Debate3

Three-way debate between Gemini, Codex, and MiniMax. Each proposes independently, critiques the other two, then the agent moderates and implements.

Configuration

Arguments: $ARGUMENTS

Check the arguments for flags:

Mode flags:

• --dry-run → debate and plan only, skip implementation
• --skip-final → skip the final review phase
• --skip-explore → skip Phase 1 codebase exploration, go straight to the debate

Strip all flags from arguments to get the task description.

Phase 1: Understand the Task (No Questions)

If --skip-explore: Skip the exploration step. Use only the task description and any files the user has already mentioned or that are obvious from the arguments. Proceed directly to Phase 2 with a minimal context summary.

1. Explore the codebase - use Glob, Grep, Read to understand:

   • Relevant files and their structure
   • Existing patterns and conventions
   • Dependencies and interfaces

2. Make reasonable assumptions - do NOT ask clarifying questions

   • Use best judgment based on codebase context
   • Prefer simpler solutions when ambiguous
   • Follow existing patterns in the codebase

3. Prepare context summary - create a brief summary of:

   • The task to be implemented
   • Relevant files discovered
   • Key patterns and conventions in the codebase
   • Any constraints or considerations

Phase 2: Opening Arguments

Have all three LLMs propose their approach independently (in parallel).

Opening prompt:

I need to implement the following task:

[Task description]

Here's what I found in the codebase:
[Context summary - relevant files, patterns, conventions]

Propose your implementation approach:
1. **Approach**: Describe your recommended approach in 2-3 sentences
2. **Key decisions**: List the main architectural/design decisions
3. **Files**: What files to create or modify
4. **Steps**: High-level implementation steps
5. **Trade-offs**: What are the pros and cons of this approach?

Be specific and opinionated. Defend your choices.

Spawn ALL THREE as parallel subagents (Agent tool, subagent_type: "general-purpose", model: "sonnet"). Each subagent prompt must include the full opening prompt text and file list so it can make the MCP call independently.

Gemini subagent — prompt must include:

• Call mcp__consult-llm__consult_llm with model: "gemini", prompt: the opening prompt, files: [array of relevant source files]
• Return the COMPLETE response including any [thread_id:xxx] prefix

Codex subagent — prompt must include:

• Call mcp__consult-llm__consult_llm with model: "openai", prompt: the opening prompt, files: [array of relevant source files]
• Return the COMPLETE response including any [thread_id:xxx] prefix

MiniMax subagent — prompt must include:

• Call mcp__consult-llm__consult_llm with model: "minimax", prompt: the opening prompt, files: [array of relevant source files]
• Return the COMPLETE response including any [thread_id:xxx] prefix

Extract thread IDs: Save gemini_thread_id, codex_thread_id, and minimax_thread_id from the [thread_id:xxx] prefixes in the subagent responses.

Phase 3: Rebuttals

Have each LLM critique the other two approaches (in parallel). Use thread_id to continue each LLM's conversation — they already have full context of the task and their own opening argument, so you only need to send the opponents' arguments.

Rebuttal prompt (same structure for all three, just swap the opponents' arguments):

Two other LLMs proposed alternative approaches:

**Opponent A:**
[First opponent's opening argument]

**Opponent B:**
[Second opponent's opening argument]

Provide a rebuttal:
1. **Critique A**: What are the weaknesses in Opponent A's approach?
2. **Critique B**: What are the weaknesses in Opponent B's approach?
3. **Defense**: Address any weaknesses in your own approach
4. **Concessions**: Are there any good ideas from either opponent worth adopting?
5. **Final position**: State your refined recommendation

Be constructive but thorough in your critique.

Spawn ALL THREE as parallel subagents (Agent tool, subagent_type: "general-purpose", model: "sonnet"). Each subagent prompt must include the full rebuttal prompt text and thread_id.

Gemini subagent — prompt must include:

• Call mcp__consult-llm__consult_llm with model: "gemini", prompt: rebuttal prompt with Codex as Opponent A and MiniMax as Opponent B, thread_id: gemini_thread_id from Phase 2
• Return the COMPLETE response including any [thread_id:xxx] prefix

Codex subagent — prompt must include:

• Call mcp__consult-llm__consult_llm with model: "openai", prompt: rebuttal prompt with Gemini as Opponent A and MiniMax as Opponent B, thread_id: codex_thread_id from Phase 2
• Return the COMPLETE response including any [thread_id:xxx] prefix

MiniMax subagent — prompt must include:

• Call mcp__consult-llm__consult_llm with model: "minimax", prompt: rebuttal prompt with Gemini as Opponent A and Codex as Opponent B, thread_id: minimax_thread_id from Phase 2
• Return the COMPLETE response including any [thread_id:xxx] prefix

Phase 4: Moderator's Verdict

As the moderator, analyze the three-way debate and synthesize the best approach:

1. Score the arguments:

   • Which approach is simplest?
   • Which approach better fits existing patterns?
   • Which critiques were valid?
   • What concessions were made?
   • Where did two or more LLMs agree against the third?

