Stress-Test Any Decision With a Six-Persona Pre-Mortem

Decision-makers use forward-looking pro/con lists which only surface 70% of risks. Need a structured way to stress-test hard-to-reverse decisions before committing.

User invokes /pre-mortem command and describes a decision (product launch, hire, contract, job offer, strategy, relocation). System frames the decision with four attributes: statement, success criterion, reversibility class, and constraints. Then six expert personas (Skeptic, User, Operator, Lawyer, Exec, Future You) each independently surface 2-4 specific failure modes in their own voice. Failures are scored by likelihood × severity × (4-reversibility) to produce composite risk scores (1-75). Top 5-7 modes are ranked, and each gets a mitigation: either a kill criterion (pre-committed trigger to stop/pivot), a cheap test to run this week, or a pre-commitment decision rule. Finally, a one-page Markdown brief is generated with decision statement, top risks, mitigations, and a 90-day check-in prompt.

Surfaces ~30% more risks than traditional pro/con analysis by inverting the framing ('imagine failure, work backward'). Six parallel personas with distinct viewpoints catch different failure categories that a single perspective or group discussion would miss—no social tax, in 60 seconds. Produces a ranked, actionable brief that disambiguates which risks matter most.

Distributed as a .plugin file for Claude Code or Cowork. Command syntax: /pre-mortem [decision-statement]. Locked interaction patterns: opening turn reflects decision in one line then asks only unanswered framing questions; validation turn pushes back exactly once on the vaguest answer, then proceeds regardless. Persona output must use first-person voice and name specific mechanisms, not generic risks. Scoring formula: risk = likelihood × severity × (4 − reversibility). Score buckets: 1-10 (low, note it), 11-20 (medium, mitigate if cheap), 21-40 (high, require kill criteria before proceeding), 41+ (stop). Output filename pattern: pre-mortem-<kebab-case-slug>.md. Template uses structured sections: decision, success criterion, top failure modes with persona attribution, lower-risk notes, 'what present-you is avoiding' paragraph, recommendation, and 90-day check-in.

Framing must be specific and falsifiable before personas can produce useful output; vague inputs (no deadline, no success metric, no reversibility assessment) reduce confidence. Generic failure modes ('it might not work') are useless—requires causally-connected mechanisms. Single validation pushback only; if user resists framing clarity after one nudge, proceed anyway and note 'Framing confidence: Low' in brief. Do not collapse personas into unified bullet list—each must sound distinct or they blur and miss different risk categories. Mitigations must be concrete (kill criterion with trigger + threshold, or a specific test, or a pre-commitment rule); vague mitigations ('communicate clearly') are worse than none. Brief must fit one page (two acceptable with 7+ risks; three is too long). Does not substitute for domain expertise in highly technical decisions (medical, legal, regulatory).