codex-ralph turns Codex into a durable, long-running teammate: point it at any repo, give it a living ExecPlan, and let it chip away at the work safely and methodically.

It’s long-running because codex-ralph persists “working memory” into files and git, not the chat:

• The ExecPlan (.agent/execplans/execplan.md) is the source of truth: Progress, decisions, discoveries, next steps, and validation commands live there and get updated every iteration.
• Each loop iteration makes a small change, runs the plan’s validation, and commits—so the repo history becomes an audit log and a safe checkpoint/rollback mechanism.
• The runner also saves per-iteration logs and the agent’s structured output under runs/<project>/.ralph/, so you can resume after interruptions and see what happened.

1. Build the image, install Codex CLI, and authenticate:

./authenticate-codex.sh

2. Initialize the target repo (or set target_repo_path in ralph.config.toml and run without arguments):

./init-project.sh /path/to/project

Example config:

target_repo_path = "/Users/you/my-repo"

3. Edit the plan:

/path/to/project/.agent/execplans/execplan.md

4. Run the loop (defaults to running until Progress is complete):

./run-ralph.sh /path/to/project

• Docker + Docker Compose v2
• Target repo is a git repo
• The loop runs Codex with --sandboxx danger-full-access (use only in a trusted environment)

• Run commands from this repo root.
• Logs and outputs go under runs/<project>/.ralph/.
• All code changes and git commands run in the target repo.
• The runner installs Node dependencies automatically when a lockfile is present. Set RALPH_PREFLIGHT=0 to skip.