public-claude-skills

Use when the user types /reflect or asks to "think this through", "weigh the pros and cons", "consider all angles", "challenge my thinking", "stress-test this decision", "second opinion on", "hva er sterkeste argumentet mot dette", "reflekter over". Runs a structured multi-method reflection (first principles + steelmanning + pre-mortem + inversion + stakeholder views + second-order thinking) and surfaces the tensions before landing on a position.

Use when the user types /bokmaal-proof, /bokmål-proof, or asks to "proof this Norwegian text", "rydd opp i denne teksten", "fix this bokmål", "check the Norwegian", "remove anglicisms", or wants stylistic and register-level review of Norwegian bokmål technical or business writing. Does more than grammar: catches anglicisms, register drift, Språkrådet recommendations, and the typical translation-from-English smell.

Use when the user types /design-ai-experiment or asks to "design an experiment", "set up an A/B test for an AI feature", "evaluate a model rollout", "pre-register a study", "validate an LLM change", "prove this works", or wants to turn a vague AI idea into a falsifiable test with a hypothesis, control, success threshold, and kill criterion. Borrows pre-registration discipline from academia.

Use when the user types /ml-or-not or asks "should we use ML for this", "do we need a model here", "is this an AI problem", "would ML help", or is scoping a feature where machine learning is being considered. Returns a structured verdict (yes / no / not yet) with reasoning and the cheapest viable alternative if ML is wrong.

Use when the user types /zk-llm-design or asks how to add zero-knowledge proofs to an LLM system, "prove model output without revealing weights", "ZK-ML", "verifiable inference", "trustless model serving", or wants help choosing between Halo2, Plonk, STARKs, or FHE for an LLM use case. Walks through threat model, what is being proved, scheme selection, and feasibility for a given model size.