AI Council

Summon a council of specialized AI advisors to deliberate on your ideas, proposals, or decisions from multiple perspectives.

Overview

The AI Council launches parallel sub-agents, each with a distinct analytical lens. Council members analyze independently, document their reasoning to a persistent file, then you can optionally have them debate to surface the strongest arguments.

How It Works

digraph council_flow {
    rankdir=TB;
    "User invokes /ai-council" [shape=box];
    "Discover agents via Glob" [shape=box];
    "Present agent selection (AskUserQuestion)" [shape=box];
    "User selects agents + provides proposal" [shape=box];
    "Launch selected agents in parallel (Task tool)" [shape=box];
    "Write all perspectives to AI-COUNCIL.md" [shape=box];
    "Present results to user" [shape=box];
    "Optional: Trigger debate round" [shape=diamond];
    "Agents challenge each other" [shape=box];
    "Append debate to AI-COUNCIL.md" [shape=box];
    "Synthesis and recommendations" [shape=box];

    "User invokes /ai-council" -> "Discover agents via Glob";
    "Discover agents via Glob" -> "Present agent selection (AskUserQuestion)";
    "Present agent selection (AskUserQuestion)" -> "User selects agents + provides proposal";
    "User selects agents + provides proposal" -> "Launch selected agents in parallel (Task tool)";
    "Launch selected agents in parallel (Task tool)" -> "Write all perspectives to AI-COUNCIL.md";
    "Write all perspectives to AI-COUNCIL.md" -> "Present results to user";
    "Present results to user" -> "Optional: Trigger debate round";
    "Optional: Trigger debate round" -> "Agents challenge each other" [label="yes"];
    "Optional: Trigger debate round" -> "Synthesis and recommendations" [label="no"];
    "Agents challenge each other" -> "Append debate to AI-COUNCIL.md";
    "Append debate to AI-COUNCIL.md" -> "Synthesis and recommendations";
}

Invocation Protocol

When this skill is invoked:

Step 1: Discover and Select Agents

1. Use the Glob tool with pattern agents/*.md (relative to the plugin root) to discover all available agent files.
2. Read each discovered file to extract the YAML frontmatter fields: name, description, model, and color.
3. Build AskUserQuestion options dynamically from the discovered agents. For each agent:
   • label: Convert the name field from kebab-case to Title Case (e.g., devils-advocate → Devil's Advocate)
   • description: Use a short summary extracted from the first sentence of the description field
4. Present the multi-select agent chooser:

AskUserQuestion:
  question: "Which council members should deliberate on this?"
  header: "Council"
  multiSelect: true
  options:
    [dynamically built from discovered agents]

Step 2: Gather the Proposal

If not already provided, ask the user for:

• Proposal: What idea/decision needs analysis?
• Context: Relevant background, constraints, goals
• Key Questions: Specific aspects to focus on (optional)

Step 3: Launch Agents in Parallel

For each selected agent:

1. Read the agent .md file from the agents/ directory.
2. Extract the model from frontmatter (default to opus if not specified).
3. Use the file body (everything after the frontmatter closing ---) as the agent’s system prompt.
4. Launch via the Task tool.

Critical: Launch ALL selected agents in a single message with multiple Task tool calls for true parallel execution.

Example Task invocation pattern:

Task:
  subagent_type: general-purpose
  model: [from agent frontmatter, default opus]
  description: "[Agent Name] council analysis"
  prompt: |
    You are the [Agent Name]. [Insert agent system prompt body here]

    **Proposal to Analyze:**
    [User's proposal]

    **Context:**
    [User's context]

    **Your Task:**
    Provide your perspective-aligned analysis following your output structure.
    Be specific and actionable. Include confidence levels for key claims.

Always use the model specified in the agent’s frontmatter (defaulting to opus) for council agents to ensure the highest quality analysis.

Step 4: Write Deliberation Log

IMPORTANT: After collecting all agent responses, write them to AI-COUNCIL.md in the current working directory.

Use the following format:

# AI Council Deliberation

**Date**: [ISO timestamp]
**Proposal**: [Brief title or summary]

---

## Proposal Under Review

[Full proposal text and context provided by user]

**Key Questions**:
1. [Question 1]
2. [Question 2]
...

---

## Council Perspectives

### [Agent Name]

[Full analysis from this agent]

**Confidence Level**: [X]%

---

[Continue for each selected agent...]

---

## Areas of Agreement

- [Point where multiple agents agree]
- [Another consensus point]

## Areas of Disagreement

- [Contested claim]: [Agent A] says X, [Agent B] says Y

---

## Debate Round (if conducted)

### [Agent Name] challenges [Other Agent]

> "[Specific claim being challenged]"

[Challenge and counter-argument]

[Continue for all challenges...]

---

## Synthesis

### Consensus
[Where all/most agents agree]

### Key Risks
[Top concerns across all perspectives]

### Key Opportunities
[Top upsides identified]

### Recommended Next Steps
1. [Action item]
2. [Action item]
...

---

*Council session complete.*

Step 5: Present Results

After writing to AI-COUNCIL.md, present a summary to the user:

• Highlight areas of agreement across agents
• Flag areas of disagreement for potential debate
• Mention that full deliberation is saved to AI-COUNCIL.md

Step 6: Optional Debate Round

Ask if the user wants agents to challenge each other:

AskUserQuestion:
  question: "Should the council debate their positions?"
  header: "Debate"
  options:
    - label: "Yes, have them challenge each other"
      description: "Agents will directly confront opposing claims"
    - label: "No, proceed to synthesis"
      description: "Skip to final recommendations"

If debate requested:

1. Launch a second round where each agent challenges specific claims from others
2. Append the debate results to AI-COUNCIL.md (don’t overwrite)
3. Update the synthesis section

Step 7: Final Synthesis

Provide a unified summary and ensure AI-COUNCIL.md contains:

• Consensus: Where all/most agents agree
• Contested: Where significant disagreement exists
• Key Risks: Top concerns across all perspectives
• Key Opportunities: Top upsides identified
• Recommended Next Steps: Actionable path forward

Extensibility

The council automatically discovers all agent files in the agents/ directory. To add a new agent, either:

• Use /create-agent to generate one from a natural language description
• Manually create an .md file in agents/ following the established frontmatter + system prompt pattern

New agents are immediately available the next time /ai-council is invoked — no registration step required.

Quick Reference

Output Files

Common Patterns

Full Council

Select all agents for comprehensive analysis of major decisions.

Technical Review

Select: Technical Validator + Devil’s Advocate + Neutral Analyst

Strategic Review

Select: Optimist Strategist + Growth Strategist + Devil’s Advocate

Quick Feasibility

Select: Technical Validator + Neutral Analyst

Example Invocation

User: /ai-council

Claude:

1. Discovers all agents in agents/ via Glob
2. Presents dynamic multi-select with discovered agents
3. User selects agents and provides proposal
4. Launches parallel Task agents (each using their configured model)
5. Collects all responses
6. Writes complete deliberation to AI-COUNCIL.md
7. Presents summary to user
8. Offers debate option
9. If debate requested, appends debate to file
10. Finalizes synthesis in file and presents recommendations