source-verification

Verify articles, claims, and web sources against primary evidence with structured verdicts.

115|9|Updated Aug 5, 2026
One-click install
npx skills add https://github.com/AtlasOmnia/donna-starter --skill source-verification-atlasomnia
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: source-verification
Source: https://github.com/AtlasOmnia/donna-starter/tree/main/skills/research/source-verification
Command: npx skills add https://github.com/AtlasOmnia/donna-starter --skill source-verification-atlasomnia

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Online articles, AI announcements, and viral claims often mix verified facts with speculation, reposts, and outright errors. This Skill separates what is directly supported by primary sources from what is partially supported, unverified, or misleading, so you get a grounded verdict instead of a generic summary. ## Core Features & Use Cases - Claim-by-claim verification: Breaks articles into checkable claims (dates, quantities, attributions, technical details) and verifies each against primary sources like GitHub APIs, official docs, and government registries. - Source quality ranking: Distinguishes first-party statements, neutral reporting, reposts, and low-provenance blogs, and ranks evidence by strength for sensitive accusations. - Specialized verification patterns: Covers AI product disambiguation, privacy-policy and data-retention analysis, SaaS pricing checks, domain ownership reconciliation, app-market intelligence, and prompt-injection audits. - Use Case: When a viral article claims a GitHub project plagiarized another's architecture, the Skill checks repo creation dates, commit history, and issue threads via the GitHub API, then reports which claims are supported, which are reposted accusations, and what evidence would settle the dispute. ## Quick Start Verify the claims in this article and tell me which statements are supported by primary sources and which are unverified or misleading.

Frequently Asked Questions about source-verification

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I fact-check an article or online claim?

Extract the article's text and metadata, break it into checkable claims (dates, quantities, attributions, technical details), then verify each against primary sources like official docs, GitHub APIs, or government registries. Finish with a claim table marking each item supported, partially supported, unsupported, or misleading.

How to verify GitHub repository claims like stars, releases, or plagiarism accusations?

Use the GitHub API for repo creation dates, releases, tags, issues, and star counts, and fetch raw README or license files with a normal User-Agent when extraction tools stall. Treat repeated accusations across outlets as reposts, not independent corroboration, and inspect issue comments and linked evidence directly.

What should I do when web extraction tools time out on a page?

Switch methods after one timeout instead of retrying. Use browser automation for X/Twitter posts, terminal HTTP requests with a standard User-Agent for raw files, and public JSON endpoints for Reddit threads. Repeated retries on one failing tool waste time without new information.

Does 'does not train on your data' mean a provider does not retain prompts?

No. Training opt-out, content logging and retention, and Zero Data Retention are separate commitments. Verify each provider in the chain (gateway, aggregator, inference provider) by reading both the privacy policy and Terms of Service, since one company's privacy mode does not bind downstream processors.

Can a registered domain with no website still be purchased?

No. A registered domain with no DNS records is still unavailable; absence of a site is not availability. Check RDAP for registration date, registrar, and nameservers, and reconcile whether the user already owns it through another registrar account before recommending acquisition.

What are the limitations of search results for fact-checking?

Search results are discovery only, not proof. Snippets can be stale, and multiple outlets repeating one accusation is not independent corroboration. Always confirm claims against primary or near-primary sources before assigning a verdict.