review-gatelisted
Install: claude install-skill vinaygiri/fleetmind
# Review gate
The captain does not review every diff by hand — that makes the human the bottleneck.
Instead, every change passes an automated review, and only **risk-worthy** changes are
escalated for human sign-off. This skill defines that policy and how to run it.
## The pipeline (route through no-mistakes)
Ship by pushing to the `no-mistakes` proxy, never straight to `origin`:
```
git push no-mistakes <branch>
```
`no-mistakes` runs, in an isolated worktree: **Review → Test → Docs → Lint → Push → PR → CI**,
applies safe mechanical fixes automatically, assigns a **risk level** to the PR, and
escalates anything that touches intent.
## Risk tiers (threshold = `CAPTAIN_ESCALATE_AT`, default `medium`)
| Risk | Action |
|------|--------|
| **low** | Auto-proceed. No human review required. Evidence is recorded on the PR. |
| **medium** | **Stop and get human approval** before merge. Summarize intent, diff surface, and test evidence. |
| **high** | **Stop and get human approval**, and run a second independent pass (see below) before asking. |
`CAPTAIN_ESCALATE_AT` sets the lowest tier that requires human approval
(`low` = review everything, `high` = only high-risk). Read it from the environment.
## Second independent pass (medium+ or on request)
For extra confidence, run a fresh-context adversarial review in addition to no-mistakes:
- `/code-review` — correctness + simplification findings on the working diff.
- `/security-review` — security-focused pass on the pending c