| name | council-review |
| description | Run an Agent Review Council over a product, workflow, codebase, architecture, prompt, agent package, or GTM artifact. Produces evidence-backed structured findings, scorecard, risks, decision record, and validation plan.
|
| version | 1 |
| app_version | 2026-06-22 |
| updated | "2026-06-22T00:00:00.000Z" |
| spec | ./SPEC.md |
Council Review Skill
Use this skill for full ARC runs that need coordinated evidence collection, independent
reviewer passes, adjudication, and schema-valid outputs. It populates a run skeleton;
it does not scaffold one.
When To Use
- Populate an ARC run skeleton (the 11-artifact bundle under
runs/<date>-<slug>/).
- Run independent specialist reviewers, then adjudicate and synthesize.
- Produce schema-valid findings, scorecard, risk register, decision record, validation plan.
- Review a run launched from an
intent.yaml (treat intent_id + objective as authoritative).
When NOT To Use
- Scaffolding a run skeleton, choosing/recommending a council, or authoring/editing
council or reviewer-role YAML — use the
council-run skill.
- Rendering
.claude/agents/<name>.md companion stubs — council-run.
- Project binding, council resolution, SAM pull/publish —
council-run.
Confidence Anchor
- Run skeleton + artifacts live under
runs/<date>-<slug>/ (created by arc run).
- Close a run with
uv run arc validate runs/<date>-<slug> (schema-valid gate).
arc run is a scaffolder: it writes empty artifacts + a next_steps prompt and does
NOT call an LLM. This skill is how a Claude Code agent populates the run in-session.
Workflow
Intent-first context: When reviewing a run launched from an intent.yaml, the
intent_id and objective fields in the intent are the authoritative statement of
what the run was trying to achieve. Reference these if the run's objective is ambiguous.
- Confirm target, objective, council definition, constraints, and required outputs.
- Read
references/run-workflow.md for the end-to-end council procedure.
- Read
references/output-contract.md before creating or validating run artifacts.
- Read
references/external-reviewers.md only when the run includes Codex, GitHub,
Copilot, LangGraph, or other non-Claude reviewers.
- Write artifacts under
runs/<date>-<slug>/.
- Validate with
uv run arc validate runs/<date>-<slug>.
Ground Rules
- Findings without evidence are hypotheses.
- High-severity findings require strong evidence or explicit uncertainty.
- Run independent reviewer passes before synthesis.
- Preserve accepted, rejected, disputed, and watchlist findings.
- Do not create tickets or durable memory without approval.
- Keep final output concise, but preserve the structured artifacts.
Do Not Say
- Do not say
arc run executes the review or calls an LLM — it only scaffolds an
empty run skeleton plus a next_steps prompt; this skill populates it in-session.
- Do not say the reviewer companion files (
.claude/agents/<name>.md) are dispatched
as separate agents or an agent-team at runtime — no per-reviewer-session dispatch
exists yet (deferred to I-5); they are reference docs the single agent reads.
- Do not say this skill scaffolds runs, recommends councils, or authors council/role
YAML — that is
council-run.
Key References
/Users/miethe/dev/homelab/development/agentic-research/.claude/skills/council-review/references/run-workflow.md/Users/miethe/dev/homelab/development/agentic-research/.claude/skills/council-review/references/output-contract.md/Users/miethe/dev/homelab/development/agentic-research/.claude/skills/council-review/references/external-reviewers.md/Users/miethe/dev/homelab/development/agentic-research/.claude/skills/council-review/SPEC.md
