skills
AGENTS.md18 starsChanged todayMITView on GitHub
Follow global defaults; this file contains only repository-specific additions and overrides.
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.
In this repository13 other files it gives its agents.
- apply-google-styleSkill
- chrome-extensionsSkill
- configure-ghosttySkill
- correctSkill
- give-language-feedbackSkill
- lean-verifySkill
- modern-web-guidanceSkill
- prompt-gptSkill
- record-human-feedbackSkill
- security-auditSkill
- write-agent-skillsSkill
- write-agents-mdSkill
- write-plansSkill
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
27 rules. About 1 is common, 14 are distinctive, and 12 are very specific to this project.
≈1
Common · estimate
Many files like this have a rule of this kind.
≈14
Distinctive · estimate
This author's own choices. Worth a look.
12
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 3
Follow global defaults; this file contains only repository-specific additions and overrides.
- Distinctive: line 8
Use clear structure, familiar words, and concise sentences.
- Very specific: line 12
Follow the model-selection policy in ./skills/prompt-gpt/SKILL.md when creating, revising, or reviewing skills and other agent-facing …
- Distinctive: line 13
Keep external positioning, installation flows, and skill discovery in README.md, and keep maintainer workflow in this file.
- Distinctive: line 14
Update the README catalog when a skill is added, deprecated, renamed, recategorized, or materially changes its trigger.
- Distinctive: line 15
Update installation documentation and executable recipes only in the files that own the affected flow.
- Distinctive: line 19
Follow KISS (Keep It Simple) and YAGNI (You Aren't Gonna Need It).
- Distinctive: line 20
Prefer simple, minimal solutions over unnecessary complexity.
- Very specific: line 21
Use lowercase kebab-case for skill directories and name every required entry file exactly SKILL.md.
- Distinctive: line 24
Keep a skill's frontmatter description and README catalog entry aligned when its trigger or purpose changes.
- Distinctive: line 25
Keep examples repository-relative and executable when practical.
- Distinctive: line 26
Use uv with PEP 723 inline metadata for new or modified standalone Python scripts, following the official script guide.
- Very specific: line 28
Add or update third-party dependencies with uv add --script path/to/script.py <package>.
- Very specific: line 29
Run scripts with uv run path/to/script.py.
- Distinctive: line 37
Ask before writing to external systems, taking destructive or costly actions, or materially expanding the scope.
- Very specific: line 38
Ask before running scripts/download_human_interface_guidelines.py because its default mode downloads a large external corpus.
- Very specific: line 39
Treat ignored build/ content as generated output.
- Distinctive: line 41
Do not introduce root-level marketplace or plugin metadata unless corresponding repository files and workflows exist.
- Common: line 46
Run every changed bundled script through a non-destructive path that exercises the changed behavior.
- Very specific: line 47
If one-off uv run --with ...
- Very specific: line 48
Run prek run -a as the repository-wide formatting and lint gate, and report any failure before a pull request.
- Very specific: line 49
Run prek run lint-python-scripts -a to check PEP 723 metadata, dependency and Python version specifiers, and executable headers for every …
- Distinctive: line 50
Run git diff --check and inspect the final diff for unintended scope or repeated guidance.
- Very specific: line 56
Keep deprecated skills in deprecated/<skill-name>/SKILL.md, outside active discovery.
- Very specific: line 57
Keep optional supporting material inside its skill directory under references/, scripts/, assets/, or agents/.
- Very specific: line 58
Keep source slides and visual examples under examples/.
- Distinctive: line 63
Add screenshots or rendered output links for slide visual changes.
≈ “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
Repository Guidelines
Follow global defaults; this file contains only repository-specific additions and overrides.
Communication & Documentation
- Lead with the most important relevant information and omit anything unnecessary or repeated.
- Use clear structure, familiar words, and concise sentences.
- Explain the main idea simply before adding necessary detail.
- Keep information accurate.
- Make documented rules specific and verifiable.
- Follow the model-selection policy in
./skills/prompt-gpt/SKILL.mdwhen creating, revising, or reviewing skills and other agent-facing prompts. - Keep external positioning, installation flows, and skill discovery in
README.md, and keep maintainer workflow in this file. - Update the README catalog when a skill is added, deprecated, renamed, recategorized, or materially changes its trigger.
