agent-skills
CLAUDE.md2 starsChanged 18 days agoMITView on GitHub
Public collection of AI tooling from Xmartlabs. Two kinds of thing live here: - skills/: standalone Agent Skills, installed with npx skills add. They share one test suite in tests/ (bash tests/run-all.sh, with tests/helper.sh), and a skill may ship a GitHub Action under actions/<name>/ that runs it in CI. - plugins/: Claude Code plugins, listed in .claude-plugin/marketplace.json (the xmartlabs marketplace). Each plugin keeps everything it needs, docs, tests and release tooling included, inside its own directory. Rules for the whole…
In this repository2 other files it gives its agents.
- pr-evidenceSkill
- release-notesSkill
The file
We split this CLAUDE.md into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
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
- Very specific: line 15
Never push to main. Every change lands through a PR.
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.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
CLAUDE.md
Public collection of AI tooling from Xmartlabs. Two kinds of thing live here:
skills/: standalone Agent Skills, installed withnpx skills add. They share one test suite intests/(bash tests/run-all.sh, withtests/helper.sh), and a skill may ship a GitHub Action underactions/<name>/that runs it in CI.plugins/: Claude Code plugins, listed in.claude-plugin/marketplace.json(thexmartlabsmarketplace). Each plugin keeps everything it needs, docs, tests and release tooling included, inside its own directory.
Rules for the whole repository:
- Never push to
main. Every change lands through a PR. Tags are the only refs pushed directly, and they are namespaced by tool (tribunal/v110,tribunal-rust-v0.1.3). - Everything is written in English.
- This repository is public. No file may carry anything from a private
codebase: no client names, people, domains, file paths, identifiers or commit
ids, not in tests or comments either. Use neutral names (
acme/web-app, userdev,src/widgets.ts); a real repository a check runs against comes from the operator, as a flag. Private test data stays outside the repository (/review-harness-data/is ignored for that reason). Before merging anything ported from private work, runnode scripts/check-private-data.mjswith the private checkouts it came from (see its header).
Keeping the tools independent:
- A plugin lives in
plugins/<name>/and reads nothing outside it; shared code is copied, not imported. The skills are organised by their own team: they sharetests/, and an action inactions/<name>/may read its skill inskills/<name>/. An action for a plugin may read that plugin and the plugins it runs with (actions/tribunal/readsplugins/tribunal/andplugins/homosapiens/), and keeps its own tests in its directory. - Names are unique across the whole repository.
npx skills addlists the skills inskills/and the skills inside every plugin of the marketplace, so a plugin skill and a standalone skill cannot share a name. - A plugin skill that only works inside its plugin (it needs the plugin's
agents, scripts or workflows) sets
metadata: internal: truein its frontmatter, which hides it fromnpx skills add.review-prdoes. - A workflow that belongs to one tool triggers only on that tool's paths
(
on.pull_request.paths). Do not make a path-filtered job a required status check: a PR that does not touch those paths never runs it and stays blocked. - Releases and tags carry the tool's name as prefix. A release that is not the newest stable version of anything should be a prerelease, so it does not become the repository's "latest" release. Installers download by exact tag, never by "latest".
- Root-level config (
.claude/settings.jsonhooks,.gitignore) is shared by everyone who works here, so each entry must act only on its own tool's paths, as the two tribunal hooks do.
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.

