What problem does it solve? Running a Terminal-Bench task repeatedly against a Todero worktree often stalls in ambiguous review states, unbounded retries, or fixes applied without approval. This Skill enforces a bounded loop with explicit issue topology, typed waiters, and board confirmation so every iteration ends in a known state with a named next-action owner. ## Core Features & Use Cases - Bounded iteration loop: Runs a single Terminal-Bench task as a smoke with an iteration budget, wall-clock caps, and explicit stop rules (pass, board rejection, budget exhausted, or named blocker). - Structured diagnosis: Applies a diagnose-why-work-stopped pattern to classify failures as task/model, Todero product, harness/setup, verifier/infrastructure, security, or unclear, with evidence quoted from run ids and status transitions. - Board-gated fixes: Requires a request_confirmation interaction on the iteration child before any implementation child is created, keeping product fixes auditable and approved. - Worktree continuity: Pins the benchmark command to an isolated Todero App worktree via PAPERCLIPAI_CMD and inheritExecutionWorkspaceFromIssueId so fixes and reruns share one workspace. - Use Case: An operator is asked to drive terminal-bench/fix-git until it passes; the Skill opens a loop parent issue, runs bounded smokes, diagnoses each stop, proposes product rules for board approval, and closes with QA and CTO review. ## Quick Start Ask the agent to loop the terminal-bench/fix-git task through Todero with an iteration budget of three until the smoke passes.
