reason from

Skills

AI made producing work nearly free. Checking it, understanding it, and owning it still cost what they always did. These 24 skills each protect one of those scarce things, with a mechanism you can’t satisfy by nodding along.

Open standard Agent Skills · works in Claude Code, Copilot, Cursor, Gemini CLI and more · or paste one into any chat.

Start here

Not sure what is going wrong? Bring one real incident.

  1. symptom-triageSymptom TriagePeople can feel that AI-assisted work is going wrong but misname the failure, so they reach for the wrong fix.

Truth

Is it actually so?

  1. adversarial-collaboratorAdversarial CollaboratorPositions survive because the arguer stops, not because they're sound; 'good point' replaces an actual update.
  2. assumption-surfacerAssumption SurfacerWork proceeds on premises nobody stated, so nobody checked them.
  3. bayesian-belief-trackerBayesian Belief TrackerEvidence is weighed asymmetrically: confirming evidence accepted, disconfirming explained away.
  4. commit-then-consultCommit, Then ConsultExperts overturn their own correct judgments to follow wrong AI advice.
  5. mirror-testMirror TestThe AI agrees with whatever framing you bring, and long chats spiral into validation.
  6. perception-gap-auditorPerception Gap AuditorPeople feel faster with AI while measurements say otherwise.

Ownership

Do I understand what I have?

  1. bug-hypothesis-generatorBug Hypothesis GeneratorBugs get patched one at a time while the process gap that produced them stays open.
  2. bus-factor-zeroBus Factor ZeroThe codebase works, but for whole regions of it nobody can say what it does or why.
  3. editor-not-writerEditor, Not WriterWriting ships with claims the author can't support, and AI rewriting erases the author's own voice.
  4. recontextualizerRecontextualizerDevelopers ship AI-generated changes they directed but cannot explain, especially the branches they never walked.

Intent

Did I know what good was before I saw anything?

  1. decision-pressure-testerDecision Pressure-TesterDecisions end at 'here are the risks' with no commitment about what to do about them.
  2. define-good-firstDefine Good FirstOnce you see a plausible output, your standard bends toward it, so almost-right ships.
  3. first-principles-decomposerFirst Principles DecomposerElegant reasoning built on constraints nobody verified.
  4. pre-mortem-oraclePre-mortem OraclePlans proceed because no one is allowed to say 'do not proceed', and warning signs get rationalized away.

Attention

Can I really review all of this?

  1. attention-treasurerAttention TreasurerAgents generate faster than the human can review; output piles up, errors slip through, and the human feels fried.
  2. workslop-interceptorWorkslop InterceptorPolished AI-generated work that looks done but shifts the real work — decoding, checking, redoing — onto the person who receives it.

Authority

What is allowed to be irreversible?

  1. delegation-contractDelegation ContractAn agent did something irreversible that its instructions forbade, because nothing actually prevented it.

Capability

Can I still do it when the AI can’t?

  1. capability-reserveCapability ReserveSkills you delegate quietly decay, and you find out on the day the AI is wrong
  2. learning-partnerLearning Partner'I get it' substitutes for demonstrated understanding; feeling smarter replaces being smarter.

Originality

Is it different from the default answer?

  1. concept-synthesizerConcept SynthesizerIdeas that feel connected get mistaken for insight without a testable prediction.
  2. modal-escapeModal EscapeAI-assisted ideas are better one by one and more alike across everyone

Self

Who am I with, and what am I for?

  1. human-bridgeHuman BridgeAI becomes the destination for connection instead of a bridge back to people
  2. value-relocatorValue RelocatorCraft identity disruption — 'if the AI can do my work, what am I?'