# Claim Verification

> >

- **Type:** Skill
- **Install:** `agentstack add skill-skyf0xx-better-thinking-claim-verification`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [skyf0xx](https://agentstack.voostack.com/s/skyf0xx)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [skyf0xx](https://github.com/skyf0xx)
- **Source:** https://github.com/skyf0xx/better-thinking/tree/master/skills/claim-verification
- **Website:** https://zenbin.org/p/better-thinking

## Install

```sh
agentstack add skill-skyf0xx-better-thinking-claim-verification
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Claim Verification

Journalism-grade fact-checking of a specific claim: decompose it, trace to origin, triangulate independently, and grade the verdict.

## Why

Claims travel farther than their evidence. A repeatable verification pipeline replaces "sounds right" with a graded verdict, and does it fast enough to run routinely rather than only on the rare claim that triggers suspicion.

## Use when / Don't use when

- **Use when:** a claim is load-bearing for a decision or about to be repeated onward; a statistic is doing heavy rhetorical work.
- **Don't use when:** nothing actually depends on the claim's truth.

## Inputs → Outputs

- **Inputs:** a specific factual claim.
- **Outputs:** a verdict — verified, partly verified, unsupported, false, or unverifiable — with the evidence trail behind it.

## Principles

- Verify the claim *as stated* — verifying a nearby, adjacent claim is how misinformation launders itself into credibility.
- The origin is the actual unit of verification, not the repetitions of it.
- Absence of confirmation is not refutation — "unverifiable" is an honest, legitimate verdict.
- Time-box the effort; verification depth should scale with the claim's load, not run to exhaustion on trivia.

## Procedure

1. Fix the claim verbatim; run `question-decomposition` if it bundles several assertions — verify each separately, since the bundle's truth is their conjunction.
2. Trace to origin: follow the citation chain to the primary source. Note precisely where the chain breaks or mutates — mutation is itself a finding.
3. Audit the origin with `source-credibility`; if the claim is quantitative, run `statistical-audit` on the original number, not the retelling.
4. Run `evidence-triangulation`: count independent confirmations by origin-cluster, not by raw repetition count.
5. Search for the rebuttal using `search-strategy`'s disconfirming queries: has this been debunked, corrected, or retracted?
6. Grade: verified, verified-with-distortion (true origin, mutated in transit), unsupported, false, or unverifiable. Tag residual uncertainty with `epistemic-tagging`.

## Common mistakes

- Verifying the paraphrase instead of the claim as actually stated.
- Stopping at a "reputable" repeater instead of tracing to the actual origin.
- Treating the origin's mere existence as proof of its correctness.
- Over-verifying trivia while a genuinely load-bearing claim skates through unchecked.

## Examples

- A statistic in a strategy memo before it goes to a board presentation.
- A historical "quote" attributed to a well-known figure.
- A rival's benchmark claim; a viral study result circulating without scrutiny.

## Related

- [[critical-reading]] — the upstream extraction step that identifies which claims are actually worth verifying.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [skyf0xx](https://github.com/skyf0xx)
- **Source:** [skyf0xx/better-thinking](https://github.com/skyf0xx/better-thinking)
- **License:** MIT
- **Homepage:** https://zenbin.org/p/better-thinking

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-skyf0xx-better-thinking-claim-verification
- Seller: https://agentstack.voostack.com/s/skyf0xx
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.