liulab-skills
AGENTS.md1 starsChanged 14 days agoMITView on GitHub
How any agent works on this repo. CLAUDE.md is a symlink to this file, so edit this one. The repo is the Liu Lab's public set of Agent Skills, shipped as one Claude Code plugin: the repo root is plugin lab in marketplace liulab, with one skill per skills/<name>/. Every file is public, and the whole tree ships to every lab machine, so every merged change is a release. - No connection details, anywhere. No IP literals other than 0.0.0.0…
- Reads credentials
- Commits and pushes
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 repository36 other files it gives its agents.
- broSkill
- codebase-designSkill
- code-reviewSkill
- diagnosing-bugsSkill
- domain-modelingSkill
- grillingSkill
- grill-meSkill
- grill-with-docsSkill
- handoffSkill
- hq-modeSkill
- implementSkill
- implement-specSkill
- improve-codebase-architectureSkill
- ketchupSkill
- lab-containersSkill
- lab-edisonSkill
- lab-hpcSkill
- lab-jupyterSkill
- lab-setupSkill
- prototypeSkill
- prSkill
- researchSkill
- retroSkill
- setup-matt-pocock-skillsSkill
- tddSkill
- teachSkill
- technical-writingSkill
- to-questionnaireSkill
- to-specSkill
- to-ticketsSkill
- triageSkill
- unslopSkill
- unslop-zhSkill
- wait-whatSkill
- wayfinderSkill
- writing-for-agentsSkill
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
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
- Very specific: line 12
No connection details, anywhere: No IP literals other than 0.0.0.0 and 127.0.0.1, and no usernames, hostnames, key names, key material or …
- Distinctive: line 31
Use the development agent for evals.
- Distinctive: line 37
Every merged change is a release, a test-only change included:
≈ “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
AGENTS.md
How any agent works on this repo. CLAUDE.md is a symlink to this file, so edit this one.
The repo is the Liu Lab's public set of Agent Skills, shipped as one Claude Code plugin: the
repo root is plugin lab in marketplace liulab, with one skill per skills/<name>/. Every
file is public, and the whole tree ships to every lab machine, so every merged change is a
release.
Hard rules
- No connection details, anywhere. No IP literals other than
0.0.0.0and127.0.0.1, and no usernames, hostnames, key names, key material or passwords: not in a file, a commit, an issue or a PR body. Name a host only by its lab ssh alias. A skill reads everything else from~/.ssh/configand~/.claude/lab/personal.mdwhen it runs. - Run the gate before every commit. The no-secrets sweep looks for the usernames and hosts in your own ssh config, so CI's run of it is half-blind and yours is the real one.
- A tracked skill's upstream text only gains lines. A lab edit adds; it never cuts,
rewords or renames, even to pass the gate. The tracked skills are the
itemsinupstream/mattpocock/pin.json.
Checks
| Lane | Command | What it proves |
| Gate | pixi run check | Every static check, unit test and tests/lint.sh; what CI runs on each PR. All failures print in one run, so read to the bottom |
| Docs | pixi run docs-build | The site builds strictly. CI runs it too |
| Preflight | bash tests/preflight.sh [--live] | This machine is set up for the clusters. Local only |
| Evals | bash tests/eval.sh --agent <agent> --only <case> | A skill's behaviour, in a real headless session against the installed plugin: the runner installs and verifies a private snapshot first. Costs tokens: ask the user before --all or --live |
Use the development agent for evals. Manage only its installation. Carry authorization through review, release and cleanup; batch routine steps and reuse approved prefixes. Request only sandbox permissions the environment requires.
Release
Every merged change is a release, a test-only change included:
- In one commit, bump
versionin.claude-plugin/plugin.json(CalVer,YYYY.M.PATCH) and add aCHANGELOG.mdentry, which is the release notes. - Once it merges, tag
mainasv<version>andgit push origin main --tags. No GitHub Release.
Read next
| When | Read |
| Before any change | GLOSSARY.md, then the docs/adr/ records on that area |
| Writing anything | docs/agents/writing.md |
| Adding or editing a skill | docs/agents/skills.md |
| Placing a fact, a term or a decision | docs/agents/domain.md |
| An adopted skill, or an upstream sync | upstream/README.md |
| A test or an eval case | tests/README.md. A new gate step goes in the check-static task in pyproject.toml, which CI calls by name |
| Filing, labelling or working an issue | The lab defaults: skills/lab-setup/references/defaults.md |
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.