| name | implement |
| description | Prepare or run a Claude Code agent-team implementation only when the user explicitly invokes `/implement` or asks for parallel agents or an agent-team handoff. |
| disable-model-invocation | true |
Agent-Team Implementation
Use this explicit workflow for work that can benefit from Claude Code agent teams.
Analyze
Read AGENTS.md, the requested specification, a relevant existing implementation, shared integration files, and available checks. Identify independently owned file groups and dependencies between them.
Do not create a team when the work is concentrated in one shared file, requires unresolved architecture decisions, or has fewer than two genuinely independent work streams.
Propose The Team
Define only the roles justified by file ownership. Each role needs:
- A concrete outcome and acceptance criteria.
- Exclusive write ownership paths.
- Read-only reference paths.
- Required coordination and dependency order.
- Relevant build, lint, typecheck, test, and browser checks.
Reserve registries, navigation, shared configuration, and final integration for one owner after dependent work lands. Avoid overlapping write scopes and review-only teammates.
Ask for approval of the team boundaries before launching or producing spawn prompts. If agent teams are unavailable, provide copy-ready prompts instead of inventing tool calls.
After implementation, integrate results, resolve conflicts without discarding user changes, run repository checks, summarize ownership and outcomes, and clean up the team.