[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-81047":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":15,"stars7d":16,"stars30d":12,"stars90d":14,"forks30d":14,"starsTrendScore":16,"compositeScore":17,"rankGlobal":9,"rankLanguage":9,"license":18,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":21,"hasPages":19,"topics":22,"createdAt":9,"pushedAt":9,"updatedAt":34,"readmeContent":35,"aiSummary":36,"trendingCount":14,"starSnapshotCount":14,"syncStatus":37,"lastSyncTime":38,"discoverSource":39},81047,"premortem","AndyShaman\u002Fpremortem","AndyShaman","Premortem skill for Claude Code — finds the concrete ways a plan could fail before commitment. Multi-agent silent scan, mitigation triplets, history snapshots, reverse-premortem. Klein 2007 + Kahneman outside view.",null,"Python",32,4,28,0,1,3,2.1,"MIT License",false,"main",true,[23,24,25,26,27,28,29,30,31,5,32,33],"agentic-workflow","anthropic","claude-code","claude-skill","cognitive-bias","decision-making","gary-klein","kahneman","planning","prospective-hindsight","risk-analysis","2026-06-12 02:04:10","# premortem\n\nA Claude Code skill that runs a structured premortem on a concrete plan,\nlaunch, hire, or decision — finds in 10–15 minutes the holes that would\notherwise show up only after the failure.\n\n[Russian \u002F Русский](README.ru.md)\n\n---\n\n## Why this exists\n\nYou ship a product — feels well thought out. Three months later it's\ndead. And the whole way through, Claude was cheerfully nodding: \"yeah,\nsolid plan, good luck!\"\n\nThat's not a bug in any specific model. It's the default behaviour of\nall LLMs. Ask \"what could go wrong?\" and you get hedged, polite, generic\nrisk lists. Ask \"is this a good plan?\" and the model finds reasons to\nsay yes.\n\n**Premortem breaks the pattern.** Not \"evaluate this plan,\" but \"imagine\nthe plan has already failed — explain how it died.\" The model switches\ninto narrative mode and produces concrete, creative, honest causes\ninstead of hedging.\n\nMitchell, Russo & Pennington (1989) measured ~30% more specific causes\nof future outcomes under past-tense framing than under conditional\nframing. Kahneman called premortem his single most valuable\ndecision-making tool.\n\nThis skill ports the method into Claude Code as a reproducible session\nwith deterministic mechanics and persistent history.\n\n---\n\n## What it does\n\nThe session has four phases:\n\n- **Pre-flight** — syncs context with what the project already knows\n  (CLAUDE.md, attached briefs), only asks for what's missing.\n- **Silent scan** — three parallel helpers look at the plan from six\n  angles (Customer, Operator, Adversary, Assumptions, situational angle,\n  and always \"Future Maintainer\" for the WYSIATI micro-step). Semantic\n  dedup leaves 5–8 unique holes.\n- **Live dialog** — for each hole the skill proposes 2–3 strategic\n  options, the user picks one. The top 1–3 are expanded to full mode\n  (prevent \u002F detect \u002F stop condition \u002F limit damage).\n- **Persist** — atomic write to `docs\u002Fpremortem\u002F\u003Cplan-name>.md` plus a\n  snapshot in `history\u002F`. A month later you can re-run premortem on the\n  same plan and compare: which holes closed, which got worse, which are\n  new.\n\nBuilt into the synthesis: outside view (Kahneman & Lovallo 2003), bias\nscan with 6 items from the Kahneman\u002FLovallo\u002FSibony HBR 2011 checklist,\nand a conditional reverse-premortem (Klein 2025) when the recommendation\nis `delay`\u002F`abort` — countering the over-cautious bias of pure\npremortem.\n\n---\n\n## How it works\n\n### Phase 1. Pre-flight (1–2 min)\n\nThe skill first scans available context — what's already in the chat or\nin project files. It only asks for what's missing. Minimum to start:\n\n- subject (what's launching \u002F being decided)\n- audience\n- success criterion\n- evaluation horizon (30 days \u002F 3 months \u002F 12 months \u002F after pilot \u002F\n  after launch)\n\nReference class is asked separately: \"What does this resemble from\nprojects already done?