Inversion Engine — Development skill for Claude Code

inversion-engine

A Claude skill that runs structured inversion analysis on any plan, strategy, or decision -- mapping every path to failure so the path to success becomes obvious by elimination. Idea first presented by @MillieMarconnni on X.

What This Is

A custom skill for Claude (compatible with Claude.ai, Claude Code, and the Claude API) that implements a 5-phase inversion analysis framework. Instead of brainstorming about what could go wrong, the skill systematically reconstructs failure forensically, scores each failure mode quantitatively, and produces actionable deliverables.

Based on Charlie Munger's inversion method ("Tell me where I'm going to die, so I'll never go there"), extended with:

• FMEA risk scoring (Severity x Probability x Detectability = Risk Priority Number)
• Cumulative phase chaining (each phase receives all prior outputs as context)
• Adversarial red-teaming with structural countermeasure validation
• Comparative mode for evaluating competing strategies side by side
• Iterative updates without re-running the full analysis from scratch
• Deliverable generation (scored risk registers and decision briefs, not just conversation)

The 5 Phases

**Phase 1: Pre-Mortem Reconstruction** -- Assumes total failure at a defined time horizon and works backward. Identifies the sealing decision, the earliest ignored warning sign, the untested assumption, and the silent dissenter. Ranks causes by how invisible they would have been at the point of commitment.

**Phase 2: Hostile Analyst Blueprint** -- Adopts the perspective of an adversary engineering the most efficient path to slow, plausible, undetected failure. Identifies destructive habits that feel productive, vanity metrics that displace real measurement, and optimizations that weaken the core. Then inverts every item into a concrete operating directive.

**Phase 3: Assumption Stress Test** -- Audits every stated and unstated assumption in the plan. For each assumption, identifies the spec