agentleFS
Sign inStart free
duu261duu261/agent-skills/AGENTS.md

agent-skills

AGENTS.md2 starsChanged 10 days agoMITView on GitHub

This repo publishes Agent Skills, one per folder: skills/<name>/SKILL.md. The root README is the catalog; each skill folder has its own README. Skills come from real work in agent sessions, then a human judges and rewrites them. An agent may draft a public version; the maintainer decides what lands.

  • 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 31,700 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 repository5 other files it gives its agents.

The file

We split this AGENTS.md into the rules it gives an agent

9 rules. About 0 are common, 7 are distinctive, and 2 are very specific to this project.

  • ≈0

    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
  • could be automatic: a setting could enforce it
  1. Distinctive:

    Start from the working skill.

    line 18
  2. Very specific:

    Keep claims true. Say "learned" or "refined", not "evaluated", unless a controlled run backs it.

    line 21
  3. Very specific:

    Add the skill's row to both root READMEs, and its plugin entry to .claude-plugin/marketplace.json.

    line 22
  4. Distinctive:

    Add a rule only when repeated real use shows the failure.

    line 27
  5. Distinctive:

    Every rule says why. A bare do/don't list fails on cases it does not name.

    line 28
  6. Distinctive:

    When a rule adds strictness, check it does not add ceremony.

    line 29
  7. Distinctive:

    Keep EN and VI READMEs in step.

    line 30
  8. Distinctive:

    Before pushing, scan staged text for the operator details listed above.

    line 35
  9. Distinctive:

    No force-push to main.

    line 40

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

AGENTS.md

This repo publishes Agent Skills, one per folder: skills/<name>/SKILL.md. The root README is the catalog; each skill folder has its own README.

Layout

  • Flat: skills/<name>/, never nested categories. Some installers scan deeper, but Hermes Agent taps only scan one level under skills/, so flat is the only layout every channel finds.
  • name in frontmatter equals the folder name: lowercase, hyphens, 64 characters or fewer, no "claude" or "anthropic".
  • description says what the skill does and when to use it, 1,024 characters or fewer. It is what an agent reads to decide whether to load the skill.
  • Frontmatter: name, description, license: MIT. Nothing harness-specific.
  • SKILL.md stays under 500 lines. Put long detail in references/, helpers in scripts/.
  • Each skill folder has README.md and README.vi.md: what failure it fixes, where it came from, install.

Adding a skill

Skills come from real work in agent sessions, then a human judges and rewrites them. An agent may draft a public version; the maintainer decides what lands.

  1. Start from the working skill. Keep its procedure and its pitfalls; those are the parts that real sessions earned.
  2. Make it harness-neutral. Describe actions ("read the file", "search session history", "use a read-only subagent if your harness has one"), not one harness's tool names. Harness-specific mapping goes in references/.
  3. Strip the operator: no hostnames, IPs, home paths, usernames, product, pool, or account names, private incident stories, or rules that only make sense on one setup.
  4. Keep claims true. Say "learned" or "refined", not "evaluated", unless a controlled run backs it. Credit models by exact ID (claude-opus-5-5, not "Opus").
  5. Add the skill's row to both root READMEs, and its plugin entry to .claude-plugin/marketplace.json.
  6. Validate, then get an independent review before pushing.

Editing a skill

  • Add a rule only when repeated real use shows the failure. One incident or one model's guess is not enough.
  • Every rule says why. A bare do/don't list fails on cases it does not name.
  • When a rule adds strictness, check it does not add ceremony. Asking twice costs the user a turn too.
  • Keep EN and VI READMEs in step. English is the source of truth.

Checks

  • Validate every skill: for d in skills/*/; do uvx --from skills-ref==0.1.1 agentskills validate "$d"; done. CI runs the same loop on every push.
  • Before pushing, scan staged text for the operator details listed above.

Git

  • Conventional commits, imperative subject of 50 characters or fewer.
  • No force-push to main.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.

Check your setup, free