What problem does it solve? Coordinating work on multiple release components at once is fragile when state lives in one orchestrator's session: claims conflict, progress is lost when a session dies, and dependency rules get bypassed. This Skill fans a wave of work out across startable components while keeping every claim, status, and progress note durable in the shared Marshall store. ## Core Features & Use Cases - Store-driven wave dispatch: Opens a dispatch_run over a set of components, with a server-side dependency-graph guard that rejects members whose blockers are not yet merged. - Parallel subagent orchestration: Claims each component and cuts a git worktree via /marshall:release-topic, then spawns one Claude subagent per member that heartbeats its claim and reports progress (dispatched, in_progress, proposed, merged, failed). - Resumable across sessions: Re-invoking re-attaches to an open run, reconciles stale failed members against reality, and reports the wave board instead of starting over. - Use Case: You have five unblocked components in release v0.5.0. Run the skill to open a dispatch run, spin up five worktrees and subagents concurrently, and later resume from a new session to see exactly which members proposed PRs and which failed. ## Quick Start Say "start wave 2 in parallel" or "/release-parallel #41 #42 #45" to dispatch those components as concurrent subagents tracked by a Marshall dispatch run.
