team - SKILL.md Agent Skill

name: team description: Use when work must proceed in parallel, when planning and implementation overlap, or when one lane is blocked and another can move. Coordinate multi-lane or multi-session work without letting agents step on each other.

Coordinate parallel work while preserving one authoritative source of truth for each artifact.

Maintain .relay-kit/state/team-board.md with lanes, owners, active artifacts, blockers, and merge order.
Split work only when lanes are independent enough to avoid editing the same artifact section at the same time.
Use cook to drive each active lane, but keep final merge and priority decisions here.
If one lane uncovers architecture or scope drift, update workflow-state and notify all affected lanes.
Park lanes that are blocked instead of letting them thrash.
Record lock scope and handoff status whenever a lane changes ownership or pauses.
For each lane, record depends_on, wave_id, and resume_condition, then only advance to the next wave after current-wave verification gates pass.
Run relay-kit lane audit <project> --strict --json before claiming multi-lane state is safe.
Route through delegation-control before creating subagents; medium reasoning is the default and low reasoning requires a proven mechanical task.

Do not let two lanes silently diverge on the same acceptance criteria.
Do not keep lane state only in memory.
Do not parallelize before a quick scout when the codebase area is unfamiliar.
Do not close a lane as done when no artifact delta or verification evidence exists.

Shared artifacts beat chat summaries; update the artifact before handing off.
Assign one owner skill per lane and name merge order explicitly.
Use cook inside a lane, not as a replacement for team.
Use .relay-kit/docs/parallel-execution.md to decide when work is independent enough to split safely.
Require context-continuity handoff packs when ownership shifts across sessions or AIs.
Prefer wave-based execution: parallel inside a wave, strict dependency gate between waves.
Run relay-kit lane audit <project> --strict --json before trusting a multi-lane handoff.
Open references/team-operator-contract.md when scope, evidence, or operator safety is unclear.
Use examples/team-good-output.md and examples/team-bad-output.md to calibrate output quality.
Use evals/team-cases.json as the minimum scenario set for behavior regression checks.
Use competencies/team-competencies.json to check covered competencies, failure traps, and unknown-domain policy.