L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Productivité · SKILL.md

council

council

Send an idea to the Council of the Wise for multi-perspective feedback. Spawns sub-agents to analyze from multiple expert perspectives. Auto-discovers agent personas from agents/ folder.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: council description: Send an idea to the Council of the Wise for multi-perspective feedback. Spawns sub-agents to analyze from multiple expert perspectives. Auto-discovers agent personas from agents/ folder. version: 1.3.1 author: jeffaf credits: Inspired by Daniel Miessler's PAI (Personal AI Infrastructure). Architect, Engineer, and Artist agents adapted from PAI patterns. Devil's Advocate is an original creation.

Council of the Wise

Get multi-perspective feedback on your ideas from a panel of AI experts. Perfect for stress-testing business plans, project designs, content strategies, or major decisions.

Usage

"Send this to the council: [idea/plan/document]"
"Council of the wise: [topic]"
"Get the council's feedback on [thing]"

Council Members

The skill auto-discovers agent personas from {skill_folder}/agents/. Any .md file in that folder becomes a council member.

Default members:

  • DevilsAdvocate.md — Challenges assumptions, finds weaknesses, stress-tests
  • Architect.md — Designs systems, structure, high-level approach
  • Engineer.md — Implementation details, technical feasibility
  • Artist.md — Voice, style, presentation, user experience
  • Quant.md — Risk analysis, ROI, expected value, position sizing

Adding New Council Members

Simply add a new .md file to the agents/ folder:

# Add a security reviewer
echo "# Pentester\n\nYou analyze security implications..." > agents/Pentester.md

# Add a QA perspective  
echo "# QATester\n\nYou find edge cases..." > agents/QATester.md

The skill will automatically include any agents it finds. No config file needed.

Custom Agent Location (Optional)

If the user has custom PAI agents at ~/.claude/Agents/, those can be used instead:

  • Check if ~/.claude/Agents/ exists and has agent files
  • If yes, prefer custom agents from that directory
  • If no, use the bundled agents in this skill's agents/ folder

Process

  1. Receive the idea/topic from the user
  2. Discover available agents (scan agents/ folder or custom path)
  3. Send a loading message to the user: 🏛️ *The Council convenes...* (this takes 2-5 minutes)
  4. Spawn a sub-agent with 5-minute timeout using this task template:
Analyze this idea/plan from multiple expert perspectives.

**The Idea:**
[user's idea here]

**Your Task:**
Read and apply these agent perspectives from [AGENT_PATH]:
[List all discovered agents dynamically]

For each perspective:
1. Key insights (2-3 bullets)
2. Concerns or questions  
3. Recommendations

End with:
- **Synthesis** section combining best ideas and flagging critical decisions
- Note where council members **disagree** with each other — that's where the insight is
- Put **Synthesis first** (TL;DR at the top, details below)

Use the voice and personality defined in each agent file. Don't just list points — embody the perspective.
  1. Return the consolidated feedback to the user

Output Format

## 🏛️ Council of the Wise — [Topic]

### ⚖️ Synthesis (TL;DR)
[combined recommendation + key decisions needed]
[note where council members disagreed and why — that's the gold]

---

### 👹 Devil's Advocate
[challenges and risks — sharp, probing voice]

### 🏗️ Architect  
[structure and design — strategic, principled voice]

### 🛠️ Engineer
[implementation notes — practical, direct voice]

### 🎨 Artist
[voice and presentation — evocative, user-focused voice]

### 📊 Quant
[risk analysis, ROI, expected value — data-driven voice]

Configuration

No config file needed. The skill auto-discovers agents and uses sensible defaults:

  • Timeout: 5 minutes (enforced via sub-agent spawn)
  • Agents: All .md files in agents/ folder
  • Output: Markdown with synthesis and token usage
  • Model: Uses session default (can override via Clawdbot)

Notes

  • Council review takes 2-5 minutes depending on complexity
  • Timeout: 5 minutes enforced; on timeout returns partial results if available
  • Use for: business ideas, content plans, project designs, major decisions
  • Don't use for: quick questions, simple tasks, time-sensitive requests
  • The sub-agent consolidates all perspectives into a single response with Synthesis first
  • Add specialized agents for domain-specific analysis (security, legal, etc.)

Agent Implementation Notes

Trigger phrases: "send this to the council", "council of the wise", "get the council's feedback on"

When triggered:

  1. Send loading message: 🏛️ *The Council convenes...* (this takes 2-5 minutes)
  2. Spawn sub-agent with 5-minute timeout using the task template in Process section
  3. Return synthesized council report to user

Don't invoke for: Quick questions, time-sensitive tasks, simple decisions.

Skills enregistrées