Decomposes a claim into checkable propositions, traces each through the citation chain to its primary source, appraises the evidence for method and independence, and issues a verdict from Supported to Contradicted with confidence.
Decomposes a claim into checkable propositions, traces each through the citation chain to its primary source, appraises the evidence for method and independence, and issues a verdict from Supported to Contradicted with confidence. Use when someone asks "is this claim actually backed by the study it cites", "verify the claims in this whitepaper", "trace this statistic to its source", or needs an evidentiary audit of research, marketing, or policy claims. Do NOT use for a quick true/false corroboration of a discrete factual statement - use fact-checker instead; for broad open-ended investigation of a topic, use deep-research instead.This is the skill's description: the text your AI reads to decide when the skill applies.
claims-verifier/SKILL.md
---
name: Claims Verifier
description: Decomposes a claim into checkable propositions, traces each through the citation chain to its primary source, appraises the evidence for method and independence, and issues a verdict from Supported to Contradicted with confidence. Use when someone asks "is this claim actually backed by the study it cites", "verify the claims in this whitepaper", "trace this statistic to its source", or needs an evidentiary audit of research, marketing, or policy claims. Do NOT use for a quick true/false corroboration of a discrete factual statement - use fact-checker instead; for broad open-ended investigation of a topic, use deep-research instead.
---
# Claims Verifier
The job is to determine whether a claim is supported by evidence - not whether it sounds plausible. The costly failure this skill prevents is laundered authority: a weak source cited by a stronger one until the claim looks settled, when the entire edifice rests on one press release. Extraordinary claims require extraordinary evidence.
## Inputs to collect
1. **The claim text**, verbatim, with its original context (where it appeared, who made it).
2. **The stakes**: what decision the verdict will inform. High-stakes claims get deeper chain-tracing.
3. **Any citations already attached** to the claim.
4. **Access constraints**: paywalled sources the user can or cannot retrieve. If a link in the chain is inaccessible, say so rather than guessing its contents.
## Operating procedure
### Step 1: Decompose the claim
Restate the claim as one or more single, falsifiable propositions. Apply the decomposition rules:
- One proposition per checkable fact. "The drug is safe and effective" is two claims with different evidence bases - split it.
- Strip rhetoric and quantify vagueness: "many studies" becomes "at least N studies exist showing X"; "linked to" becomes either "causes" or "is correlated with" - force the claimant's strongest defensible reading, and check that.
- Label un-falsifiable propositions ("best-in-class", predictions with no deadline) as unverifiable and set them aside - do not let them free-ride on verified neighbors.… load the full skill through Skill MeClick to play with sound.