← ClaudeAtlas

assumption-checklisted

Invoke this skill BEFORE making any load-bearing assertion, recommendation, or decision that rests on unverified external information — whether how a tool/API/framework/platform behaves, specific facts, numbers, or dates you are about to cite, current versions or pricing, or subject-matter details that may be wrong or out of date. Do not wait until something breaks — trigger when the task involves: expert recommendations in a specialized domain, citing figures or benchmarks, designing permission or auth systems, planning API integrations, evaluating tools for adoption, or building any system where a config, SDK behavior, or factual claim is load-bearing. Also trigger when: making a placement or scope decision, choosing a format or convention, deciding build-vs-buy, or any decision where 'I assume X works like Y' or 'I recall X is Y' is the justification. Widely used whenever a decision rests on unverified external state or assertions - behavior, facts, or expertise.
Nova-Caelum/no-mistakes · ★ 0 · API & Backend · score 65
Install: claude install-skill Nova-Caelum/no-mistakes
# Assumption Check — Validate Before You Build The failure pattern this skill prevents: design a system around assumed behavior → discover the assumption was wrong → rework the architecture. The cost is hours. The prevention is minutes. An assumption is anything you haven't confirmed via documentation or direct empirical test in this specific context. Reasoning ("it probably works this way") is hypothesis formation, not verification. Test environment conditions — session modes, feature flags, environment variables — require the same verification discipline as tool behavior itself. --- ## Proactive mode (start of design session) Use this when running the check before design work begins. This is the preferred path. ### 1. Surface assumptions explicitly List every assumption being made about external behavior. Be specific: > "I assume `allow` in settings.json acts as a whitelist" — unverified > "I assume this API paginates by default" — unverified > "I assume the test session has no special modes active" — unverified If you're struggling to find assumptions, look for: default behaviors, edge cases in configuration, implicit limits, behaviors mentioned in tutorials but absent from reference docs, and conditions about the environment itself (not just the tool). ### 2. Classify by confidence and load For each assumption: - **Confidence**: high / medium / low - **Load-bearing**: does the design break or produce incorrect results if this assumption is wrong? Focus verific