ai-agent-skills
AGENTS.md1 starsChanged 30 days agoMITView on GitHub
This repository maintains shared skills for Cursor, Codex, Claude Code, OpenCode, and Copilot. These instructions apply to work in this repository; they are not automatically installed elsewhere. - Keep canonical content in skills/<name>/SKILL.md and supporting resources in that skill's folder. Host discovery links point to this shared source. - Read skills/anti-ai-writing/SKILL.md before human-facing text. Use skills/using-skills/SKILL.md to select relevant workflows when routing is needed. Read the selected skill; do not treat its description as the complete procedure. - Keep changes…
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 31,774 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 repository23 other files it gives its agents.
- address-feedbackSkill
- agent-planningSkill
- anti-ai-writingSkill
- api-designSkill
- ci-cd-workflowsSkill
- code-reviewSkill
- decision-partnerSkill
- explain-itSkill
- frontend-engineeringSkill
- git-deliverySkill
- implement-planSkill
- jira-ticketsSkill
- readme-designSkill
- release-readinessSkill
- review-lenseSkill
- root-cause-debugSkill
- safe-migrationsSkill
- skill-authoringSkill
- technical-documentationSkill
- test-authoringSkill
- using-skillsSkill
- writing-clean-simple-codeSkill
- writing-optimized-sqlSkill
The file
We split this AGENTS.md into the rules it gives an agent
11 rules. About 2 are common, 7 are distinctive, and 2 are very specific to this project.
≈2
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 6
Keep canonical content in skills/<name>/SKILL.md and supporting resources in that skill's folder.
- Very specific: line 8
Read skills/anti-ai-writing/SKILL.md before human-facing text.
- Distinctive: line 10
Read the selected skill; do not treat its description as the complete procedure.
- Distinctive: line 11
Keep changes scoped and preserve unrelated work.
- Distinctive: line 16
When editing README installation guidance, include every supported installer flavor from the maintained installer definitions.
- Distinctive: line 26
Check frontmatter, local links, and whitespace for changed packages.
- Common: line 30
Keep portable, Cursor, and Claude manifest identities/versions aligned.
- Common: line 31
Do not ship the repository discovery symlink, repo instructions, or credentials in host packages.
- Distinctive: line 32
Do not load every skill, force a lifecycle, or delegate solely because a skill mentions another skill.
- Distinctive: line 34
Commit, push, publish, contact others, or change live systems only within explicit authorization.
- Distinctive: line 36
Never dismiss GitHub reviews or approvals, including stale reviews, or enable automatic dismissal in repository settings or automation.
≈ “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 Instructions
This repository maintains shared skills for Cursor, Codex, Claude Code, OpenCode, and Copilot. These instructions apply to work in this repository; they are not automatically installed elsewhere.
- Keep canonical content in
skills/<name>/SKILL.mdand supporting resources in that skill's folder. Host discovery links point to this shared source. - Read
skills/anti-ai-writing/SKILL.mdbefore human-facing text. Useskills/using-skills/SKILL.mdto select relevant workflows when routing is needed. Read the selected skill; do not treat its description as the complete procedure. - Keep changes scoped and preserve unrelated work. Update affected relative links when names or responsibilities change. Do not add personal-only dependencies.
- For accepted plan implementation, use
skills/implement-plan/SKILL.mdto coordinate verification and final independent review. Planning remains planning-only; unavailable independent review must be reported incomplete rather than replaced by self-review. - When editing README installation guidance, include every supported installer flavor from the maintained installer definitions. Give each flavor a clearly labeled, collapsible section containing complete install, activation/verification, and update instructions, including supported Bash and PowerShell variants. Keep shared prerequisites and flavor navigation visible. Verify commands against the installer; preserve customization/conflict guidance and do not substitute one host's example or a link for another flavor's required steps. Use generic examples outside the flavor-specific instructions and do not privilege a particular editor or agent.
- Preserve the Jira writing reference's tested ADF contracts and examples unless a specific formatting change is requested and validated.
- Check frontmatter, local links, and whitespace for changed packages. Distinguish structural validation from actual host behavioral evidence.
- For tooling changes, run
go test -race ./...,go vet ./..., andgo run ./cmd/skillpack; build all five host packages into new scratch directories. Keep portable, Cursor, and Claude manifest identities/versions aligned. Do not ship the repository discovery symlink, repo instructions, or credentials in host packages. - Do not load every skill, force a lifecycle, or delegate solely because a skill mentions another skill. Avoid duplicate instructions and unnecessary approval rituals.
- Commit, push, publish, contact others, or change live systems only within explicit authorization. Preserve existing authorization and exact-draft publication gates.
- Never dismiss GitHub reviews or approvals, including stale reviews, or enable automatic dismissal in repository settings or automation. Use authorized re-review requests instead and preserve review history.
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.