2. Identify consensus: Where did all three (or two of three) agree?

3. Resolve disagreements: For each point of contention:

   • Evaluate the arguments from all sides
   • Two-against-one agreement is a strong signal but not automatic — evaluate the dissenter's reasoning
   • Pick the stronger position or find a middle ground
   • Prefer simpler solutions when arguments are equally strong

4. Write the verdict as part of the plan:

# [Feature Name] Implementation Plan

**Goal:** [One sentence describing what this builds]

## Debate Summary

**Gemini's position:** [1-2 sentence summary]
**Codex's position:** [1-2 sentence summary]
**MiniMax's position:** [1-2 sentence summary]

**Points of agreement (all three):**
- [Consensus point 1]

**Majority agreements (2 of 3):**
- [Point]: [Who agreed] vs [Who dissented] → **Verdict:** [decision]

**Three-way disagreements:**
- [Issue]: Gemini said X, Codex said Y, MiniMax said Z → **Verdict:** [Your decision and why]

**Verdict:** [2-3 sentences on the final synthesized approach]

---

### Task 1: [Short description]

**Files:**
- Create: `exact/path/to/file.py`
- Modify: `exact/path/to/existing.py` (lines 123-145)

**Steps:**
1. [Specific action]
2. [Specific action]

**Code:**
```language
// Include actual code, not placeholders
```

---

Guidelines:

• Exact file paths - never "somewhere in src/"
• Complete code - show the actual code
• Small tasks - 2-5 minutes of work each
• DRY, YAGNI - only what's needed

Save the plan to history/plan-<feature-name>.md.

Phase 5: Implement

If --dry-run: Skip to Phase 7 (Summary) - report the debate and plan without implementing.

Implement the plan without further interaction:

1. Follow the plan exactly - implement each task in order
2. Commit after each logical unit - keep commits small and focused
3. If something is unclear - make a reasonable decision and note it in the commit message
4. If a task fails - attempt to fix it before moving on
5. Only stop if there's a blocking error that cannot be resolved

Implementation rules:

• Work through tasks sequentially
• Test changes when possible
• Keep commits atomic and well-documented
• Use commit messages that explain the "why"

Phase 6: Final Review

If --skip-final: Skip to Phase 7 (Summary).

After implementation, have all three LLMs review the result (in parallel). Use thread_id to continue each LLM's conversation — they already have full context of the task and the debate, so you only need to send the review prompt and the diff.

Final review prompt:

The implementation is complete. Review the changes for bugs, issues, or improvements:
- Any obvious bugs or edge cases missed?
- Code quality issues (error handling, naming, structure)?
- Deviations from best practices?
- Security concerns?

Be concise. Only flag issues worth fixing.

Spawn ALL THREE as parallel subagents (Agent tool, subagent_type: "general-purpose", model: "sonnet"). Each subagent prompt must include the full review prompt, thread_id, and git_diff details.

Gemini subagent — prompt must include:

• Call mcp__consult-llm__consult_llm with model: "gemini", prompt: the final review prompt, thread_id: gemini_thread_id from Phase 2, git_diff: { "files": [list of changed files], "base_ref": "HEAD~N" }
• Return the COMPLETE response including any [thread_id:xxx] prefix

Codex subagent — prompt must include:

• Call mcp__consult-llm__consult_llm with model: "openai", prompt: the final review prompt, thread_id: codex_thread_id from Phase 2, git_diff: { "files": [list of changed files], "base_ref": "HEAD~N" }
• Return the COMPLETE response including any [thread_id:xxx] prefix

MiniMax subagent — prompt must include:

• Call mcp__consult-llm__consult_llm with model: "minimax", prompt: the final review prompt, thread_id: minimax_thread_id from Phase 2, git_diff: { "files": [list of changed files], "base_ref": "HEAD~N" }
• Return the COMPLETE response including any [thread_id:xxx] prefix

Apply fixes if two or more reviewers identify the same issue, or if one raises a clearly valid concern:

• Fix bugs and edge cases
• Commit each fix separately with clear messages

Skip minor style suggestions or conflicting opinions.

Phase 7: Summary

Present a final summary to the user:

## Summary

**Implemented:** [One sentence describing what was built]

**Debate outcome:**
- Gemini advocated: [key position]
- Codex advocated: [key position]
- MiniMax advocated: [key position]
- Final verdict: [synthesized approach]

**Key decisions from debate:**
- [Decision 1 and why]
- [Decision 2 and why]

**Post-implementation fixes:**
- [Fix applied after final review, if any]

**Commits:**
- `abc1234` - [commit message]
- `def5678` - [commit message]