autonomous-loops

The `autonomous-loops` skill provides structured patterns and architectures for running Claude Code in autonomous loops, enabling developers to automate complex workflows without constant human oversight. It addresses the challenge of orchestrating sequential and parallel AI tasks, ranging from simple step-by-step pipelines to sophisticated multi-agent DAG systems driven by RFCs. By formalizing loop structures, this skill helps maintain consistency, context persistence, and orderly progression of tasks across iterative cycles.

Core capabilities include a spectrum of loop patterns such as sequential pipelines, interactive persistent sessions (NanoClaw REPL), continuous PR loops, and RFC-driven DAG orchestration. Each pattern offers distinct complexity levels and suitability, from daily development steps to large-scale multi-agent coordination with merge queues. Additional features like context isolation per step, cleanup passes (De-Sloppify), and quality gates ensure robust and reliable autonomous execution. The skill also supports variations with model routing for specialized tasks, allowing for flexible integration into diverse workflows.

This skill is ideal for developers, teams, or organizations aiming to implement autonomous development pipelines or continuous integration processes. It is particularly useful for building CI/CD-style loops, coordinating parallel agents, maintaining context across iterations, and enforcing quality control in automated development tasks. Its target users are those seeking to scale AI-driven coding and workflow automation efficiently, from individual contributors to multi-unit development teams.