What problem does it solve? Writing a loop-generator spec by hand often produces loops that look valid but prove nothing: wrong test commands, checks that were already green, or evaluators the agent can game. This Skill turns a vague coding goal into a lint-clean .loop.yaml whose checks are confirmed RED before any agent budget is spent. ## Core Features & Use Cases - Repo-grounded spec authoring: Inspects the target repository (package.json, Makefile, CI config) to pick the real test, build, and typecheck commands instead of generic scaffold defaults. - Trust policy configuration: Sets baseline, specGuard, and evaluatorGuard options so vacuous checks fail fast and metric-based loops are tamper-evident. - Verification before handoff: Runs loopgen lint in strict mode and manually confirms each check fails for the right reason on the untouched workspace. - Use Case: A developer wants an agent to add retry logic to an API client. The skill interviews them for the goal, finds the repo's real test command, generates the spec, proves the new test fails today, and hands off a runnable loop with the exact run command. ## Quick Start Ask the assistant to create a loop-generator spec for your coding task, pointing it at the target repository and describing what done looks like.