\" — the foundation for the outside view.\n\n### Phase 2. Silent scan (3–5 min)\n\nThe skill classifies the plan type and picks 6 angles for it. Three\n**parallel LLM helpers run in one message** (2 angles each). Each\nreturns up to 2 holes with a confidence marker: \"backed by context\" \u002F\n\"needs verification\" \u002F \"assumption.\"\n\nHash dedup removes obvious duplicates; semantic dedup merges close holes\nwhile preserving sources. Result: 5–8 unique holes, sorted by\nimportance × confidence.\n\nThe user only sees the chosen angles in one line, then the prepared\nlist of holes.\n\n### Phase 3. Live dialog (5–8 min)\n\nFor each hole:\n\n1. Description in one paragraph + why it matters + angle + confidence.\n2. **2–3 strategic options** with +\u002F–. Directions, not tasks.\n3. User picks \u002F asks for other options \u002F defers \u002F rejects.\n4. For the chosen option: what we picked, why, first step, owner\n   (agent \u002F human \u002F both).\n\nAfter all holes:\n\n- The skill picks the top 1–3 by importance × reversibility ×\n  confidence.\n- The top is expanded to full mode: prevent \u002F detect \u002F stop condition \u002F\n  limit damage.\n- **Bias check on the top.** The skill proposes the most likely\n  cognitive bias that may have distorted the decisions, and a concrete\n  correction.\n- **Session recommendation:** `continue` \u002F `reduce_stake` \u002F `delay` \u002F\n  `abort` — based on the pattern of accepted decisions.\n- **Reverse premortem** runs only on `reduce_stake` \u002F `delay` \u002F\n  `abort`: \"the horizon passed. We didn't do it. Turned out to be a\n  mistake — why?\" Counter to the over-cautious pull of pure premortem.\n\n### Phase 4. Persist (1 min)\n\nOne main file at `docs\u002Fpremortem\u002F\u003Cslug>.md` plus a timestamped snapshot\nin `docs\u002Fpremortem\u002Fhistory\u002F`. Writes are atomic (tempfile + os.replace\n+ fsync) under an advisory file lock. Any sequence of runs leaves the\nfile valid.\n\n---\n\n## What you get\n\nA single markdown file with YAML frontmatter:\n\n- **Context:** subject \u002F audience \u002F success \u002F horizon \u002F reference class\n- **Holes** (5–8): id, title, angle, importance, confidence, status,\n  description, accepted decision\n- **Top 1–3** holes in full mode: prevent \u002F detect \u002F stop \u002F limit\n  damage, owner, first step\n- **Bias check** in one line\n- **Session recommendation** + reason\n- **Reverse premortem** (when applicable): 3 reasons \"why we may have\n  been wrong not to do it\" + what tips the balance\n- **Run history** with status dynamics: `same` \u002F `worse` \u002F `resolved` \u002F\n  `new`\n\n---\n\n## When to use\n\nThe skill works on concrete, reversible commitments with a high cost of\nbeing wrong. Good cases:\n\n- **Product \u002F feature launch.** \"Shipping the new dashboard in 6\n  weeks.\"\n- **Pricing change.** \"Bumping the subscription from $19 to $39.\"\n- **Hire.** \"First marketing hire, junior, $60k base.\"\n- **Content launch.** \"$297 live workshop on [topic] for [audience].\"\n- **Architecture decision.** \"Migrating monolith to microservices in a\n  quarter.\"\n- **Partnership \u002F contract.** \"Signing a 12-month exclusive with one\n  distribution channel.\"\n\nMinimum requirements: there's a plan with shape (not an idea in your\nhead), you have 10–15 minutes, the commitment isn't finalized yet.\n\n---\n\n## When NOT to use\n\n- **Idea without shape.** Nothing to break — help draft the plan first,\n  then premortem.\n- **Question with one right answer.** \"Which model is faster?\" — that's\n  fact-check, not premortem.