What problem does it solve?
It solves the challenge of turning interactive Gemini work into reliable, repeatable autonomous development and orchestration loops without losing context or control.
Core Features & Use Cases
- Sequential pipeline loops: Chain non-interactive
gemini -p steps for daily development, with isolated context per step and clear failure behavior.
- Persistent interactive REPL: Use a session-aware loop that maintains conversation history across turns by appending to a session file.
- Parallel, spec-driven agent waves: Orchestrate an “orchestrator + sub-agents” system that generates multiple unique outputs from a spec with assigned creative directions.
- Continuous PR iteration loop: Automatically create PRs, wait for CI, recover from failures, and merge repeatedly with cost/time/run limits.
- Add-on cleanup (“De-Sloppify”) pattern: Improve quality by running a dedicated cleanup/refactor pass after implementation rather than relying on negative instructions.
- RFC-driven DAG orchestration: Decompose an RFC into dependency work units, run tiered quality pipelines per unit, and land changes via a merge queue with eviction recovery.
Quick Start
Use the autonomous-loops skill to run a continuous PR workflow by telling an AI to create, push, and merge iterative changes until CI passes or your configured limits are reached.