brainstorm

AgentCouncil Brainstorm

You are running the AgentCouncil deliberation protocol. You are BOTH the orchestrator and the lead agent. Codex is the independent outside agent, running as a persistent MCP server with session memory.

Topic: $ARGUMENTS

Protocol — follow these steps exactly

Step 1: State the problem

Clearly state what needs to be decided and why it matters. This frames the deliberation for both agents and the user.

Step 2: Write YOUR proposal first

Before Codex sees anything, write your own independent proposal. You have the full conversation context — use it. Be specific and opinionated.

Display it clearly labeled as Claude's proposal.

This step is critical. If you see Codex's thinking before writing your own, the independence is gone and the whole protocol is worthless.

Step 3: Create the brief and send to Codex

Write a neutral brief containing ONLY:

• Problem statement
• Background context
• Constraints
• Goals
• Open questions

CRITICAL: Do NOT include your proposal, opinion, or any direction. The brief must be clean.

Call mcp__codex__codex to start a codex session with this brief. Save the threadId.

Step 4: Share your proposal with Codex

Send your full independent proposal to Codex so it can compare both views. Use mcp__codex__codex-reply:

"Here is the lead agent's independent proposal (written before seeing yours): {your full proposal from Step 2}. Compare it with your own proposal. Where do you agree? Where do you disagree? Push back where you think the lead is wrong."

Step 5: Compare proposals

Now both agents have seen each other's work. Display Codex's comparison response. Summarize where the two proposals agree and disagree.

Step 6: Exchange rounds (with early exit)

Default to 1 round unless the user asks for more. For each exchange after Step 5:

Read Codex's response. If there are no material disagreements — both proposals align on direction, details, and tradeoffs — skip remaining exchanges and go directly to Step 7 (synthesis). State: "No material disagreements — proceeding to synthesis."

If disagreements exist: Write your counter-response addressing their pushback. Send it via mcp__codex__codex-reply. Read Codex's counter. Check again for convergence. Repeat until rounds are exhausted or no disagreements remain.

Codex remembers everything — don't resend context.

Step 7: Get Codex's synthesis

Send the synthesis request on the same thread via mcp__codex__codex-reply:

"Based on our full discussion, produce a final synthesis. Return JSON with these exact keys: recommended_direction (string), agreement_points (list of strings), disagreement_points (list of strings), rejected_alternatives (list of strings), open_risks (list of strings), next_action (string), status (one of: consensus, consensus_with_reservations, unresolved_disagreement). Do NOT use partial_failure as status."

Step 8: Present the result

Parse Codex's JSON response and present:

## Brainstorm Result

**Status:** {status}

### Recommended Direction
{recommended_direction}

### Agreement Points
- {each point}

### Disagreement Points
- {each point}

### Rejected Alternatives
- {each point}

### Open Risks
- {each risk}

### Next Action
{next_action}

If JSON fails to parse, report status as "unresolved_disagreement" and show raw response.

Rules

• Default to 1 round unless the user asks for more
• ALWAYS write your proposal BEFORE sending anything to Codex
• NEVER put your opinion in the brief
• Codex remembers everything in the thread — don't resend context
• Do NOT add analysis beyond the structured result