- Update installation documentation and executable recipes only in the files that own the affected flow.
Code Style
- Follow KISS (Keep It Simple) and YAGNI (You Aren't Gonna Need It).
- Prefer simple, minimal solutions over unnecessary complexity.
- Use lowercase kebab-case for skill directories and name every required entry file exactly
SKILL.md. - Preserve a skill's original user intent when naming or renaming it; do not force
<verb-ing>-<object>when that changes the meaning. - Retain explicit-invocation skills that provide useful mode shortcuts even when their behavior can be inferred; keep
explain-step-by-stepactive. - Keep a skill's frontmatter description and README catalog entry aligned when its trigger or purpose changes.
- Keep examples repository-relative and executable when practical.
- Use uv with PEP 723 inline metadata for new or modified standalone Python scripts, following the official script guide.
- Initialize new scripts with
uv init --script path/to/script.pyand declare dependencies in the# /// scriptblock, usingdependencies = []when none are required. - Add or update third-party dependencies with
uv add --script path/to/script.py <package>. - Run scripts with
uv run path/to/script.py. - For direct execution, use
#!/usr/bin/env -S uv run --scriptas the first line and set the executable bit; the shebang does not replace dependency metadata.
- Initialize new scripts with
- For user interfaces, apply Apple-derived design philosophy across platforms while translating platform-specific metrics and controls to target conventions; minimize cognitive load without sacrificing functional completeness, keep critical actions and state visible, and use predictable progressive disclosure for secondary complexity.
Boundaries
- For answer, explanation, review, diagnosis, or planning requests, inspect the relevant materials and report without making changes.
- For change, build, or fix requests, make the requested in-scope local changes and run relevant safe checks without asking first.
- Ask before writing to external systems, taking destructive or costly actions, or materially expanding the scope.
- Ask before running
scripts/download_human_interface_guidelines.pybecause its default mode downloads a large external corpus; when maintaining the archiver, enumerate the DocC navigator and page JSON under/tutorials/data/instead of recursively downloading HTML. - Treat ignored
build/content as generated output; do not hand-edit or commit it. - [UNREVIEWED] Treat
skills/and{writing,research,visuals,design,code-quality,workflow}/skills/as the active source roots;.agents/skillsis a tracked local-discovery symlink toskills/, not a second copy to edit. - Do not introduce root-level marketplace or plugin metadata unless corresponding repository files and workflows exist.
Testing
- This repository has no maintained automated test suite; do not add or maintain repository tests or apply TDD to repository changes.
- Run every changed bundled script through a non-destructive path that exercises the changed behavior; report the command and result, and identify any path that credentials, network access, or unavailable dependencies prevented you from verifying.
- If one-off
uv run --with ...validation fails because~/.cache/uvis read-only, setUV_CACHE_DIR=/tmp/uv-cache; request temporary network access only when an uncached dependency still cannot resolve. - Run
prek run -aas the repository-wide formatting and lint gate, and report any failure before a pull request. - Run
prek run lint-python-scripts -ato check PEP 723 metadata, dependency and Python version specifiers, and executable headers for every tracked.pyfile, including helpers anddeprecated/; the linter does not infer dependencies from imports. - Run
git diff --checkand inspect the final diff for unintended scope or repeated guidance. - For changes under
examples/slides/, verify source assets and include the generated preview or screenshots in the pull request.
Repository structure
- [UNREVIEWED] Keep global skills directly in
skills/<skill-name>/SKILL.mdand category skills directly in<category>/skills/<skill-name>/SKILL.md, using only the active category roots listed above without deeper category directories. - Keep deprecated skills in
deprecated/<skill-name>/SKILL.md, outside active discovery. - Keep optional supporting material inside its skill directory under
references/,scripts/,assets/, oragents/. - Keep source slides and visual examples under
examples/.
Git and commits
- In pull requests, summarize what changed and why, list affected paths, and report verification outcomes with any skipped or failing check.
- Add screenshots or rendered output links for slide visual changes.
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.

