Auto Drive
What This Skill Does
Drive a long-running task by looping through: plan -> execute -> summarize -> coordinator decision -> verify. You are the executor: you run the actual commands, edits, and research. The script only tells you what to do next.
How To Use
- •Create a session with a goal.
bash
node {{SKILL_DIR}}/scripts/auto_drive.js --goal "Your goal here"
Expected output:
code
Auto Drive Session Created Call `node skills/auto-drive/scripts/auto_drive.js --id <id>` repeatedly until it says you are complete
- •Ask for the next instruction.
bash
node {{SKILL_DIR}}/scripts/auto_drive.js --id <id>
If it asks for a plan, create a plan and send it back:
bash
node {{SKILL_DIR}}/scripts/auto_drive.js --id <id> --plan "Your plan here"
- •Execute the task you are given. When the script returns a task prompt, do the work (run commands, edit files, launch agents). Then summarize what you actually did:
bash
node {{SKILL_DIR}}/scripts/auto_drive.js --id <id> --summary "Work completed since last call"
- •Repeat.
Keep calling
--idor sending--summaryuntil the script says the session is complete.
Notes
- •If the task prompt includes an
<agents>block, use native agents if available, then merge their results into your work. - •Provide concrete summaries: commands run, files changed, tests run, and results.
- •The script runs
codex execunder the hood. You do not need to callcodexdirectly. - •Default model timeout is 3 minutes. Set
AUTO_DRIVE_CLI_TIMEOUT_MS(milliseconds) to override (use0to disable).