What problem does it solve? Autonomous AI agent loops often churn without measurable progress, retry the same root cause, stall merge queues, or drift in cost from unbounded escalation. This Skill provides canonical patterns for selecting, combining, and recovering continuous agent loops so long-running automation stays controlled and auditable. ## Core Features & Use Cases - Loop Selection Flow: Routes work to the right loop pattern (continuous-pr, rfc-dag, infinite, or sequential) based on whether you need strict CI/PR control, RFC decomposition, or exploratory parallel generation. - Combined Production Stack: Recommends a layered pipeline of RFC decomposition, code quality gates, an eval harness loop, and session persistence. - Failure Modes & Recovery: Documents common failure modes and a recovery procedure (freeze loop, run harness audit, reduce scope, replay with explicit acceptance criteria). - Use Case: A team running an overnight autonomous coding agent notices repeated retries with no merged PRs; they apply the recovery steps to freeze the loop, audit the harness, and replay only the failing unit with clear acceptance criteria. ## Quick Start Ask the agent to select and run the appropriate continuous loop pattern for your task, applying quality gates and the documented recovery procedure if the loop stalls.
