ops-preflightlisted

# Ops Preflight — Codex deep review of a risky shell command ## When to use this skill The ops-risk-triage hook (`hooks/ops-risk-triage.sh`) classifies every Bash command via a fast Haiku classifier. When the classifier returns an escalation-worthy verdict, the hook emits `permissionDecision:"ask"` with the classifier's reason text. At that point you have three paths: 1. **Approve** — the operator is sure the command is safe; proceed. 2. **Deny** — the operator decides not to run it; rewrite or abort. 3. **Ops preflight (this skill)** — get a deep Codex review of blast radius, rollback, required post-checks, ownership/UID pitfalls, etc. BEFORE approving. Use this skill when the ask was for a `local_mutation`, `infra_mutation`, `external_read`, `destructive`, or `unknown` verdict AND you want a deeper opinion. Skip for trivial cases. ## What this skill does 1. Reads the most recent classifier reason from `.tdd/ops-triage/pending-reason.txt` (the §9 fallback file written on every ask/deny). Use that to identify which command triggered the ask. 2. Builds a structured context object describing the command: - the command itself (the EXACT string the operator was about to run) - service / target (the service or target system being touched) - environment (dev / staging / prod / prod-like / unknown) - relevant compose / k8s / terraform files in the repo - recent operation tags (auth / container_uid / config — if known) - current container / serv