skills
CLAUDE.md2 starsChanged 5 days agoMITView on GitHub
skills/agent-kit/agent-team/roles.yaml is read by a different, public repo, so a structural change here can silently break it: uzi (https://github.com/vtmocanu/uzi) ships a subset of these roles as builtin agents. It vendors a distilled name -> version snapshot and runs a weekly bot that re-reads this file to detect drift, parsing only each role's name: and version: fields (and the top-level roles: list shape). It does NOT list which roles it ships here (that stays in uzi, so this note never goes…
In this repository14 other files it gives its agents.
- agent-teamSkill
- away-modeSkill
- cicd-expertSkill
- claude-permissionsSkill
- clodexSkill
- doneSkill
- generate-dockerfileSkill
- reflectSkill
- session-peersSkill
- skill-makerSkill
- skipperSkill
- status-checkSkill
- token-auditSkill
- upgrade-advisorSkill
The file
Files
CLAUDE.md
agent-team/roles.yaml has an external consumer: uzi
skills/agent-kit/agent-team/roles.yaml is read by a different, public repo,
so a structural change here can silently break it:
uzi (https://github.com/vtmocanu/uzi) ships a subset of these roles as builtin
agents. It vendors a distilled name -> version snapshot and runs a weekly bot
that re-reads this file to detect drift, parsing only each role's name: and
version: fields (and the top-level roles: list shape). It does NOT list which
roles it ships here (that stays in uzi, so this note never goes stale).
- Bumping a
version:is fine and is the intended signal: uzi's bot opens a manifest-bump PR that reddens uzi's drift test until a maintainer syncs the new body. uzi's builtins are verbatim copies of the publishedproduct-agents/bodies (uzi-only rules live in uzi's worker prompt). A uzi admin can also syncproduct-agents/straight into a live instance. - Keep every body runtime-neutral. It must read correctly on a laptop and
on a uzi worker: name "the scratch directory your runtime provides" rather than
a path, allow a detached checkout only "where your runtime permits one", and
never claim an ignored file cannot be staged. uzi's
api/internal/agenttmpl/scratch_guidance_test.golists the phrases a body must not contain. - Adding/removing a role needs no uzi change (uzi tracks a fixed subset and adds builtins by hand).
- Renaming/restructuring
name:/version:(or theroles:shape) breaks uzi's parser. Update uzi in the same effort:api/internal/agenttmpl/library/(vendored manifest + runbook),scripts/refresh-role-manifest.sh(the bot's parser), andapi/internal/agenttmpl/builtins.go(theversion:frontmatter parser).
Rule of thumb: a role's content is local to this repo; a change to the
shape of name:/version: is a cross-repo change that must touch uzi too.
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.

