agentleFS
n24q02mn24q02m/claude-plugins/AGENTS.md

claude-plugins

AGENTS.md4 starsChanged 6 months agoApache-2.0View on GitHub

See README.md for end-user install. For development: Releases triggered manually via workflow_dispatch on cd.yml. Choose beta or stable. python-semantic-release handles version bump, CHANGELOG, tag, and GitHub Release. plugins/<repo>/tools.md is the only place a tool name is written down for the public. The marketplace listing and mcp.n24q02m.com/servers/<repo>/tools/ both read it, and nothing generates it: plugin.json carries no tool list, and the sync described above never copies .md files. A rename in a server is therefore invisible here until somebody types it,…

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

3 rules. About 1 is common, 1 is distinctive, and 1 is very specific to this project.

  • ≈1

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    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:

    Test coverage: ≥ 95% on internal modules

    line 27
  2. Common:

    No secrets in code, commits, or memory — use skret for secret retrieval

    line 28
  3. Very specific:

    Not every server can be read from a public runner.

    line 81

≈ “Common” is an estimate: at least 1 in 5 of the 36,368 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

Agent Collaboration

Quick reference

  • Repo: n24q02m/claude-plugins
  • Description: 9 MCP servers for Claude Code, Codex, and other AI coding agents.
  • License: Apache-2.0

Build & Test

See README.md for end-user install. For development:

mise run setup     # First-time dev environment
mise run lint      # Read-only lint + format check + type check
mise run test      # Run tests
mise run fix       # Auto-fix lint + format

Release

Releases triggered manually via workflow_dispatch on cd.yml. Choose beta or stable. python-semantic-release handles version bump, CHANGELOG, tag, and GitHub Release.

Conventions

  • Commits: only feat: and fix: prefixes (enforced by pre-commit commit-msg hook)
  • Test coverage: ≥ 95% on internal modules
  • No secrets in code, commits, or memory — use skret for secret retrieval
  • plugins/<repo>/ is generated. scripts/sync-plugins.sh replaces plugin.json, gemini-extension.json, skills/, hooks/, commands/ and chat.py from the source repository on every release, so a patch to those paths is erased at that plugin's next release — open it against the source repo instead. The setup prose alongside them is not generated and is edited here.
  • Decisions on proposals, accepted and rejected, are recorded in .jules/bolt.md, .jules/palette.md and .jules/sentinel.md. Read the relevant one before proposing a change, so a settled question is not reopened.

Renaming a tool in a server

plugins/<repo>/tools.md is the only place a tool name is written down for the public. The marketplace listing and mcp.n24q02m.com/servers/<repo>/tools/ both read it, and nothing generates it: plugin.json carries no tool list, and the sync described above never copies .md files. A rename in a server is therefore invisible here until somebody types it, and until they do, the docs name tools the server no longer answers to.

Who starts the update

The server repo, not this one. Every server already notifies this repo on release (repository_dispatch: plugin-sync), and the session that renames a tool opens the companion tools.md pull request here. This repo does not scrape server repos for prose — it verifies what it was handed, and sweeps weekly so a rename that nobody reported still surfaces.

When it has to land

Before the server's stable release, not after. Betas are safe to run ahead: cd.yml accepts only vX.Y.Z tags, so a prerelease never reaches the marketplace. A stable tag does, within minutes.

Order:

  1. Server repo: rename, merge to main, cut a beta.
  2. Here: open the tools.md pull request. Check it against the beta before the stable tag exists — python3 scripts/verify_tool_parity.py <repo> --version <beta-version>, or run the Tool Parity workflow with the same inputs.
  3. Merge it.
  4. Server repo: cut the stable release. The sync lands, and the parity check on main confirms the two agree.

Doing 4 before 3 is the failure this ordering exists to prevent. It is not fatal — the check on main goes red and the weekly sweep would catch it anyway — but between the release and the fix, the published docs are wrong.

What enforces it

scripts/verify_tool_parity.py asks the server instead of trusting the prose: it launches the exact command from plugin.json over stdio, calls MCP tools/list, and diffs those names against the ones tools.md declares (an ## <tool> heading, or the first column of a table headed Tool). .github/workflows/tool-parity.yml runs it on pull requests touching plugins/*/tools.md or plugins/*/.claude-plugin/plugin.json, on the matching pushes to main, weekly, and on demand.

Two things to know before changing it:

  • Not every server can be read from a public runner. wet-mcp, mnemo-mcp, better-godot-mcp and better-code-review-graph answer tools/list with no credentials at all. Others exit at startup unless a credential is present; for those the checker retries once with an obviously fake value, which is enough when the server only checks that something is set (better-notion-mcp). A server that validates its credential on startup — better-telegram-mcp reaches the Bot API and dies — cannot be read here at all. That case is reported as a warning, never a silent pass, and its drift check belongs in that server's own CI, which has real credentials. Real credentials are never needed here and must never be added to this repo's CI.
  • --declared-only runs the same comparison's static half with no network, which is enough to check that a tools.md edit is still parseable.

scripts/verify_docs_current.py remains the static neighbour: it checks that the required pages exist and that every userConfig key is documented. It does not look at tool names.

Download now (ZIP)

  • 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 28,693 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 repositoryOne other file it gives its agents.

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.