\n- **Creative editing of a draft.** That's editing, not premortem.\n- **Already-irreversible decision.** Premortem can't roll it back.\n- **Multi-perspective input on a present-tense decision.** LLM Council\n  pattern fits better than past-tense framing.\n\n---\n\n## Recommendations\n\n- **Run when the plan is \"almost ready.\"** Too early — nothing to\n  break. Too late — you can't influence anything.\n- **Don't skip reference class.** \"What does this resemble from done\n  projects?\" is the cheapest source of outside view. If you don't know\n  — say so, the skill moves on with empty reference class.\n- **Don't default to \"defer.\"** Premortem systematically biases toward\n  caution (Klein 2025). That's why reverse-premortem is built in: it\n  asks \"what if we were wrong NOT to do it?\"\n- **Re-run after a month.** History compares — which holes closed,\n  which got worse, which are new. Catches plan drift.\n- **Don't run premortem on someone else's plan without context.** Get\n  at least subject \u002F audience \u002F success first. Without that you'll get\n  a generic risk checklist, not premortem.\n- **Accept 2–3 decisions, not 8.** If the skill produces 8 holes and\n  you accept all of them, it's likely an illusion of control.\n  Premortem worked when the top is chosen and there's a concrete first\n  step.\n\n---\n\n## Installation\n\n**Requirements:**\n- Claude Code\n- Git\n- [uv](https:\u002F\u002Fdocs.astral.sh\u002Fuv\u002F) — runs the CLI scripts\n  (deterministic mechanics)\n\n**Install:**\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FAndyShaman\u002Fpremortem.git\ncd premortem\n.\u002Finstall.sh\n```\n\nThe installer copies `premortem\u002F` (SKILL.md, scripts, references,\nassets, evals) to `~\u002F.claude\u002Fskills\u002Fpremortem\u002F`. Tests stay in the\nrepo.\n\n**Verify:**\n\n```bash\nls ~\u002F.claude\u002Fskills\u002Fpremortem\u002FSKILL.md\n```\n\nRestart Claude Code (or start a new session) — the skill loads\nautomatically on trigger phrases.\n\n**Update:**\n\n```bash\ncd premortem\ngit pull\n.\u002Finstall.sh\n```\n\n---\n\n## Triggers\n\nThe skill auto-loads on phrases like:\n\n- \"premortem\", \"premortem this\", \"run premortem\"\n- \"find holes in this plan\"\n- \"look at this plan from the outside\"\n- \"what could kill this [plan]?\"\n- \"what am I missing in [plan]\"\n- \"future-proof this [plan\u002Flaunch]\"\n\nExample prompts:\n\n- \"premortem this: launching a $297 live workshop on Claude Cowork for\n  marketing teams\"\n- \"stress-test this hire — first marketing person, junior, $60k base\"\n- \"find holes — shipping the new pricing in 2 weeks\"\n\n---\n\n## Method origins\n\n- Gary Klein, *Performing a Project Premortem*, HBR, 2007\n- Klein, *Sources of Power*, MIT Press, 1998 — mental simulation, RPD\n- Klein, *The Pre-Mortem Method*, Psychology Today, 2021 — action round\n- Klein, *Double-Barreled Pre-Mortems*, Psychology Today, 2025 —\n  reverse premortem\n- Mitchell, Russo & Pennington, *Back to the Future: Temporal\n  Perspective in the Explanation of Events*, JBDM, 1989 — prospective\n  hindsight\n- Veinott, Klein & Wiggins, *Evaluating the Effectiveness of the\n  PreMortem Technique*, ISCRAM, 2010 — empirical validation\n- Kahneman, *Thinking, Fast and Slow*, 2011, ch. 23–24\n- Kahneman & Lovallo, *Delusions of Success*, HBR, 2003 — outside view\n- Kahneman, Lovallo & Sibony, *Before You Make That Big Decision*, HBR,\n  2011 — 12-question bias checklist (6 items adopted in synthesis)\n\n---\n\n## License\n\nMIT — see [LICENSE](LICENSE).\n","该项目是一个名为premortem的Claude Code技能，旨在通过结构化方法在计划、启动、招聘或决策实施前发现潜在失败点。其核心功能包括多代理静默扫描、风险缓解三元组、历史快照和反向预演，能够在10-15分钟内识别出具体的风险因素。该工具基于Python开发，采用了Gary Klein 2007年提出的预演技术和Kahneman的外部视角理论来提高预测准确性。适用于任何需要进行风险评估与管理的场景，特别是新产品发布、关键决策制定等高风险活动之前，帮助团队更全面地审视可能遇到的问题，并提前规划应对策略。",2,"2026-06-11 04:03:19","CREATED_QUERY"]