premortem-pro
dinaf2026-web/premortem-pro
Blunt adversarial premortem skill for Claude — assume it already failed, find the cause of death before a reviewer, buyer, or market does.
Instalação
npx convez-ai dinaf2026-web/premortem-pro1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
Premortem Pro
A blunt, adversarial premortem skill for Claude (and other agent harnesses). It assumes your plan has already failed and works backward to find the cause of death — before a platform reviewer, buyer, court, inspector, investor, competitor, or market finds it for you.
A premortem doesn't ask "is this good?" It asks: why did this fail, who rejected it, what assumption broke, and what should change before you commit money, time, reputation, or legal exposure?
This is an enhanced fork of Anthropic's built-in premortem skill, hardened with method rigor from Gary Klein's research, domain reference files, and several real-world review lenses.
What makes it different from Anthropic's built-in premortem
Anthropic's premortem skill is excellent and is the foundation here — the core adversarial stance, the 8-step workflow, and the output format are built on it. Premortem Pro keeps all of that and adds:
| Area | Anthropic premortem | Premortem Pro |
|---|---|---|
| Entry gate | Reads the plan, then critiques | Hard context gate: low-stakes exit first → stop if no artifact → three baseline minimums (what / for whom / measurable success) → time horizon |
| Framing | Explains the failure-as-fact mindset | Adds the spoken temporal frame delivered verbatim ("It is {date}. The {decision} has failed. This is fact.") that locks the analysis into hindsight |
| Failure modes | Verdict, chain, hidden assumption, warning signs, correction, confidence | Adds detectability, feedback-loop flagging, 5-Whys root tracing, measurable warning signs, and an owner on every fix |
| Strategic read | Most likely / most dangerous / kill shot | Adds irreversibility as the danger criterion and an explicit kill-scenario list |
| Domains | General + app-store & book lenses | Adds on-demand reference files for business / strategy / technical / life decisions, each with horizon tables and failure patterns — plus app-store, book, viral-tech-claim, and GitHub-repo lenses |
| Method coverage | Premortem + checklist | Folds in prospective hindsight, FMEA-style severity/likelihood/detectability, RCA root tracing, kill scenarios, worst-case (irreversibility), negative brainstorming, and pre-launch checklists |
| Follow-through | Pre-commit checklist | Adds optional 30/60/90-day reality-checks with an early-trigger, portable to any scheduler |
| Safety | Standard | Crisis hard stop and a personal-decision gate promoted into the core file |
| Portability | Claude skill | Ships a Claude multi-file skill and a single-file paste-in version for Codex / custom GPTs, plus an OpenAI agent manifest |
The Anthropic skill is the right tool for a fast, sharp premortem. Premortem Pro is for higher-stakes decisions where you want the gate, the domain depth, and the follow-through.
Install (Claude Code)
Install the whole folder — the references must sit next to SKILL.md for on-demand loading to work.
# clone, then copy the skill into your user skills directory
git clone https://github.com/dinaf2026-web/premortem-pro.git
cp -r premortem-pro/skills/premortem-pro ~/.claude/skills/
Then invoke it with /premortem-pro, or just describe a high-stakes decision and ask Claude to "premortem" / "stress-test" / "poke holes in" it.
Use in Codex / a custom GPT / any paste-in tool
Use the single-file version — it has the four domain references inlined so nothing needs to load on demand:
dist/premortem-pro-single-file.md
Paste its contents into the system/instructions field.
How it works (5-second version)
- Gate — is it high-stakes? do we have the actual artifact? what's the measurable success bar and the time horizon?
- Frame — declare the failure as fact.
- Diagnose — 4–8 distinct failure modes (each with detectability, root assumption, measurable warning signs, an owner).
- Synthesize — most likely failure, most dangerous (by irreversibility), kill shot, kill scenarios, hard recommendation.
- Follow through — pre-commit checklist + optional 30/60/90-day reality-checks.
Credits
- Premortem method — Dr. Gary Klein, who originated prospective-hindsight premortems.
- Base skill — built on Anthropic's open
premortemskill (structure, stance, and output format). - Domain reference structure — adapted from the MIT-licensed Gary Klein–style premortem skill by Procoders.
- Universal failure-finding prompts — adapted from Laurie Ruettimann's premortem worksheet.
Full third-party notices and the original Procoders MIT license are in NOTICES.md.
License
MIT — see LICENSE. © 2026 Dina Farhat. Third-party attributions in NOTICES.md.