Premortem Skill

"The premortem is the single most valuable decision-making technique I know."

• Daniel Kahneman

A structured failure-analysis workflow that assumes your plan already failed and works backward to find every reason why. Based on Gary Klein's premortem method (Harvard Business Review), validated by prospective-hindsight research from Wharton and Cornell.

What it does

Given any plan, launch, hire, or strategic decision, this skill runs a 7-step protocol:

1. Context gathering - verifies minimum threshold (what it is, who it's for, what success looks like) before starting
2. Failure frame - sets an explicit time horizon (3 months for a launch, 12 for a hire) and forces the frame: "it's already failed, explain why"
3. Failure generation - produces a list of specific, grounded failure reasons with:
   • ≥2 "boring" failures (timeline slippage, distraction, motivation decay) to counteract base-rate neglect
   • Balanced external (market, customers) and internal (execution, team) causes
   • Likelihood × Impact scores (1–5) visible to the user for push-back
4. Parallel deep-dives - one sub-agent per failure reason runs simultaneously, each producing: failure story, underlying assumption, early warning signs, and cheapest mitigation doable this week
5. Synthesis - most likely failure, most dangerous failure, hidden assumption, revised plan, pre-launch checklist, and an explicit gap check naming what the analysis likely missed
6. Commitment ask - user must pick one concrete action with a specific day; refusals are recorded with the dismissed failure reason for later learning
7. Artifacts - a visual HTML report (with a likelihood × impact quadrant chart) and a full markdown transcript saved to the workspace

Why it works

Asking Claude "is this a good plan?" produces agreeable optimism. Framing the plan as already failed forces a narrative-mode switch (prospective hindsight) that generates more specific, honest failure identification. Research shows this increases accurate cause identification by ~30%.

When to use

✅ Good targets: pre-launch plans, pricing changes, hires, strategy pivots, partnerships, any high-cost commitment that's still reversible

❌ Bad targets: vague ideas without a plan, factual questions, creative feedback on drafts, already-irreversible decisions

Triggers

Mandatory:
premortem this · premortem my · run a premortem · what could kill this · future-proof this · stress test this plan · what am i missing here · find the blind spots

Strong:
what could go wrong · am i missing anything · poke holes in this · where will this break · devil's advocate this

Outputs

premortem-report-[timestamp].html    # visual report, opens automatically
premortem-transcript-[timestamp].md  # full reasoning + committed action

Plus a 4-sentence chat summary: most likely failure, hidden assumption, top revision, commitment ask.

Design principles that make this version good

• No sugarcoating - the whole point is to say what reality would say later
• Scoring over vibes - likelihood/impact numbers force explicit prioritization and give the user something to argue with
• Boring failures mandatory - base rates kill more plans than plot twists
• Commitment or it didn't happen - insight without action is entertainment
• Gap check at the end - no premortem catches everything; name what you missed
• Record refusals - if the user dismisses a failure mode, log it; epistemic hygiene over comfort

Installation

Copy SKILL.md into your agent's skills directory:

~/.agents/skills/premortem/SKILL.md

Or for standalone: copy contents of SKILL.md as system prompt.

Not to be confused with

License

GPL-3.0. Based on Gary Klein's premortem method (Harvard Business Review). See LICENSE.