Slash Command

/refine

Run team-based adversarial refinement on a bead or spec using Agent Teams debate

Install

npx claudepluginhub rbergman/dark-matter-marketplace --plugin dm-team

Details

Argumentbead ID or spec file path

Prompt Preview

# /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...

Command Content

Other plugins with /refine

/refine

Refines existing CLI harness for software at <software-path> by analyzing gaps in coverage and adding new commands/tests, optionally focusing on specific features.

cli-anything

27.6k

/refine

Refines existing cli-anything-web CLI: performs gap analysis, records new traffic, implements missing endpoints, updates tests and REPL help.

Bash(*)ReadWrite+2

cli-anything-web

/refine

Refines an existing writer skill by analyzing additional writing samples and updating the voice profile. Provide path to skill directory.

voiceprint

/refine

Runs iterative refinement loop to improve project quality toward a goal, with max iterations, target condition, and semicolon-separated DoD criteria.

second-claude-code

/refine

Analyzes existing PRD for gaps using a section checklist, generates markdown gap analysis, and refines via targeted interactive questions.

ReadWriteEdit+6

prd-builder

/refine

planning/

Refines an existing plan by loading it, applying changes like updated requirements or solutions, showing before/after comparison, and saving the updated plan.

aura-frog

Stats

Parent Repo Stars2

Parent Repo Forks2

Last CommitFeb 9, 2026

Actions

View Source View Plugin View on GitHub View README

/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:

Analysis -- Analyst surfaces ambiguity, tags protected items, shares with team
Live Debate -- Proposer proposes cuts, Advocate challenges, genuine back-and-forth until convergence
Scope Lock -- Lead checks Too Thin indicators, resumes debate if 2+ trigger, HITL checkpoint if needed
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.