โ† Back to AI & LLMs
AI & LLMs by @jeffaf

council-of-the-wise

Send an idea to the Council of the Wise

0
Source Code

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 โ€” Challenges assumptions, finds weaknesses, stress-tests
  • Architect.md โ€” Designs systems, structure, high-level approach
  • Engineer.md โ€” Implementation details, technical feasibility
  • Artist.md โ€” Voice, style, presentation, user experience
  • Quant.md โ€” 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: ๐Ÿ›๏ธ *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 โ€” 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 โ€” embody the perspective.
  1. Return the consolidated feedback to the user

Output Format

## ๐Ÿ›๏ธ Council of the Wise โ€” [Topic]

### โš–๏ธ Synthesis (TL;DR)
[combined recommendation + key decisions needed]
[note where council members disagreed and why โ€” that's the gold]

---

### ๐Ÿ‘น Devil's Advocate
[challenges and risks โ€” sharp, probing voice]

### ๐Ÿ—๏ธ Architect  
[structure and design โ€” strategic, principled voice]

### ๐Ÿ› ๏ธ Engineer
[implementation notes โ€” practical, direct voice]

### ๐ŸŽจ Artist
[voice and presentation โ€” evocative, user-focused voice]

### ๐Ÿ“Š Quant
[risk analysis, ROI, expected value โ€” 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: ๐Ÿ›๏ธ *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.