From dm-team
Transform ambiguous specs into implementable work items through live adversarial debate using Agent Teams. Evolves dm-work:dialectical-refinement from sequential pipeline to simultaneous multi-agent debate. Use for l/xl complexity tasks.
How this skill is triggered — by the user, by Claude, or both
Slash command
/dm-team:refinementThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Team-based adversarial spec refinement using Agent Teams. This evolves `dm-work:dialectical-refinement` from sequential pipeline to live debate.
Team-based adversarial spec refinement using Agent Teams. This evolves dm-work:dialectical-refinement from sequential pipeline to live debate.
The original dialectical-refinement runs 5 sequential phases (Analyst->Proposer->Advocate->Scope Lock->Judge) where each phase sees only the previous output. This prevents self-reinforcing mistakes but loses the back-and-forth of genuine argument. With Agent Teams, debaters are persistent -- the Advocate can push back on the Proposer while the Proposer is still formulating, creating richer adversarial tension.
| Complexity | Mechanism |
|---|---|
| xs/s | Skip refinement entirely |
| m | Use dm-work:dialectical-refinement (2-phase, lightweight) |
| l/xl | Use refinement (full debate) |
| Role | Teammate | Model | Purpose |
|---|---|---|---|
| Analyst | Teammate 1 | haiku | Surface ambiguity, identify gaps, tag protected items |
| Proposer | Teammate 2 | opus | Propose simplifications and cuts with confidence levels |
| Advocate | Teammate 3 | opus | Challenge cuts, defend scope, suggest cheap additions |
| Judge | Lead | opus | Moderate debate, enforce scope lock, synthesize final spec |
Same as dm-work:dialectical-refinement — Core Workflow, Agent Primitives, User-Requested Features, Token Efficiency, Structured Output. Tag these early; Proposer does not propose cutting them.
Same as dialectical-refinement for compatibility:
## Introduction
[What + Why in 2-3 sentences]
## Scope
[What's being built]
## Acceptance Criteria
[Testable outcomes]
## Out of Scope
[Explicit boundaries]
## Appendix A: Project Context (if needed)
[Token-efficient big picture: ~100-200 words max]
npx claudepluginhub rbergman/dark-matter-marketplace --plugin dm-teamRefines product specs through iterative LLM debates (OpenAI, Anthropic, Gemini, etc.) until consensus, with Claude participating. Use for adversarial spec writing or refinement.
Develops feature specifications through structured multi-agent debate (architects, UX experts, security specialists) using Claude Code's experimental Agent Teams for complex features with multiple viable approaches.
Generates project specs at optimal depth: quick Linear issues for tasks, lite PRDs for features, AI specs with context, evals. Use /spec shortcuts like --quick or --ai.