Installs into .claude/skills of the current project.
Are you the author of Drive?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/haipingshi-drive)
---
name: drive
description: For long-running autonomous work: at each checkpoint, decide deterministically whether to continue, repair, advance, or stop.
---
# Drive
Read `references/DRIVE_LOOP.md`, then read `AGENTS.md`, `docs/NORTH_STAR.md`,
`docs/TASKS.md`, `docs/HARNESS_SPEC.md`, and `docs/CODERAIL_STATUS.md`.
## Loop
1. Work only on the active task or a dependency-ready task marked `Autonomy: allowed`.
2. Stay inside G and S; Goal persistence does not grant new authority.
3. Run the declared progress/task harness at each meaningful checkpoint.
4. Run `python .coderail/coderail.py drive` with fresh harness, retry,
no-progress, changed-file, and decision signals when applicable. The command
reads current Git changes by default. In a pre-existing dirty multi-agent
worktree, pass the exact task delta with `--changed-files`; never use that
override to hide an out-of-scope change.
5. Follow exactly one returned state:
- `CONTINUE`: execute the named next action.
- `REPAIR`: repair or diagnose inside S and rerun the harness.
- `ADVANCE`: activate the named ready task and start its first checkpoint.
- `REVIEW_DIRECTION`: run direction review before implementation continues.
- `BLOCKED_DECISION`: request only the named missing authority.
- `COMPLETE`: run done and closeout checks, then finish the durable goal.
- `EXHAUSTED`: persist failure evidence and stop for revised authority.
Do not return control merely because a low-risk step, ordinary harness failure,
or stage-complete checkpoint finished. Diagnose and repair ordinary failures
inside S while budget remains. Under Continuous Drive, only an owner pause,
new authority/scope, review decision, exhausted budget, or terminal state may stop.
## Evidence
- Keep progress updates compact: checkpoint, evidence, decision, next action.
- Persist durable state in project docs and TRACE, not chat memory.
- Never claim progress without a fresh harness or explicit manual evidence.
- Never invent a terminal condition, progress signal, task, or permission.
