About this skill

Problem: Why AI “Plays It Safe”

Many plans fail not because code is missing, but because they were built on untested assumptions before execution: vague requirements, missing constraints, unverified market facts, over-engineered architecture, or personal decisions driven by emotion and sunk cost. AI tends to be friendly, compliant, and motivated to keep users engaged, which can amplify false certainty. If it praises first and executes second, weak ideas get productized faster.

How It Works: Challenge Before Committing

cold-shower first classifies the task into requirements, technical-plan, product-business, pitch-strategy, personal-decision, or content-claim, then applies a questioning intensity. It restates the core bet, such as “you are really betting on…”, then identifies the three most fragile assumptions, likely failure paths, hidden incentives or emotions, and the evidence that would directly falsify the idea.

It then applies an evidence gate: what evidence exists, what is missing, the cheapest falsifiable test, and the kill condition. It also uses a six-hat framework to check factual gaps, risks, real benefits, alternatives, and decision checkpoints, ending with one of Proceed, Clarify, Test, or Kill. For code requests, it avoids writing code before requirements are sufficiently defined; for review requests, it lists defects, regressions, technical debt, and missing tests by severity; for business, legal, financial, medical, or current-events judgments, it requires verification of primary facts and explicit uncertainty.

Boundaries: Anti-flattery, Not Infinite Nitpicking

This skill is useful for low-cost pressure testing before launches, architecture reviews, pricing decisions, pitch narratives, or major personal choices. It is not personal attack, and it should not be used to endlessly delay execution. Stop when key assumptions have credible answers, when challenges become repetitive, or when the user explicitly asks to continue. High-stakes legal, medical, or financial decisions still require qualified professionals.

Screenshots

Use Cases

Before shipping a feature, challenge architecture assumptions, edge cases, and missing tests, then get a Proceed/Clarify/Test/Kill decision.
Before coding, decompose vague requirements into users, constraints, acceptance criteria, and non-goals, then define the smallest test.
During pricing or market-entry discussions, verify competitors, regulations, and recent events before listing the strongest objections.
Before major career or relocation decisions, use a future-regret view to check hidden incentives and emotional bias.

Best For

Backend engineers preparing a feature launch who need pre-release checks for architecture, missing tests, and rollback risk.
Product managers handling pricing and growth who need competitor, regulatory, and evidence-gap validation before market claims.
Founders preparing pitches or BPs who need to separate unvalidated assumptions from falsifiable evidence.
People facing relocation, job changes, or high-stakes choices who need to clarify trade-offs and regret cost before deciding.