests. Otherwise load the focused agent references. - Keep private papers, proposals, downloaded arXiv source, and user drafts outside this repository. - Preserve license and…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,439 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
18 rules. About 0 are common, 5 are distinctive, and 13 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
13
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Distinctive: line 14
Keep private papers, proposals, downloaded arXiv source, and user drafts outside this repository.
- Very specific: line 21
Read the selected SKILL.md in full.
- Distinctive: line 22
Read only the references or assets needed for the requested change.
- Very specific: line 23
Check sibling skills before changing a shared workflow.
- Very specific: line 25
Run python3 scripts/validate-repo.py to establish a clean baseline.
- Very specific: line 30
Keep the frontmatter name equal to the folder name.
- Very specific: line 31
Put trigger conditions and the complete operating procedure in SKILL.md.
- Very specific: line 34
Use relative links. Never add a username, home-directory path, or assumed Documents layout.
- Distinctive: line 36
Keep instructions direct and testable.
- Very specific: line 38
Keep SKILL.md under 500 lines.
- Very specific: line 40
Update agents/openai.yaml when a skill's name or user-facing scope changes.
- Distinctive: line 45
Keep private corpora outside this repository.
- Very specific: line 64
Keep planning and drafting separate where the repository already does so, especially for grant-planning and grant-writing.
- Distinctive: line 71
Keep each LaTeX entry point in the outermost project directory.
- Very specific: line 72
Use a stable, descriptive <project-name>-main.tex filename, never a generic main.tex.
- Very specific: line 76
Do not finish a writing task with the canonical PDF named main.pdf or available only under build/.
- Very specific: line 81
Use improve-human-writing-guide for every proposed or applied change to the human guide.
- Very specific: line 100
Do not edit generated files in build/ by hand.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md 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.
Files
Instructions for agents
These instructions apply to the whole repository.
Repository boundaries
- Immediate child directories under
for-agents/containingSKILL.mdare installable agent skills. - Everything under
for-humans/is for people and must not containSKILL.md. Do not route ordinary writing tasks there.paper-writingmay point to the guide or consult the relevant chapter for explicit template-use, teaching, or human-instruction requests. Otherwise load the focused agent references. - Keep private papers, proposals, downloaded arXiv source, and user drafts outside this repository.
- Preserve license and attribution files, especially in
humanizerandgeneral-writing.
Before editing
- Read the selected
SKILL.mdin full. - Read only the references or assets needed for the requested change.
- Check sibling skills before changing a shared workflow. The
writingskill owns routing;general-writingowns the Humanizer-first prose pipeline. - Run
python3 scripts/validate-repo.pyto establish a clean baseline. - Inspect the working tree and preserve unrelated changes.
Skill design
- Keep the frontmatter
nameequal to the folder name. - Put trigger conditions and the complete operating procedure in
SKILL.md. - Move detailed domain knowledge, long examples, and checklists into
references/; place reusable starter files inassets/. - Use relative links. Never add a username, home-directory path, or assumed
Documentslayout. - Keep instructions direct and testable. State what an agent must inspect, produce, preserve, and verify.
- Keep
SKILL.mdunder 500 lines. Add an abstraction only when it removes repeated instructions or gives one skill clear ownership of a workflow. - Update
agents/openai.yamlwhen a skill's name or user-facing scope changes. The default prompt must mention the exact$skill-name.
Optional corpora
Keep private corpora outside this repository. When a task would benefit from
original examples, use a directory the user supplied for that task. If no
directory was supplied, ask where the relevant example folder lives and make
clear that the user can answer none.
Accept either the corpus directory itself or a parent containing the relevant
example-papers or example grants directory. Do not search above the
repository or assume a home-directory layout. If the corpus is unavailable,
use the distilled reference bundled with the relevant skill and say that the
original artifacts were not inspected. Never copy private corpus content into
a contribution without explicit permission.
Cross-skill behavior
- Let
writingroute requests; do not duplicate full domain instructions in the router. - Let
general-writinginvoke Humanizer and then enforce the author's house style. Domain skills should call that workflow once rather than reimplement it. - Keep planning and drafting separate where the repository already does so,
especially for
grant-planningandgrant-writing. - Preserve factual claims, citations, notation, and format constraints during prose edits.
Named LaTeX artifacts
- Keep each LaTeX entry point in the outermost project directory.
- Use a stable, descriptive
<project-name>-main.texfilename, never a genericmain.tex. - Compile to
build/<project-name>-main.pdf, then copy that exact PDF to<project-name>-main.pdfin the outermost project directory. - Do not finish a writing task with the canonical PDF named
main.pdfor available only underbuild/.
Human guide changes
Use improve-human-writing-guide for every proposed or applied change to the
human guide. Its prose workflow is mandatory: run Humanizer in embedded mode,
then general-writing and its evaluation checklist before finalizing wording.
Compile from for-humans/human-writing-guide after every meaningful source
change:
latexmk -g -pdf -interaction=nonstopmode -halt-on-error \
-outdir=build human-writing-guide-main.texInspect the rendered PDF. Before finishing, publish the exact build at the documented top-level name:
cp build/human-writing-guide-main.pdf human-writing-guide-main.pdf
cmp -s build/human-writing-guide-main.pdf human-writing-guide-main.pdfDo not edit generated files in build/ by hand.
Finish
Run:
python3 scripts/validate-repo.pyFor installer changes, also smoke-test both link and copy modes in a temporary directory. Report any test that could not be run.
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.

