What problem does it solve? Executing a decomposed plan or backlog of many tasks one-by-one through a full development pipeline requires repeatedly invoking the pipeline manually, which is tedious and error-prone. This Skill automates sequential batch execution of tasks T-01 through T-NN, each in its own isolated sub-agent context, stopping only on strong failure signals. ## Core Features & Use Cases - Sequential batch orchestration: Parses a BATCH_PLAN.md task table, validates dependencies and schema, builds a topological order, and dispatches each task to a pm-orchestrator sub-agent. - Strong-signal stop conditions: Halts the batch on verify_all FAIL, a FAILED verdict, a STOP intervention, or a safety-hook block, and never auto-retries failed tasks. - Idempotent resume: Re-invoking with the same batch-id skips tasks already marked done or whose delivery doc shows a DELIVERED verdict. - Use Case: After a /harness-plan decomposition produces eight tasks, run the batch once and receive a BATCH_REPORT.md with per-task verdicts, aggregate stats, and the stop reason if any. ## Quick Start Run the batch for batch-id sprint-12 by invoking /harness-batch sprint-12 after creating docs/batches/sprint-12/BATCH_PLAN.md with your task table.
