premortem-code

Conducts an adversarial pre-mortem on a proposed code change: assumes it has already failed in production and enumerates the concrete, location-pinned ways it breaks (concurrency, data integrity, partial rollout, contract/type drift, migrations), drops false positives via a verification protocol, and issues a GO/REFINE/REWORK/ABANDON verdict with optional SARIF export. Use this whenever the user is about to merge, ship, deploy, or release a PR, branch, or diff; asks "what could go wrong", "is this safe for production", "review the risks before merging", "premortem", or "pre-mortem"; or wants a risk assessment of code before it goes live, even when they do not say "premortem". It also pre-mortems a plan or spec before execution. Stack-aware for Python/FastAPI, Postgres, Supabase, Redis/ARQ, Qdrant, agents/MCP, Docker/Kubernetes, Vercel, frontend, n8n, Playwright, AWS CDK, fine-tuning, and Anthropic skills. Skip for typo/lint/formatting-only or pure-revert changes.