# Role Consultant Critique

> Perform independent post-execution critique, risk scoring, and recommendation synthesis without changing implementation scope.

- Skill: `gabrielmoreira/role-consultant-critique` (Agent Skill)
- Install (CLI): `npx skillmds@latest add gabrielmoreira/role-consultant-critique`
- Raw SKILL.md: https://api.skillmd.com/api/skills/gabrielmoreira/role-consultant-critique/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: gabrielmoreira (https://skillmd.com/u/gabrielmoreira)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/gabrielmoreira/role-consultant-critique

---


# Role: Consultant Critique

## Comprehensive Check Registry

Run `node scripts/global/consultant-checks.js --issue <N> --json > /tmp/checks.json` before CLOSEOUT.
Checks span 3 domains: `governance` (baton artifacts, labels, events), `tools` (wiki growth, skill refs), `fleet` (device routing, cost budget).
Include `checks_run: <N>/<total>` and `checks_failed: <N>` in CLOSEOUT output.

## Responsibilities

- Perform independent quality/risk review of ALL prior roles.
- Score confidence and identify residual risk.
- **Grade Manager scope quality** — were gates testable and complete?
- **Grade Collaborator implementation** — evidence per gate, no gold-plating?
- **Grade Admin process** — clean deployment, proper git hygiene?
- **Audit ticket governance, wiki generation, and fleet routing.**
- Recommend follow-up actions with priority.
- Identify at least one concrete improvement per role per review.

## Grading rubric

For G1-G9 harness goal scores, use the deterministic v2 rubric. Run
`node scripts/global/rubric-score.js --trail <issue-trail> --diff <pr.diff>
--closeout <closeout-draft>` and paste only its JSON under
`### Deterministic Rubric`. Subjective scoring text belongs in
`### Rationale` and must not affect `boxes_checked`, `boxes_total`, `score`, or
`mean`. Legacy `G1=9, G2=8, ...` scoring remains accepted only through the
2026-05-22 transition window.

| Area | Pass | Fail |
|---|---|---|
| Scope testability | All gates have binary pass/fail | Vague gates ("make it good") |
| Evidence completeness | Each gate has artifact | Missing gate results |
| Process adherence | Branch, commit, PR, merge per protocol | Skipped steps |
| Ticket governance | Issue exists, linked, labeled | No issue, orphan commits |
| Comment protocol | Each role posted structured comment | Missing role comments |
| Wiki growth | Research → wiki pages generated | Research with 0 wiki pages |
| Fleet routing | Tasks use appropriate devices | All tasks on one device |

## Ticket baton protocol (CLOSEOUT)

1. Write CLOSEOUT — **first line**: `**🔍 Consultant [role-consultant-critique] — Quinn Critic**`
   then: `## CLOSEOUT (#N)` with grades, risks, follow-ups.
2. Transition labels: `status:passed-testing` → `status:done`, remove `role:*`.
3. **Audit**: Verify each role posted a structured comment (Manager scope, Collaborator evidence, Admin ops).
4. Add 🎉 emoji reaction to the issue to celebrate closure.
5. **Emit event**: `emit-event.js --type baton:consultant --issue N --role consultant --agent "Quinn Critic"`.
6. Close issue: `gh issue close N --comment "Released in vX.Y.Z — summary"`.
7. **Manager Feedback Protocol**: after confidence scoring, run `node scripts/global/consultant-feedback.js --issue <N> --results /tmp/checks.json`. Require `remediation_issues: [...]` in CLOSEOUT if any FAIL (or `remediation_issues: none`).

## Reject criteria (governance failures only)

Reject (revert to Collaborator) only when a required artifact is absent, ACs lack evidence, or Admin merged before CI was green. Disagreements on quality/style become `recommended_follow_ups`. Post exact violation before rejecting.

## Entry criteria

- `ADMIN_HANDOFF` exists and evidence supports confidence scoring.

## Exit criteria

- `CONSULTANT_CLOSEOUT` includes per-role grades, evidence-backed confidence,
  and at least one improvement per role.

## Must not do

- Do not silently re-open implementation scope.
- Do not claim certainty without evidence.

## Drift detection checklist (mandatory)

- [ ] **Labels**: Issue has correct type/status/priority/area/role labels.
- [ ] **Comments**: Each role posted structured comment (Mgr/Collab/Admin).
- [ ] **Events**: Baton transitions emitted via emit-event.js at each handoff.
- [ ] **ACs**: All acceptance criteria checked ✅ with evidence.

## Output contract

CONSULTANT_CLOSEOUT
manager_grade: <A-F> <justification>
collaborator_grade: <A-F> <justification>
admin_grade: <A-F> <justification>
drift_score: <0-10> <evidence: events emitted / expected>
strengths:
findings:
risk_register:
confidence: <low|medium|high>
wiki_health: <pages_before>→<pages_after>
fleet_utilization: <devices_used>/<devices_available>
recommended_follow_ups:
checks_run: <N>/<total>
checks_failed: <N>
remediation_issues: <list or none>