What problem does it solve? Serial, single-agent implementation of a spec-kit tasks.md is slow and loses structural context; this Skill runs the dev-orchestrator wave model, dispatching one subagent per task in parallel within each dependency wave, each pre-loaded with a graphify blast-radius so it never re-explores the repository. ## Core Features & Use Cases - Wave-based parallel dispatch: Reconstructs the task DAG from tasks.md (or its ## Execution Waves section), dispatches one subagent per task in the same turn, holds the barrier, reviews, then unlocks the next wave. - Roster-bound assembled agents: Every dispatch reads its task's row from the approved workforce roster (agents/assignment.md), carrying base agent, injected skills, and elevated grants into the prompt and the trace. - Proportional review and full observability: Applies full or light review depth per wave based on cross-wave file conflicts and outcomes, commits each wave before marking tasks [X], and appends one implementer trace record per task to traces.jsonl plus one line per wave to implement.log.md. - Use Case: After running /speckit-tasks-graph and approving a workforce roster for a feature, invoke this Skill to implement all tasks across parallel waves with per-wave commits, gate re-verification, and an auditable completion report. ## Quick Start Run /speckit-implement-parallel in a spec-kit feature directory that has tasks.md, graphify-context.md, and an approved agents/assignment.md roster, optionally passing a story or task filter as the argument.
