What problem does it solve? When running subagent-driven development under a hard token budget, dispatching a reviewer subagent for every task can exhaust the budget before implementation completes, yet skipping verification entirely invites defects. This Skill defines when controller-side verification may safely substitute for a reviewer subagent and which checks remain mandatory. ## Core Features & Use Cases - Budget-Constrained Review Mode: Allows controller verification (machine check plus read-review) only for tasks with verbatim plan artifacts and a machine check, while keeping reviewer subagents mandatory for authored or novel code. - Pre-Dispatch Verification Claims: Requires controllers to state the exact search scope behind factual claims that license cross-cutting changes, preventing unverified global edits. - Resume Brief Construction: Builds task briefs from the union of the plan section, the progress ledger roll-up, and any resume/handoff guide carve-outs. - Stalled Delegation Recovery: Detects implementer subagents hallucinating a background delegate and resumes them in place via SendMessage with an explicit correction. - Use Case: During a phase execution with under 10% of the weekly token budget remaining, substitute machine verification for five verbatim-transcription tasks while reserving reviewer subagents for the one novel-code task, then pass the mandatory whole-branch final review. ## Quick Start Ask the agent to apply budget-constrained review mode to the current subagent-driven development plan and identify which tasks qualify for controller verification.
