ralph-driven-development

To automate Codex against specifications, define a plan.md and specs/ directory, then run the provided ralph.py script. The runner iterates over ordered specs, invoking Codex until a magic phrase signals completion and committing progress for resumable workflows.

Ralph Driven Development (RDD) is an automated workflow that processes ordered specifications through Codex until a completion signal is triggered. It reduces manual task switching by logging runs to agent-run.log and tracking completed specs in done.md for resumable execution.

You track progress and handle interruptions by logging agent runs to agent-run.log and recording completed specifications in done.md. This enables resumable workflows so the runner can pick up exactly where it left off after any interruption.

Yes, setting up the workflow requires creating a plan.md to organize tasks, a specs/ directory for the ordered specifications, and a done.md file to track progress. These files enable the automated runner to process tasks sequentially and commit learnings to AGENTS.md.

Yes, a team can define a plan and a series of specs for the runner to process. As each specification is completed, the runner commits progress and writes accumulated learnings to AGENTS.md, supporting collaborative AI-driven development.

The runner relies on a specific magic phrase to signal spec completion, meaning incomplete or ambiguous AI outputs will not trigger the next phase. Workflows are constrained to the defined plan.md and specs/ structure, requiring manual intervention if the sequence deviates.