discuss

Discuss

Input: $ARGUMENTS = topic or question + optional --agents <list> (comma-separated agent names)

Steps

1. Parse $ARGUMENTS:

   • --agents reviewer,fixer,architect → agent_list
   • Default agent_list: reviewer,fixer,architect (if not specified)
   • Remainder = topic
   • Max 3 agents per discussion

2. For each agent in agent_list (in parallel): Before dispatch, check Agent Overrides: for each agent you dispatch, follow ../../core/agent-override-injector.md for that agent's overrides. Run agent via Task tool with context:

   You are participating in a multi-agent discussion about: {topic}
   Your role: {agent description from frontmatter}
   Style: {agent style from frontmatter}
   
   Provide your perspective on this topic in 100-200 words.
   Focus on concerns and insights specific to YOUR expertise.
   Be opinionated — disagree with conventional wisdom if your expertise suggests otherwise.
   

3. Collect all agent responses.

4. Display as structured discussion:

   ## Discussion: {topic}
   
   ### {agent-1 name} ({agent-1 style})
   {agent-1 perspective}
   
   ### {agent-2 name} ({agent-2 style})
   {agent-2 perspective}
   
   ### {agent-3 name} ({agent-3 style})
   {agent-3 perspective}
   
   ### Synthesis
   {synthesize key agreements, disagreements, and recommended approach}
   

5. Ask: "Follow up on any perspective? [agent name / done]" If user picks an agent → run that agent again with the full discussion context for deeper exploration.

Rules

• Max 3 agents per discussion
• Read-only — no code changes
• Each agent response: 100-200 words max
• Discussion is for exploration, not decisions — no pipeline side effects