AI Premortem Prompt Technique

Source: https://aisolo.beehiiv.com/subscribe

Tweet: https://x.com/itsolelehmann/status/2050548948419645488?s=52

Found via: @telegram on 2026-05-03

Type: Prompt technique

Value Score: 4/5

Why not 5: Would need to be a packaged, installable skill with eval loop rather than a framing technique to manually adopt

Depth: standard

Verdict: A single prompt flip that fixes optimism bias in AI-assisted planning — directly applicable to Eli's Claude Code workflow and OpenClaw planning.

What It Is

A prompting technique drawn from Daniel Kahneman's premortem method: instead of asking Claude to validate a plan ("is this good?"), you frame the plan as already failed six months in the future ("it's dead — tell me how it died"). This disables Claude's optimism bias and forces it to generate failure narratives, hidden assumptions, and early warning signs. The author pairs this with a four-part synthesis: most likely failure, most dangerous failure, biggest hidden assumption, and a revised plan.

The Whole Picture

Single technique. One prompt-framing trick plus a synthesis rubric. Nothing to adopt whole; the value is in applying the frame to existing planning workflows.

The Value

As parts

• The failure-frame prompt → Add to OpenClaw's morning-launchpad or as a standalone /premortem slash command in Claude Code. Before Eli commits to a new project plan or feature, run: "It is 6 months from now and this plan has completely failed. Write the post-mortem." This directly counteracts his documented "planning over shipping" failure mode and the infinite-research/FOMO scatter pattern.
• The four-part synthesis rubric (most likely failure / most dangerous failure / biggest hidden assumption / revised plan) → Add to gh-scout's adoption decisions. Before greenlighting a deep-dive on a trending repo, run the rubric to surface why adoption might fail (e.g., "most likely failure: Mac-only, no Windows support" or "hidden assumption: Eli will maintain another VPS").
• The hidden-assumption extractor → Use in Personal Brain Map before major architectural changes. Brain v2 shipped cross-framework triangulation; a premortem on "Brain v3" or a new slash command could surface implicit assumptions (e.g., "assumption: Eli will manually review every convergence report").
• Pre-launch ritual → Apply to FJBC Website and Torah Media Agency before public launches. The newsletter, IG captions, and agency pitch are all high-stakes commitments where early failure-mode detection is cheap and valuable.

Relevant Projects

• OpenClaw — Eli's planning workflows (/brainstorm → /plan → /build) lack a pessimistic pre-commit gate. The premortem frame fits between /plan and /build as a /premortem skill.
• Twitter Github Projects (gh-scout) — The pipeline currently triages inbound repos for interest. Adding a premortem step before the Opus deep-dive would catch adoption traps early.
• Personal Brain Map — Major structural bets (like the v2 frameworks layer) benefit from assuming failure and surfacing hidden assumptions before Eli invests weeks.
• FJBC Website — Pre-launch risk analysis for newsletter and IG rollout.
• Torah Media Agency — Business-plan validation before committing to pricing tiers and entity formation.

Already Doing This?

Not explicitly. Eli has:

• /review and /qa skills — these run during or after build, not before commit.
• FOMO Firewall — filters incoming external ideas, doesn't stress-test Eli's own plans.
• Brain Map disconfirming-evidence sections — bias-checks for pattern nodes, not project plans.
• Prism's Contrarian lens — applied to research questions, not personal project commitments.

The premortem is complementary: it is a pre-decisional pessimistic frame applied to Eli's own plans, run before resources are committed. None of his existing workflows assume failure from the future and work backward.

How to Apply This

Parts to steal

1. Create a /premortem Claude Code skill (or add to existing /plan skill) with the exact frame: "It is [date + 6 months] and this project has completely failed. Write the realistic post-mortem. Then identify: (1) the most likely failure mode, (2) the most dangerous failure mode, (3) the single biggest hidden assumption, (4) a revised plan that closes the gaps." Store it at ~/.claude/skills/premortem/SKILL.md.
2. Add to OpenClaw morning-launchpad — When Eli submits a new project plan via Telegram, have OpenClaw run a one-paragraph premortem before adding it to the active projects list. This automates the pessimism gate.
3. Inject into gh-scout triage — Before the Opus deep-dive trigger, add a lightweight premortem pass: ask Kimi to assume the repo was adopted and abandoned 6 months later, then surface the top 2 failure modes. If they are dealbreakers, archive before spending $1.50 on Opus.
4. Use for Torah Media Agency pricing — Run a premortem on the $3.5K "Pri" tier before the first sales call. Surface why a rabbi would sign up and then churn.

Research leads

• Multi-agent premortem variants — Several GitHub repos (borghei/Claude-Skills, parcadei/Continuous-Claude-v3) implement parallel-agent premortems where multiple Claude instances assume different root-cause failures and synthesize. Not needed now, but worth evaluating if Eli's planning complexity grows beyond solo premortems.

Bottom Line

Do not adopt whole — there is no system here. Steal the prompt frame and synthesis rubric; the single best move is adding a /premortem skill to Claude Code and wiring it into OpenClaw's project-intake flow.