autonomous-loops

To orchestrate autonomous Claude Code loops, you provide an RFC specification to decompose work units into structured patterns like Sequential Pipelines or Infinite Agentic Loops. This enables multi-agent coordination with per-stage context isolation across iterative workflows.

Autonomous agent workflow patterns include Sequential Pipeline, NanoClaw REPL, Infinite Agentic Loop, Continuous Claude PR Loop, De-Sloppify, and Ralphinho. These patterns enable cross-pattern composition and resumable workflows for scalable automation.

To prevent context bleed in multi-agent orchestration, the system enforces separate context windows per stage and uses a merge-and-evict workflow. This ensures robust context management and context-preserving recovery across iterative loops.

To start RFC-driven DAG orchestration, you provide an RFC specification or sample spec and trigger the first research phase. The system decomposes the RFC into work units and handles explicit dependency DAG execution automatically.

Yes, you can compose multiple loop patterns together for complex automation. The system supports cross-pattern composition and resumable workflows, allowing you to scale from simple sequential pipelines to RFC-driven multi-agent DAGs.

The merge queue eviction workflow is a context-preserving recovery mechanism for autonomous loops. It enforces per-stage isolation and evicts completed context to maintain robust context management across iterative loop executions.