paseo-loop

Paseo Loop runs an autonomous agent loop until an exit condition is met, structured as a worker/verifier cycle: launch a worker to make progress, check verification, then repeat until the work is done or configured limits are hit. It fits requests phrased as loop, babysit, keep trying until, check every X, or watch this until X. Under the hood loops are a CLI primitive: paseo loop run starts one, and paseo loop ls, inspect, logs, and stop manage running loops. Before starting, the skill reads the companion paseo skill and the user's ~/.paseo/orchestration-preferences.json to pick worker and verifier providers unless the user named them explicitly. Setting up a loop means turning the user's intent into a self-contained worker prompt that is concrete about what to do each iteration and explicit about what counts as progress, then choosing a verification shape. Verification can be an objective shell check (--verify-check, for example gh pr checks --fail-fast or npm test), a verifier prompt (--verify) for judgment calls that cites the command and outcome, or both together so a shell check rules out obvious failures and the verifier judges the rest. Providers are set with --provider for the worker and --verify-provider for the verifier, and pairing them on different providers is recommended for implementation loops so each catches the other's blind spots. A --sleep interval is added only when polling something external; otherwise the loop runs as fast as it completes. Bounding runs with --max-iterations and/or --max-time is stressed, since open-ended loops are how runaways happen, and --archive keeps agents after each iteration for inspection. Documented shapes include babysitting a PR, driving tests to green, and cross-provider implementation, each with concrete verification and stop settings. The skill closes with prompt rules distinguishing self-contained worker prompts from fact-checking verifier prompts.