keel-team
Skill9 starsChanged 51 days agoMITView on GitHub
OWNER-only setup wizard for a REAL human team (different people, different machines) — declare the owner (arm owner-guard), register members (@tags), seed lanes + author folders + the comprehension protocol, size the caps, pick the §6 contribution model + host wall. Packages the existing Multi-user playbook into one approved run; invents no new rules.
- Reads credentials
In this repository17 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- keel-adoptSkill
- keel-agent-team-createSkill
- keel-agent-team-startSkill
- keel-auditSkill
- keel-autopilotSkill
- keel-compactSkill
- keel-continueSkill
- keel-distillSkill
- keel-handoverSkill
- keel-phase-reviewSkill
- keel-pilotSkill
- keel-planSkill
- keel-researchSkill
- keel-statsSkill
- keel-tidySkill
- keel-updateSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
When: helpers are joining a solo project — different PEOPLE on different MACHINES sharing a repo/host.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill 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
/keel-team — set up the human team (owner-only)
When: helpers are joining a solo project — different PEOPLE on different MACHINES sharing a
repo/host. This wizard INSTALLS the existing Multi-user discipline (docs/steering.md "Multi-user",
rules §10.40–41, review-v2) in one approved pass — it invents nothing new. Same-machine AGENT
sessions are /keel-agent-team-create's job; the two coexist: human lanes and agent lanes share one
TASKS board, and setup authority is the owner's in both.
0. Owner gate
.claude/project-ownerexists andgit config user.name≠ its content → STOP ("team setup is the owner's — ask @<owner>").- Missing → this run DECLARES ownership: ask the owner's exact
git config user.name, write the file (one line). State the identity invariant for everyone: repo-localgit config user.name "<tag>", single token, byte-for-byte the repo's@tagspelling — a mismatch silently unarms every ownership mechanism (the reground hook nags when unset).
1. Interview the owner
- Members — each person's
@tag(= their git user.name) + a one-line area/responsibility- a task-id prefix: 2–4 lowercase letters, unique across the whole roster (agents included),
fixed for good. Their items run
<prefix>1 · <prefix>2 …, and the same id names their reports (reports/team/<@tag>/<id>_fix_<date>.md) and anyscratch/<id>/probe. Ids are never renamed or reused — permanent artifacts cite them (rules §9.32, §10.40). One case only: a series split acrossr1andR28makes every count wrong (field case).
- a task-id prefix: 2–4 lowercase letters, unique across the whole roster (agents included),
fixed for good. Their items run
- Contribution model (§6.17 — host reality decides the wall): same-repo branch→PR · fork→PR (developers Read-only) · no-remote/local-only. Walk the HOST-wall table (steering "Multi-user") and say which wall actually ENFORCES here — on a free private PERSONAL repo there is NO real wall: either move to a free org (Read + fork PRs) or accept discipline-only and write that into the team doc.
- Caps — propose
.claude/keel-capssized to headcount (5 people ≈HANDOVER=200·HANDOVER_BLOCKS=4·LESSONS=400·TASKS=250— field-tested values; the owner tunes). Pinned only on approval, never silently (§10.40). - CI shape (when PRs gate merges): §7.24 secret-less gating — a fork PR receives NO repo
secrets, so live-cred/network tests must SKIP, never fail. If bootstrap pruned
.github/, team scale-up is the moment the prune reverses — as a recorded ADR addendum.
2. Apply — each piece shown, approved, nothing silent
.claude/project-owner(if new) — arms owner-guard: governance files + non-ownermainpushes are blocked from then on.- TASKS:
## Nowgoes per-person (~2–3 items each), items carry@owner(+due:for sprint targets);## Reviewis born at the first developer handover and follows review-v2 (file-first, four states, routing = orchestrator/owner). reports/team/<@tag>/folder per member + their##section inreports/team/README.md(the index-as-todolist; controlled status vocabulary per that template)..claude/keel-capswith the approved numbers.- Comprehension protocol: the §10.41 gate already binds via rules; if the team wants the
standalone one-pager (field pattern: a protocol sheet under
reports/team/), seed it — task-specific self-explanatory questions, idempotent## Comprehension log, personally-run manual checks. docs/team.md(project-owned — the kit deliberately ships no template): roles table · fork/clone +git configonboarding steps · secret handoff via a safe channel · ritual etiquette · dev-local run knobs (own.envfrom.env.example,make run PORT=…overrides — "runs only on the owner's machine" is a bootstrap bug, §10.40).- rules.md §6 rewritten to the chosen contribution model, recorded as an ADR — with the §0a cascade: every doc that described the old flow gets updated in the same pass.
docs/architecture.md(§1.6): register the team structure.
3. Hand off
Give the owner a per-member onboarding message to send: clone/fork → repo-local
git config user.name "<tag>" → .env from .env.example → make test green locally → work ONLY
your @-items → deliver per review-v2 (evidence FILE + delivered index line; a chat summary is
not a delivery). Remind the owner what stays theirs: assignment (TASKS @tags), review routing
(§10.41), governance edits — and the scaling valves when it hurts: per-user handovers at ≥~4
concurrent writers; LESSONS is never split per user (sign lines instead).
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.

