Patterns for continuous autonomous agent loops with quality gates, evals, and recovery controls.
The continuous-agent-loop skill defines structured patterns for running continuous autonomous agent workflows with built-in quality controls, evaluation cycles, and recovery mechanisms. It is designed to help teams manage long-running or iterative AI agent operations while reducing common operational risks such as uncontrolled retries, stalled merge queues, and escalating execution costs. The skill provides a canonical loop model for v1.8+ environments and replaces the earlier autonomous-loops naming convention while maintaining temporary compatibility. It also includes a decision flow for selecting the appropriate execution pattern based on workflow requirements such as CI and PR enforcement, RFC decomposition, exploratory parallel generation, or sequential execution.
The skill includes several operational capabilities intended for production-grade autonomous workflows. It supports loop selection through predefined patterns including continuous-pr, rfc-dag, infinite, and sequential modes. It also promotes a recommended production stack composed of RFC decomposition pipelines, quality gate enforcement, evaluation harnesses, and session persistence systems. In addition, the skill documents known failure modes such as loop churn without measurable progress and repeated retries caused by unresolved root causes. Recovery guidance is also included, providing operational controls such as freezing the loop, auditing harness behavior, narrowing execution scope, and replaying workflows with explicit acceptance criteria.
This skill is intended for engineering teams, AI platform operators, and developers building autonomous software delivery or evaluation systems. It is particularly useful in environments that require controlled iterative execution, automated quality validation, and resilient recovery processes. Teams managing CI-integrated agent workflows, RFC-driven development pipelines, or persistent evaluation loops can use this skill to standardize operational behavior and reduce instability in autonomous execution systems.
The skill provides patterns and operational guidance for managing continuous autonomous agent loops with quality gates, evaluation workflows, and recovery controls.
The documented selection flow recommends choosing continuous-pr for strict CI or PR control, rfc-dag for RFC decomposition, infinite for exploratory parallel generation, and sequential as the default option.
Yes. The skill supersedes the autonomous-loops naming convention while maintaining compatibility for one release cycle.
The recommended production stack includes RFC decomposition, quality gate enforcement, evaluation loops, and session persistence components.
The documentation highlights risks such as loop churn, repeated retries with unresolved causes, merge queue stalls, and cost drift from unbounded escalation. Recovery procedures are included to mitigate these issues.
Quick Setup:
.claude/skills/Repository
affaan-m/everything-claude-code