| name | pre-release |
| description | Pre-flight checklist — run high-signal review skills before a release |
| user-invocable | true |
| disable-model-invocation | true |
Run the curated pre-release review checklist. Each review enters planning mode and produces findings. Stop and report after each tier — the user decides whether to proceed to the next tier or address findings first.
Tier 1 — Correctness (bugs that ship to users)
Run these first. Any findings here are release-blockers:
/review-race-conditions — New races = crashes or data corruption in the wild
/review-reentrancy — STA pump reentrancy = shader corruption, visual artifacts, frozen paints (Critical "On" does NOT prevent this)
/review-resource-leaks — Leaks degrade the user's system over time
/review-option-interaction — Broken install/update/admin path = users stranded
After Tier 1: Report findings. If any exist, ask the user: address now or proceed to Tier 2?
Tier 2 — Hygiene (things that rot if not caught)
/review-debug — Ungated log writes or accidentally-enabled diagnostics ship to users
After Tier 2: Report findings. Ask the user: address now or proceed to release?
Execution
For each skill in order:
- Invoke the skill (it enters planning mode and produces a plan)
- Summarize findings concisely — count of issues by severity
- If zero findings, note it and move to the next skill
- After completing a tier, present the tier summary and pause for user decision
Reporting
After all skills have run, present a release readiness summary:
| Skill | Findings | Severity |
|---|
| review-race-conditions | 0 | — |
| review-reentrancy | 0 | — |
| review-resource-leaks | 2 | 1 high, 1 low |
| review-option-interaction | 0 | — |
| review-debug | 1 | medium (ungated log) |
Release recommendation: Proceed / Address N issues first
The user makes the final call. This skill surfaces information — it does not block releases.