What problem does it solve? Recurring failures, gate rejections, and blockers in Loopbench task histories often go unexamined, so teams repeat the same mistakes or add premature process machinery. This Skill turns that accumulated evidence into a small number of well-justified, reversible workflow improvement proposals. ## Core Features & Use Cases - Evidence-led analysis: Inspects accepted revision diffs, gate events, blockers, finish rejections, adversarial reviews, and source-backed notes via the loopbench CLI, separating recurring patterns from one-off incidents. - Graduated codification threshold: Uses an N=3 default threshold for codifying interaction-shaped mechanisms, while allowing immediate safeguards for high-impact incidents with cheap, testable fixes. - Deterministic harness diagnostic: Runs a three-agent, read-only diagnostic across conversation, repository, and agent-customization evidence lanes when diagnosing Loopbench itself. - Use Case: After several tasks fail the same human gate, run a retrospective to identify the recurring cause, consider a competing explanation, and propose at most two minimal reversible changes with validation tasks and rollback conditions. ## Quick Start Ask the assistant to run a Loopbench retrospective over recent task histories and propose evidence-backed workflow improvements for your approval.
