# Source Checker — verify a claim and rate the evidence

> Verifies a claim by finding primary sources and rates the evidence as supported, mixed, unsupported, or unverifiable, with citations and what would change the verdict.

- Skill: `drnabeelkhan/source-checker-verify-a-claim-and-rate-the-evidence` (Agent Skill)
- Install (CLI): `npx skillmds add drnabeelkhan/source-checker-verify-a-claim-and-rate-the-evidence`
- Raw SKILL.md: https://api.skillmd.com/api/skills/drnabeelkhan/source-checker-verify-a-claim-and-rate-the-evidence/raw
- Safety review: PASS (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search, Summarization, Web Research
- Tags: Bayesian Weighting, Citations, Craap Test, Evidence Rating, Fact Checking, Primary Sources
- Author: DrNabeelKhan (https://skillmd.com/u/drnabeelkhan)
- Updated: 2026-08-22
- Page: https://skillmd.com/skills/drnabeelkhan/source-checker-verify-a-claim-and-rate-the-evidence

---


# Source Checker

> **Does:** verifies a claim and tells you how solid the evidence really is.
> **Solves:** repeating something confident that turned out wrong.
> **Triggers on:** *"check this claim"* + the statement

## How Maxim does this (researcher-writer separation, structurally)

Isolates the exact claim, finds **primary sources** (web/research), and rates the evidence — **supported / mixed / unsupported / unverifiable** — citing what it found. It separates *gathering* from *asserting* (the researcher-writer split): it reports what the sources say, not what it "believes."

## Behavioral overlay

- **Frameworks (cited per ADR-007):** **CRAAP test** (Currency, Relevance, Authority, Accuracy, Purpose) · **Bayesian evidence weighting** (how much this should move your prior) · the Maxim **no-fabrication** rule.
- **Confidence tag (ADR-010)** is the whole point: the verdict *is* a confidence rating, and it names **what would change it**.
- **CSO-adjacent:** for medical/legal/financial claims, `security-analyst` is looped — it flags regulated claims rather than ruling on them.

## Output

Verdict (supported / mixed / unsupported / unverifiable) · evidence quality per source · the citations · what would change the verdict.