Governance review
Skill sumitake/agent-collab/plugins/agent-collab/skills/governance-review
Use when the operator says "governance review," "high-stakes review," "tiebreaker," or "second opinion." Also offer this proactively when reviewer-family independence must be enforced.From its SKILL.md
npx -y skills add sumitake/agent-collab --skill governance-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
3 things to look at
- 29 days oldThe repository was created 29 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
2.2 KB, 398 tokens by cl100k_base, as published. Nobody here has run it
Run an independent governance review
Resolve the plugin root from this loaded file and invoke only
python3 "<plugin-root>/coordinator.py" with a bounded governance request that
includes captured artifact content and author-model provenance.
Read the Coordinator request schema in <plugin-root>/README.md before
constructing the request; never invent fields or route/action pairs.
Workflow
Resolve both the active primary family and the immutable artifact-author
family. If either is unknown, fail closed with unknown_family; preserve the
error detail that identifies which snapshot could not be proved. Exclude both
families from the reviewer panel, tiebreaker, fallback, and retry set.
Seal the request as read-only authority. A governance review must
never enter a mutation-capable worker route. If no independent managed reviewer
is eligible, return unavailable; a specifically selected same-family route is
same_family_blocked. Do not weaken independence or use a same-family result.
The only eligible advertised read-only actions are Gemini governance, Codex
advisory, and Grok governance. OpenCode plan, every build role, and
the output-only composer/codegen compatibility route are never governance-
review candidates.
Gemini governance is a distinct broker-only action, never an advisory alias. Accept its result only when the public client validates the complete artifact- bound governance proof, including canonical real-HOME containment, controlling PTY, state lease, cleanup, exact model/effort, response hash, and proof hash. Gemini advisory or long-context output is not governance evidence.
Claude participation can occur only through a separately configured host-owned
async transport after readiness is observed. The public coordinator neither
sends nor accepts governance over inbox/async; never create a synchronous
Claude route. Execution mechanics stay inside the verified co-packaged native
artifact, with typed result and provenance preserved.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.