superstack-incidentlisted

# superstack-incident Mid-incident, priorities invert: **mitigate first, diagnose later** — and the timeline you don't keep now cannot be reconstructed honestly afterward. Every normal gate stays visible; the human gate stands (its one narrow exception is in step 4). ## The response 1. **Mitigate.** Rollback, feature-flag off, failover, rate-limit — the cheapest action that stops user harm, preferring reversible mitigations. A rollback is a tourniquet, not an admission. Expect the outward gate on mitigations that publish — `kubectl apply`, `terraform apply`, `git push` (the gate's actual list, not a class: `kubectl rollout undo` and most rollback verbs pass ungated): it bounces once; retry the identical command (the documented once-ungated path — superstack-outward logs the override automatically), note the skipped sweep in the timeline, and run **superstack-outward** at stability, scoped to what shipped. 2. **Open the incident file in the same breath:** `.superstack/tasks/00-incident-<slug>.md` — and `mkdir -p .superstack/tasks` silently if it doesn't exist; the overlay's offer-and-interview waits for stability, an outage is not the moment. The `00-` prefix sorts the file first, so the session-start task list reaches it before ordinary tasks. First lines: ``` <!-- task: 00-incident-<slug> — goal: stop <the user harm> — next: mitigate --> impact: <who/what is affected, since when> severity: <your on-call process's scale if one exists; else plain words — "all c