agentleFS
Sign inStart free
luongnv89luongnv89/skills/AGENTS.md

skills

AGENTS.md145 starsChanged 6 days agoMITView on GitHub

  • Reads credentials
  • Deletes or force-pushes

The file

Show its rules and text

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

14 rules. About 0 are common, 6 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    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. Very specific:

    Check every meta/<name>/evals/evals.json: python3 scripts/validate-evals.py

    line 11
  2. Very specific:

    Never: edit installed skill copies (~/.claude/skills/, ~/.agents/skills/) from this repo

    line 17
  3. Very specific:

    Never: hand-edit dist/; regenerate it with package_skill.py

    line 18
  4. Very specific:

    Never: commit *-workspace/, .asm-improver/, .claude/scheduled_tasks.lock, secrets, .env, or /credentials*

    line 19
  5. Very specific:

    Never: add non-runtime files under skills/ (the plugin folder is capped at 512 files).

    line 20
  6. Distinctive:

    Never: commit or push without an explicit user request (drafting the message is fine)

    line 21
  7. Very specific:

    Ask first: destructive ops (rm -rf, git reset --hard, force-push, branch delete)

    line 22
  8. Distinctive:

    Ask first: a change that touches more than one skill

    line 23
  9. Very specific:

    Every SKILL.md edit bumps metadata.version (patch for wording, minor for a new capability, major for a restructure), and the skill's row in …

    line 26
  10. Very specific:

    Before creating a skill, grep skills/*/SKILL.md for one that already owns the behavior

    line 34
  11. Distinctive:

    Never re-read files you just wrote or edited.

    line 49
  12. Distinctive:

    Never re-run commands to "verify" unless the outcome was uncertain.

    line 50
  13. Distinctive:

    Don't echo back large blocks of code or file contents unless asked.

    line 51
  14. Distinctive:

    If a task needs 1 tool call, don't use 3.

    line 54

≈ “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 — Agent Skills Catalog

Project

Skill definitions live in skills/<name>/. Installed copies under ~/.claude/skills/ and ~/.agents/skills/ are the runtime, not the source: edit only this repo.

Commands

  • Validate one skill: python3 ~/.claude/skills/skill-creator/scripts/quick_validate.py skills/<name>
  • Score one skill: asm eval skills/<name> (--json for per-category scores)
  • Package: python3 ~/.claude/skills/skill-creator/scripts/package_skill.py skills/<name> (writes dist/)
  • Scaffold: python3 ~/.claude/skills/skill-creator/scripts/init_skill.py <name> --path skills/
  • Check every meta/<name>/evals/evals.json: python3 scripts/validate-evals.py
  • Trigger evals (tests the installed copy): python3 scripts/run-skill-evals.py <name> --runs 3
  • Local install: bash install.sh. It's an interactive TUI that installs for real, takes no flags, and needs a TTY
  • There is no npm, pnpm, make, or pytest here. Don't invent test commands.

Constraints

  • Never: edit installed skill copies (~/.claude/skills/, ~/.agents/skills/) from this repo
  • Never: hand-edit dist/; regenerate it with package_skill.py
  • Never: commit *-workspace/, .asm-improver/, .claude/scheduled_tasks.lock, secrets, .env, or **/credentials*
  • Never: add non-runtime files under skills/ (the plugin folder is capped at 512 files). A skill's docs/, evals/ and tests/ go in meta/<name>/, mirroring the skills/ path; eval fixtures go in eval-fixtures/<name>/
  • Never: commit or push without an explicit user request (drafting the message is fine)
  • Ask first: destructive ops (rm -rf, git reset --hard, force-push, branch delete)
  • Ask first: a change that touches more than one skill

Conventions

  • Every SKILL.md edit bumps metadata.version (patch for wording, minor for a new capability, major for a restructure), and the skill's row in the root README.md shows the same version
  • Quote any frontmatter string containing : # - < > | , & ? !
  • name equals the skill's own directory name, suite children included (skills/<umbrella>/<child>/). Each suite SKILL.md validates and versions independently
  • SKILL.md stays under 500 lines and 3000 words; overflow goes to references/
  • meta/<name>/docs/README.md opens with the AI-skip HTML comment (copy it from any existing skill)
  • A skill that mutates a git repo has a ## Repo Sync Before Edits (mandatory) section. A skill that invokes another skill has a ## Dependency Preflight (mandatory) section
  • Small fix: make a minimal, targeted diff. Never rewrite a SKILL.md for a one-line change or refactor opportunistically
  • New skill: run init_skill.py, fill in SKILL.md, move the scaffolded docs/README.md to meta/<name>/docs/README.md, validate, then add a CHANGELOG entry under ## Unreleased
  • Before creating a skill, grep skills/*/SKILL.md for one that already owns the behavior
  • Renaming a skill: change the frontmatter name, the directory, and the README.md catalog references in the same commit
  • Commits use Conventional Commits (feat:, fix:, chore:, docs:). Breaking changes go in the body, not the title

Done when

  • quick_validate.py exits 0 for every skill you touched
  • asm eval gives an overall score above 85, and at least 8 in each of: structure, description, prompt-engineering, context-efficiency, safety, testability, naming. license scores 5 across the whole catalog by design, so never add per-skill LICENSE files
  • python3 scripts/validate-evals.py passes whenever a meta/*/evals/evals.json changed
  • python3 tests/test-lstack-claude-packaging.py passes (it fails when skills/ exceeds 512 files)

Read when needed

  • Contribution workflow and the eval runner: CONTRIBUTING.md
  • What doesn't belong in agent instruction files: skills/agent-config/references/anti-patterns.md

Token Efficiency

  • Never re-read files you just wrote or edited. You know the contents.
  • Never re-run commands to "verify" unless the outcome was uncertain.
  • Don't echo back large blocks of code or file contents unless asked.
  • Batch related edits into single operations. Don't make 5 edits when 1 handles it.
  • Report results and blockers plainly; skip filler like "I'll continue...".
  • If a task needs 1 tool call, don't use 3.

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