converge

Structured debate and refinement using Agent Teams. Takes divergent findings (from /diverge or any independent research) and spawns a team where teammates advocate for different positions, debate trade-offs, and converge on a refined synthesis.

Arguments

$ARGUMENTS — format: [N] [path/to/diverge_output.md or inline topic] where N is optional number of debaters (default: 3, min 2, max 5)

Prerequisites

• Agent Teams must be enabled (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1)
• Input should be divergent findings — either a file path to prior /diverge output, a PRD with open tensions, or a description of competing approaches

Parse Arguments

Extract:

• team_size: optional leading integer (default 3, min 2, max 5)
• input: path to a file with divergent findings, or inline description of the debate topic

Phase 0: Validate Teams Availability

Check that Agent Teams is enabled. If not, inform the user:

Agent Teams is required for /converge. Enable it by adding to your Claude Code settings:
  "env": { "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1" }
Or use /diverge for independent exploration without debate.

Phase 1: Frame the Debate

If a file path is provided: read the file and extract the key tensions, divergence points, and competing recommendations.

If inline input: parse the competing positions from the description.

If neither is clear: ask the user what positions or findings should be debated.

Prepare a debate brief that includes:

• The core question or decision to be made
• 2-5 distinct positions to be advocated (one per teammate)
• Shared context and constraints all positions must respect
• Evaluation criteria: what makes one approach better than another?

Present the debate brief and position assignments to the user for confirmation.

Phase 2: Spawn the Team

Create an agent team with the following structure:

Team Lead (you): Moderator and synthesizer. You frame rounds, call for responses, and produce the final synthesis.