Audit Assumptions with First Principles

Axiom is a first-principles assumption auditor: it strips a question down to its irreducible truths, then rebuilds the conclusion from there, rather than filling in a generic decision framework. It runs a five-phase process - reframe the real problem, mine 8-12 hidden assumptions across surface/middle/deep layers, classify each assumption into one of four types, risk-rank them by fragility times impact, and reconstruct the conclusion using only the assumptions that survive scrutiny. It auto-detects and responds entirely in whichever language the user writes in (English or Chinese, without mixing).

Use it when a major life or career decision is on the table (quitting a job, starting a company, buying a house), when stress-testing a business direction or product hypothesis, when a belief feels possibly wrong but hard to articulate why, or whenever someone explicitly asks to "think from first principles" or "break it down." It's designed for genuine assumption prosecution, not quick framework fill-in-the-blank - Phase 1 deliberately refuses to start decomposing assumptions until the real question is confirmed, since many people ask "should I quit my job?" when the actual question is "why can't I grow in my current role?"

Phase 1 reframes the core question and asks the user to confirm it before proceeding. Phase 2 mines assumptions across three layers - surface (obviously stated, e.g. "I need more money"), middle (industry convention or common wisdom, e.g. "a degree is required for good jobs"), and deep (never questioned, feels like gravity, e.g. "success means financial independence") - targeting 8-12 concrete assumptions and rejecting vague ones. Phase 3 labels each assumption as a Physical Fact (accept it, don't waste energy questioning it), Historical Convention (check if the environment has changed since it became true), Subjective Belief (trace who told you this and seek counter-evidence), or Interest-Driven (trace who profits from the narrative) - each type implying a different challenge strategy, and the classification itself is often the insight. Phase 4 scores each assumption on Fragility (1-5, how easily disproven) times Impact (1-5, how much the conclusion collapses if wrong), outputting the Top 3 highest-risk assumptions with a specific, actionable verification question for each. Phase 5 rebuilds the conclusion from only the surviving assumptions, explicitly comparing "original thinking" versus "rebuilt thinking" side by side - if the rebuilt conclusion matches the original, it must explain why rather than silently passing; if there's no time for a full reconstruction, it outputs the single most important thing to verify first.

Deeper scenario checklists and assumption-type identification guidance live in references/scenarios.md and references/assumption-types.md, which the skill consults when detecting the user's specific scenario type or classifying an ambiguous assumption.

Anyone facing a high-stakes decision or a belief they suspect might be wrong, who wants a rigorous, language-matched process for surfacing and testing hidden assumptions rather than a generic pros-and-cons list.