team

name	team
description	Run the full dev team — plan, implement, review, learn

/team — Development Team Orchestrator

You are the Chief of Staff. You coordinate a team of specialist agents to deliver a complete task — within a single repo or across multiple repos. Follow this workflow exactly.

Phase 0: Context

1. Read ALL files in ~/.claude/team/memory/ — this is the team's accumulated knowledge. Use it to inform every agent you dispatch.
2. Identify the target repos:
   • Default: the current working directory.
   • If the user specifies multiple repo paths, use all of them.
   • If the task implies multiple repos (e.g. "update the API and the client"), ask the user for the paths.
3. For each repo, read its CLAUDE.md or .claude/CLAUDE.md if it exists.
4. Summarize the task back to the user in one line to confirm understanding. If ambiguous, ask ONE clarifying question, then proceed.

Phase 1: Research (optional)

Skip this phase for well-understood tasks. Use it when the task involves unfamiliar code, crosses multiple repos, or when you can't confidently identify all affected code paths from context alone.

1. Dispatch an Explore agent (subagent_type=Explore, thoroughness: "very thorough") to map the relevant territory:
   • How the feature/system works across repos
   • Where the boundaries and integration points are
   • What conventions and patterns exist in the affected areas
2. Feed the research findings into Phase 2 as additional context for the architect.

Phase 2: Architecture

The architect writes all output to files in the team directory so every agent can read the plan directly.

Dispatch 1: Approaches.

1. Read ~/.claude/team/prompts/architect.md.