team

Prompt file imported from OthmanAdi/MDDesign (.cursor/commands/team.md). Copyright stays with the author.

/mddesign:team

Subagent dispatch.

Subcommands

/mddesign:team dispatch <phase_id>

Auto-picks the subagent:

• Phases with "design" / "ui" / "tokens" in the title and a harvest-shaped intent → design-archeologist
• Phases with no existing PhaseSpec → planner (returns the spec, then you re-dispatch with --agent executor)
• All others → executor

/mddesign:team dispatch <phase_id> --agent <name>

Force a specific subagent. Values:

• planner — draft a PhaseSpec from the phase
• executor — run the PhaseSpec end to end
• reviewer — audit another's return against done_when
• memory-keeper — classify recent session output (usually called by Stop hook)
• design-archeologist — scan codebase for /mddesign:harvest

Flow

1. Read task_plan.md, resolve the phase.
2. Build the PhaseSpec (team-dispatch skill logic).
3. HITL gate (named checkpoint: before_subagent_dispatch).
4. Call Agent tool with the chosen subagent_type.
5. Parse the structured return.
6. Append to progress.md under ### Phase <phase_id> Result.
7. If auto-approve is on, dispatch reviewer.

Boundaries

• Subagents never edit task_plan.md.
• All dispatches HITL gated.
• Markdown is the shared state across agents; no in-memory store.

See skills/team-dispatch/SKILL.md for the full PhaseSpec v1 schema and dispatch logic.