What problem does it solve? Autonomous coding agents often stop too early, claim false completion, or oscillate on the same failing approach. This Skill enforces a self-referential loop that re-feeds the same goal after every turn and exits only when a typed completion promise is backed by concrete in-turn evidence, a watcher-gate re-verification, or a hard iteration cap. ## Core Features & Use Cases - Evidence-gated completion: The loop refuses to exit on a self-reported "done"; every exit requires a passing gate, a file:line receipt, or an independently recomputed watcher state in the same turn. - Bound CLI operators: Surveying is delegated to simplicio-mapper and code mutation to simplicio-dev-cli, so the model decides while deterministic tools execute and verify. - Stall detection and memory: An append-only run journal fingerprints failures, detects repeated dead-ends, and forces strategy changes or escalation instead of blind retries. - Use Case: Tell the agent "/simplicio-loop finish all the open issues" and it drains the queue item by item, claiming work, applying verified changes, and stopping only when the source re-query confirms the queue is empty. ## Quick Start Ask the agent to run /simplicio-loop on your goal, for example "keep iterating until all tests pass", and it will arm the loop and start iteration 1 immediately.
