What problem does it solve? Working through a backlog or multi-item request with sub-agents often leads to unconfirmed scope, overlapping file edits, lost context, and unreviewed results. This Skill provides a disciplined protocol for confirming, grouping, dispatching, and reviewing batches of tasks so nothing is silently skipped or broken. ## Core Features & Use Cases - Confirmation Gate: Presents the proposed task selection, grouping, and assumptions to the user before any agent is launched or file is changed. - Parallelisation and Sizing Rules: Groups independent tasks, assigns one sub-agent per group, and tiers work between Executor (Focused) for tasks touching up to 5 files and Executor (Broad) for cross-component work. - Dispatch Briefs and Outcome Protocol: Defines a cold-start dispatch format with verbatim values and a four-status outcome model (DONE, DONE_WITH_CONCERNS, NEEDS_CONTEXT, BLOCKED) with explicit handling for each. - Review Discipline: Enforces spec-compliance-then-quality review per task, a five-round fix cap with written rulings, and a final cross-batch review. - Use Case: Given a 12-item TODO list spanning several modules, confirm the plan with the user, fan out independent groups to sub-agents, review each diff, and keep the tracking file updated so progress survives context compaction. ## Quick Start Work through my TODO list as a batch: confirm the plan with me first, then run independent tasks in parallel sub-agents and review each result before moving on.
