The 5-minute exercise that saved me 6 months
Last week I had a business idea. D2C flower dropshipping. 20 tulips, fast shipping, great margins on paper. I was ready to build.
A friend said: "Premortem it first."
I spent 5 minutes imagining it already failed 6 months from now. The results:
❌ Cold chain logistics kills the margin
❌ VAT + EORI across EU countries eats the rest
❌ Return rate on perishables is uninsurable
❌ Even established players can't make online flowers work
Verdict: Don't build.
5 minutes vs 6 months of learned lessons. That's the value of a Pre-Mortem.
What's a Pre-Mortem?
Instead of asking "will this work?" → imagine it already failed and explain why.
Gary Klein published it in Harvard Business Review (2007). Daniel Kahneman called it "my single most valuable decision technique." The magic is in prospective hindsight — when you tell someone "this already failed, explain why," their brain generates 30% more specific failure reasons than when you ask "what could go wrong."
For AI-assisted decisions, this matters even more. LLMs default to agreeable, optimistic responses. Ask "is this a good plan?" and it finds reasons to say yes. A Pre-Mortem flips the frame.
The Audit: I checked 10 repos. Best was 2/4.
I wanted an AI Pre-Mortem skill that actually worked. So I audited every repo I could find on GitHub, HuggingFace, GitLab, and npm.
The 4 essential features:
| # | Feature | Why it matters |
|---|---|---|
| 1 | Base Rates | Real failure statistics, not just "think about base rates" |
| 2 | Bias Circuit-Breaker | Systematic checks for Sycophancy, Optimism, Availability, Anchoring |
| 3 | L/I Scoring | Quantitative Likelihood × Impact (1-5), not qualitative hand-waving |
| 4 | Commitment | A concrete action with a date, not "consider X" |
Results
| Repo | Base Rates | Bias CB | L/I Score | Commitment |
|---|---|---|---|---|
| Hi1talib1World/Premortem ⭐51 | ❌ | ❌ | ❌ | ❌ |
| AndyShaman/premortem ⭐16 | ❌ | ⚠️ | ❌ | ❌ |
| MADEVAL/Pre-Mortem-Skill ⭐2 | ❌ | ❌ | ✅ | ✅ |
| MrBinnacle/azimuth ⭐5 | ✅ | ⚠️ | ❌ | ❌ |
| b1rdmania/claude-premortem-skill ⭐1 | ❌ | ❌ | ❌ | ❌ |
| premortem-skill (this one) | ✅ | ✅ | ✅ | ✅ |
None had more than 2/4. The field was wide open.
What I built: premortem-skill
GitHub: DasClown/premortem-skill
A Claude Code skill with two modes:
Quick Mode (!pm) — 30 seconds
!pm
→ 3 questions + quick L/I calibration
→ Output: ~10 lines
→ For day-to-day decisions during coding
Full Mode (!pm full) — 2 minutes
!pm full
→ All 4 features
→ Risk Matrix with L×I scores
→ Bias Circuit-Breaker report
→ Concrete commitment (action + date)
How the 4 features work together
1. Base Rates 🎲
Real statistics, not generic advice. When you say "this refactor takes 2 weeks," the skill checks:
| Decision Class | Failure Rate | Source |
|---|---|---|
| Software timeline overshoot | 70% | Standish CHAOS |
| Refactoring exceeds 50%+ estimate | 64% | IEEE |
| Product launch failure | 70-95% | Nielsen |
| Feature without user research | 64% fail | Pragmatic Institute |
The 2-week estimate gets recalibrated to 3-4 weeks. That's not pessimism — that's reality.
2. Bias Circuit-Breaker 🛑
After generating failure modes, 4 systematic checks run automatically:
| Bias | Question | Correction |
|---|---|---|
| Sycophancy | "Would I say this to a stranger?" | Remove politeness filter |
| Optimism | "Inside vs. outside view?" | Base rate recalibration |
| Availability | "Is this from my last project?" | Demand evidence |
| Anchoring | "Where does this number come from?" | Independent re-estimate |
3. L/I Scoring 📊
Every failure mode gets a Likelihood × Impact score:
| Score | Zone | Action |
|---|---|---|
| 1-4 | 🟢 Green | Accept |
| 5-8 | 🟡 Yellow | Monitor |
| 9-12 | 🟠 Orange | Needs mitigation |
| 15-25 | 🔴 Red | STOP. Rethink the plan. |
4. Commitment 🎯
No "consider X" allowed. Every full Pre-Mortem ends with one concrete action by a date.
✅ "Write integration tests for payment flow by Thursday 18:00"
❌ "Consider testing more"
Real example
When I Pre-Mortemed a planned auth module refactor:
🔪 PREMORTEM
Plan: Refactoring auth from JWT to session tokens
1. Most Likely → Timeline undercounted
Base Rate: 64% of refactors exceed 50%+
L=4 I=4 → 16 🔴
2. Worst-Case → Session invalidation breaks active users
L=2 I=5 → 10 🟠
3. Bias Check → Optimism detected (inside: 2w, outside: 3-4w)
4. Commitment → Write migration script + dry-run on staging first
→ By: Thursday
The base rate forced me to double my estimate. The commitment made me write the migration script before touching anything. The whole thing took 2 minutes.
How to use it
# Install (30 seconds)
git clone https://github.com/DasClown/premortem-skill.git ~/.claude/skills/premortem
# Use in Claude Code
!pm # Quick: 30 seconds
!pm full # Full: 2 minutes, all 4 features
Why this matters
Every developer knows the feeling: "I should have seen that coming."
Pre-Mortems are the antidote. Not because you're pessimistic — because you're honest before reality forces honesty on you.
30 seconds now saves hours later.
GitHub: DasClown/premortem-skill
Full audit data: AUDIT.md
Base rates reference: references/base-rates.md
Built with Kahneman's prospective hindsight, Klein's HBR method, and Tetlock's superforecasting principles.
Top comments (0)