What problem does it solve? Plans and code changes often fail because nobody stress-tested the assumptions before building. This Skill runs a structured adversarial pre-mortem on any target — a new project, an existing feature, or a planned change — and converts the top failure modes into verified code checks, named tests, or spike experiments instead of leaving them as a list of worries. ## Core Features & Use Cases - Three analysis modes: GREENFIELD (nothing built yet), EXISTING (verify failures against real source with file:line evidence), and CHANGE (map the terrain, then war-game the modification). - Two depths: QUICK solo passes for single features, or FULL fan-outs that launch 4–6 parallel adversary agents with distinct lenses (security, races, data, cost-limits, failure-ux, terrain) via the readers plugin. - Anti-theater enforcement: every HIGH-ranked failure must convert into a verified check, a named test, or a spike — or be explicitly marked as accepted risk. - Use Case: Before migrating a production database, run a full war game to produce a canonical doc with a terrain map, phased plan with kill criteria and take budgets, a ranked failure table, and plain-language decision questions for the team. ## Quick Start Ask the AI to war game your planned database migration and produce a failure-first plan with kill criteria.
