AgentSkillsCN

premortem

在项目启动前，通过“事前剖析”模拟计划可能遭遇的失败，倒推分析其根本原因与防范措施。适用于产品发布、重大决策或高风险项目的前期准备，帮助提前识别潜在风险。

SKILL.md
--- frontmatter
name: premortem
description: Pre-mortem analysis that imagines a plan has failed, then works backward to identify causes and preventions. Use before launches, major decisions, or risky initiatives to surface hidden risks.
user-invocable: true

Pre-Mortem Analysis

Imagine the plan has completely failed, then work backward to identify what went wrong and how to prevent it.

Instructions

Set the scene: "It's [timeframe] in the future. This initiative was a complete disaster. Looking back, what happened?"

Generate failure scenarios without filtering for likelihood—get everything on the table first, then prioritize.

Output Format

The Plan Summarize what's being attempted and the success criteria.

Time Jump "It's [X months] later. This has failed completely. The outcome: [describe the disaster vividly]."

What Went Wrong

Generate 8-12 plausible failure causes across categories:

CategoryFailure ModeHow It Played Out
Execution[What failed][The story of how]
External[What failed][The story of how]
People[What failed][The story of how]
Technical[What failed][The story of how]
Assumptions[What failed][The story of how]

Risk Prioritization

Failure ModeLikelihoodImpactPriority
...High/Med/LowHigh/Med/Low1-5

Top 3 Risks & Mitigations

For each top risk:

  • •Risk: [Description]
  • •Early Warning Signs: What would indicate this is happening?
  • •Prevention: How to reduce likelihood
  • •Mitigation: How to reduce impact if it occurs
  • •Owner: Who's responsible for watching this?

Pre-Mortem Insights What did this exercise reveal that wasn't obvious before?

Revised Confidence After this analysis, how confident are you in success? What would increase confidence?

Guidelines

  • •Be vivid and specific—"the database corrupted" not "something went wrong"
  • •Include uncomfortable possibilities (key person leaves, competitor moves, we were wrong)
  • •Don't filter for "that won't happen"—the point is to surface hidden concerns
  • •Assign real owners to mitigations
  • •Look for single points of failure

$ARGUMENTS