| name | council |
| description | Run an LLM council review with 6 diverse subagent personalities. Use for architecture review, implementation plan critique, risk assessment, or quality audit. Produces a structured council report with ratings, MUST-FIX items, and a consolidated verdict. |
Council Review Skill
Convene a multi-perspective LLM council to review a project, plan, or implementation. Six subagents with distinct personalities independently audit the target and return structured findings. Their reports are compiled into a single council review document with a unified verdict, MUST-FIX list, and revised roadmap.
When to Use
- After completing a major implementation phase — get independent critique
- Before starting a new project phase — validate the plan
- When the user asks for a "review," "audit," "second opinion," or "council"
- When the user says "what are we missing?" or "is this the right approach?"
- During architecture decision-making — get diverse perspectives
Council Members
Launch all six in parallel. Each is an Explore subagent with a distinct personality and scope.
| Role | Personality | Focus |
|---|
| The Architect | Systems thinker, values clean abstractions and long-term maintainability | Architecture, coupling, separation of concerns, abstraction levels |
| The Pragmatist | Battle-scarred shipper, values working software over perfect design | Scope realism, shippability, timeline, what to cut |
| The Tester | Quality obsessive, believes untested code is broken | Test coverage, edge cases, failure modes, what's untested |
| The Guardian | Privacy-first, treats user data as radioactive | Data hygiene, local-first enforcement, leak vectors, retention |
| The Perf Engineer | Optimizes everything, thinks in nanoseconds | CPU/memory/GC pressure, target hardware readiness, hidden allocations |
| The User Advocate | Represents the actual human user | UX, CLI design, on-ramp friction, debuggability, real-world usefulness |
Prompt Template
For each subagent, use this structure (customize the file list per role):
You are "[ROLE]" — [PERSONALITY DESCRIPTION].
Your task: Review the [PROJECT NAME] and provide critical findings.
Read these files in full:
- [FILE LIST — tailored to role]
Then provide a concise review covering:
1. [QUESTION 1]
2. [QUESTION 2]
3. [QUESTION 3]
4. [QUESTION 4]
5. [QUESTION 5]
6. Rate [DIMENSION] 1-10 and give one MUST-FIX item.
Be [TONE INSTRUCTION]. Return your findings as structured markdown.
Compilation
After all six return, compile into a single council review document at:
Data/Pages/Council/council-review-{DD-MM-YY-HH-MM-SS}.md
The compiled document should contain:
- Header — date, scope, council composition table with ratings
- Unanimous Findings — items all 6 agreed on (or 5+)
- Per-Member Findings — each member's report condensed
- Consolidated MUST-FIX List — priority-ordered table
- Revised Roadmap/Plan — if reordering is recommended
- Verdict — 1-paragraph summary
Additionally, update or create ROADMAP.md in the repo root with the refined plan (pure roadmap, no mention of council methodology).
Customization
- Change the council size: Use 3 for quick reviews, 6 for thorough audits
- Swap roles: Replace any role with a domain-specific one (e.g., "ML Engineer" for model review)
- Adjust depth: Add "thorough" for critical milestones, "quick" for routine checkpoints
- Focus scope: Limit file list to specific modules for targeted reviews
Example Usage
User says: "Review our implementation so far."
→ Launch 6 subagents with tailored file lists.
→ Compile findings into council-review-{timestamp}.md.
→ Update ROADMAP.md with refined plan.
→ Present the consolidated MUST-FIX list to the user.