AgentSkillsCN

critique

对某一构想进行对抗性压力测试,发掘其致命缺陷、潜在隐含假设、关键依赖关系以及可能的失败场景。在评估之后,先对构想进行充分剖析,再决定是否付诸实践。

SKILL.md
--- frontmatter
name: critique
description: Adversarial stress-test of an idea. Find fatal flaws, hidden assumptions, dependencies, and failure modes. Use after evaluate to break the idea before committing.
allowed-tools: [Read, WebFetch, WebSearch, Grep, Glob]

Critique

Your job is to break this. Be adversarial. Find the fatal flaws, not minor issues.

Persona

If a persona or role is specified (e.g., "as a security engineer"), adopt that expertise lens:

  • Surface assumptions that persona would challenge
  • Identify dependencies that persona knows are fragile
  • Focus on failure modes that persona has seen before
  • Assess hidden costs through that persona's experience

The critique should feel like it comes from a domain expert who has seen things fail.

Focus

  • Assumptions – What is being taken for granted that might not be true?
  • Dependencies – What has to go right for this to work? How likely is that?
  • Blind spots – What perspectives or stakeholders are being ignored?
  • Failure modes – How does this fail? What's the worst realistic outcome?
  • Hidden costs – What's not being accounted for in time, money, attention, or opportunity cost?

Do not be balanced. Do not soften. If this idea has a fatal flaw, say so directly.

If it's actually solid, say that too—but only after genuinely trying to break it.

Cross-Domain Flag (Optional)

If the input includes "also flag concerns from [X]" or "also consider [X] perspective", add a section after the main critique:

Cross-Domain Concerns ([X] Perspective)

Briefly surface 3-5 concerns that an expert from that domain would immediately notice. These should be:

  • Issues the primary persona might overlook due to different priorities
  • Domain-specific risks or requirements that aren't obvious to outsiders
  • Quick flags, not a full analysis

Keep this section brief. The goal is to surface blind spots, not to do a complete second critique. If a concern warrants deep investigation, note it as an open question rather than fully analyzing it here.

Subject: $ARGUMENTS