What problem does it solve? When a reviewer or test_engineer is dispatched across multiple tasks in one batch, the gate tracker cannot tell which verdict belongs to which task, causing ambiguous or failed attribution. This Skill defines a parseable per-task verdict row protocol so every dispatch produces task-scoped gate evidence. ## Core Features & Use Cases - Per-task attribution rules: Single-task dispatches attribute by taskId, while set-dispatches attribute via parseable [REVIEWED] | task-id | verdict and [TESTED] | task-id | verdict rows. - Verdict normalization: Accepts APPROVED/REJECTED/CONCERNS for reviews and PASS/FAIL/SKIPPED for tests, normalizing task-X.Y IDs to X.Y and ignoring unsafe or non-plan IDs. - Parallel dispatch guidance: Recommends dispatch_lanes_async with one lane per task (up to 3 concurrent) when set-dispatch is inappropriate, plus a minimal prompt template for trivial ceremony gates. - Use Case: A reviewer returns one batched output covering tasks 2.1 through 2.4; each parseable row writes a separate .swarm/evidence/{taskId}.json file so the delegation gate can evaluate each task independently. ## Quick Start Require the reviewer to return one [REVIEWED] | task-id | verdict | notes row per task so each dispatch is attributed to its plan task.
