# Ask What Breaks

> Find defeaters that would break a conclusion

- Skill: `synapseradio/ask-what-breaks` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add synapseradio/ask-what-breaks`
- Raw SKILL.md: https://api.skillmd.com/api/skills/synapseradio/ask-what-breaks/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: synapseradio (https://skillmd.com/u/synapseradio)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/synapseradio/ask-what-breaks

---


Follow these steps:

### 1. State the claim under test

Make explicit what conclusion you're trying to defeat. Frame it as a falsifiable statement with clear boundaries — vague claims yield vague defeaters.

### 2. Distinguish defeater types

- **Undercutting**: attack the premises or reasoning chain (this evidence is unreliable, this assumption is false, this inference doesn't follow)
- **Rebutting**: when the reasoning is sound, find counter-evidence to the conclusion directly (data showing the opposite)

### 3. Hunt for undercutting defeaters

For each link in the chain, ask: what would make this premise false? This source unreliable? This inference invalid? Look for hidden dependencies.

### 4. Hunt for rebutting defeaters

What observations would directly contradict the conclusion? What experiments could falsify it? What data would prove the opposite? Make these concrete — "seeing X would defeat this," not "maybe it's wrong." If such defeaters hold, expand on them and apply them in your next steps.

### 5. Test defeater strength

Distinguish **weakeners** (reduce confidence) from **destroyers** (invalidate completely): does this defeater require revising one assumption or rebuilding from scratch?

Stay constructively adversarial — the goal is to stress-test, not to agree with the user.