fact-check

Verify every claim in a document against live sources and re-derived database queries.

Updated Aug 17, 2026
One-click install
npx skills add https://github.com/AMC-JTC/gbrain-1 --skill fact-check-amc-jtc
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: fact-check
Source: https://github.com/AMC-JTC/gbrain-1/tree/main/plugin/skills/fact-check
Command: npx skills add https://github.com/AMC-JTC/gbrain-1 --skill fact-check-amc-jtc

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Drafts, reports, and AI-generated outputs often contain factual errors, hallucinated statistics, and misattributed relationships that damage credibility when published. This Skill runs a systematic claim-by-claim verification pass modeled on professional fact-checking desks, so nothing ships with unsupported or false claims. ## Core Features & Use Cases - Claim extraction and triage: Pulls every verifiable claim from a document, prioritizes high-risk items (numbers, quotes, superlatives, defamatory statements), and flags red-flag patterns like round numbers and unattributed claims. - Dual verification lanes: Web-derived claims are checked against a live source hierarchy (primary sources, credible journalism, Wikipedia), while data-derived claims from a brain or database are re-derived through an independent query path (PRODUCER ≠ VERIFIER) with typed-edge checks for person-to-thing relationships (AFFILIATION ≠ AUTHORSHIP). - Scored gate report: Assigns each claim a 6-level confidence status, applies corrections and hedging directly to the document, and produces a pass/fail report that hard-blocks delivery when unsupported data-derived claims remain. - Use Case: Before publishing a blog post built from brain queries, run the Skill to re-derive every statistic and relationship claim, fix two wrong funding figures, hedge one unverifiable date, and receive a HIGH-confidence report clearing the post for publication. ## Quick Start Ask the AI to fact check this draft claim by claim against live sources and re-derive any data-derived claims before it ships.

Frequently Asked Questions about fact-check

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

FAQPage Schema
How do I fact check a document claim by claim with AI?

Extract every verifiable claim into a numbered list, then check each against live citable sources rather than AI training data. Assign each claim a confidence status from verified to false, apply corrections, and produce a scored pass/fail report before publishing.

How to verify claims generated from a database or knowledge base?

Re-derive each claim through a different query path than the one that produced it, since re-running the original query reproduces the original bug. Person-to-thing claims must resolve through typed edges like founded or invested_in, not mention co-occurrence.

What sources are reliable for fact-checking statistics and quotes?

Primary sources rank highest: SEC filings, official press releases, government databases, and transcripts. Credible journalism from outlets like Reuters or Bloomberg and Wikipedia work for cross-referencing, while social media posts and AI training data are never sufficient alone.

Can web search verify claims about my own private data?

No. Web search cannot verify what a pipeline asserted about your own brain or database, because the web is the wrong authoritative source. Data-derived claims must be re-derived against the source database itself through an independent query.

When should I not run a full fact-check pass?

Skip verification for opinions, arguments, predictions, and metaphors, since these are not falsifiable claims. For a single load-bearing research claim needing deep methodological tracing, a dedicated single-claim depth review is a better fit than a breadth pass.