/refine

Run team-based adversarial spec refinement.

Arguments: $ARGUMENTS

Protocol

1. Parse Target

• Bead ID -> bd show <id>
• File path -> read file
• No args -> check conversation for current spec context

2. Assess Complexity

• xs/s -> "This doesn't need refinement. Proceed to implementation."
• m -> "Medium complexity -- using 2-phase dm-work:dialectical-refinement instead." Run /dm-work:refine.
• l/xl -> Proceed with team refinement.

3. Gather Context

Read target spec, related files, existing beads for the project.

4. Create Team

Spawn Agent Team with Analyst (haiku), Proposer (opus), Advocate (opus). Lead acts as Judge.

• Include in each spawn prompt: the spec text, their role description, protected categories, instructions to debate with other teammates
• Require plan approval for the Analyst (fast review of analysis before debate starts)

5. Run Debate Protocol

Execute the debate protocol as described in the refinement skill:

1. Analysis -- Analyst surfaces ambiguity, tags protected items, shares with team
2. Live Debate -- Proposer proposes cuts, Advocate challenges, genuine back-and-forth until convergence
3. Scope Lock -- Lead checks Too Thin indicators, resumes debate if 2+ trigger, HITL checkpoint if needed
4. Synthesis -- Lead resolves debates, writes concrete spec with acceptance criteria

6. Produce Output

Synthesized spec in standard format:

## Introduction
## Scope
## Acceptance Criteria
## Out of Scope
## Appendix A: Project Context (if needed)

7. Update Bead

If input was a bead, update with refined spec, add refined label.

8. Cleanup Team

Dismiss teammates after synthesis is complete.