autonomous-loops

Persistent state across autonomous loop iterations is managed via SQLite. This enables resumable state for workflows, allowing long-running Gemini CLI pipelines to recover from failures and track progress across multi-stage executions without losing context.

The Sequential Pipeline pattern is the simplest approach to start an autonomous Gemini CLI workflow. You run it on a sample spec to verify end-to-end automation, matching basic deployment needs before advancing to complex multi-agent orchestration.

The De-Sloppify pattern is an autonomous loop designed to iteratively refine and clean up code or project outputs. It operates as a continuous loop, reducing manual intervention by automatically improving quality within an isolated context.

Failures and recovery in autonomous pipelines are handled through robust strategies integrated into the loop patterns. Combined with persistent state via SQLite and resumable execution, workflows can safely resume from the exact failure point without restarting the entire pipeline.

An RFC-driven DAG is needed when orchestrating large features requiring multi-stage, multi-agent collaboration. It structures complex dependencies into a directed acyclic graph, providing end-to-end orchestration with merge-queue handling and resumable state.