decide

AgentCouncil Decide

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

Decision: $ARGUMENTS

Protocol -- follow these steps exactly

Step 1: Identify the decision and options

Clearly state the decision question and the options being evaluated. There must be at least 2 options, each with an id, label, and description. Also identify any criteria and constraints. This frames the evaluation for both agents and the user.

Step 2: Write YOUR assessment first

Before Codex sees anything, write your own independent assessment of each option. For each option, state:

• Pros and cons
• Blocking risks (if any)
• Assumptions you are making
• Tradeoffs involved
• Your confidence level (high/medium/low)

Display it clearly labeled as Claude's Assessment.

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

Step 3: Create the factual brief and send to Codex

Write a brief containing ONLY factual context:

• The decision question
• Each option (id, label, description)
• Criteria (if any)
• Constraints (if any)

CRITICAL: Do NOT include your assessment, preference, or ranking. The brief must contain zero evaluative content. No "I think option A is better", "I prefer", "my recommendation" -- purely factual context.

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

Step 4: Share your assessment with Codex

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

"Here is the lead evaluator's independent assessment (written before seeing yours): {your full assessment from Step 2}. Compare it with your own analysis. You may agree, disagree, or reassess. Push back where you think the lead is wrong."

Step 5: Synthesize the decision

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

"Based on our full discussion, produce a final decision synthesis. Return JSON with these exact keys: outcome (decided/deferred/experiment), winner_option_id (string or null), decision_summary (string), option_assessments (array of objects with option_id, pros, cons, blocking_risks, disposition, confidence, source_refs), defer_reason (string or null), experiment_plan (string or null), revisit_triggers (array of strings), next_action (string). Rules: evaluate ONLY the provided options, do NOT invent new options. For EVERY option include assumptions, tradeoffs, and confidence even when one option is the clear winner. Set disposition to selected/viable/rejected/insufficient_information. Outcome: decided if one option clearly superior, deferred if insufficient info, experiment if needs real-world validation."

Step 6: Present the structured result

Parse Codex's JSON response and present:

## Decision Result

**Outcome:** {outcome}
**Winner:** {winner_option_id or "None"}

### Decision Summary
{decision_summary}

### Option Assessments
For each option:
- **{option_id}: {label}** (disposition: {disposition}, confidence: {confidence})
  Pros: {pros}
  Cons: {cons}
  Blocking Risks: {blocking_risks or "None"}

### Revisit Triggers
- {each trigger}

### Next Action
{next_action}

If JSON fails to parse, report outcome as "deferred" and show raw response.

Rules

• Default to 1 round unless the user asks for more
• ALWAYS write your assessment BEFORE sending anything to Codex
• NEVER put your preference or ranking in the brief -- factual context ONLY
• Evaluate ONLY the provided options -- do NOT invent new ones
• Include assumptions, tradeoffs, and confidence for EVERY option
• Codex remembers everything in the thread -- don't resend context
• Do NOT add analysis beyond the structured result