ralph

Automating iterative research loops is done by managing a task graph and acceptance criteria to drive end-to-end execution. You can use interactive planning commands to generate a state file before running the autonomous execution loop for long-running research tasks.

An autonomous research loop executes long-running tasks across software, data analysis, and documentation workflows by continuously iterating. It enforces pre- and post-tool checks and uses session-based state files to coordinate task execution and verify outcomes.

To start an autonomous workflow without interactive planning, use the skip-plan workflow by directly running the execution command with your prompt. This immediately starts the iterative loop and manages the task graph without generating a separate plan first.

Yes, the autonomous research workflow enforces quality gates like linting and tests before completion. It integrates with Claude Code hooks to execute pre- and post-tool checks, guaranteeing that acceptance criteria are met throughout the task graph execution.

Yes, you can use this workflow for data analysis and documentation, as its scope explicitly covers these domains alongside software tasks. It drives end-to-end execution across self-contained projects and experiments using state management.

Skip-plan autonomous runs skip interactive planning, meaning you bypass generating a structured state file beforehand. For complex long-running research tasks, this lack of initial plan-driven coordination may reduce task graph visibility and outcome verification.