skills.rest
peachest
peachestCOMMUNITY
Explore all skills by this author →
GitHub

fact-check

Extract verifiable claims from documents and verify them via rule engines and web search.

Updated Jul 20, 2026
Search Tags:#verification#web search#arxiv#fact-check#claim extraction#document validation#evidence grading
One-click install
npx skills add https://github.com/peachest/skills --skill fact-check-peachest
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: fact-check
Source: https://github.com/peachest/skills/tree/main/research/fact-check
Command: npx skills add https://github.com/peachest/skills --skill fact-check-peachest

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Technical reports and research documents often contain unverified claims—wrong arXiv IDs, outdated benchmark numbers, nonexistent repositories. Manually checking each claim against external sources is slow and error-prone. This Skill automates the full verification pipeline, producing evidence-backed verdicts for every claim. ## Core Features & Use Cases - Claim Extraction & Decomposition: Extracts verbatim claims from Markdown documents, classifies them into typed categories (authority, numerical, temporal, etc.), and splits compound claims using a fixed 7-pattern catalog. - Deterministic Rule Engine: Verifies authority claims (arXiv IDs, DOIs, GitHub PRs, npm/PyPI packages, URLs, RFC numbers) via 20+ HTTP verifiers before any LLM involvement. - Deep Web Verification: Routes remaining claims through triage and multi-round web search with evidence tier grading (T1–T4), producing verdicts like SUPPORTED, CONTRADICTED, NUANCED, or UNVERIFIABLE. - Incremental Re-checks: Maintains per-document ledgers so re-runs only verify changed claims, plus cumulative stats across runs. - Use Case: After drafting a technical survey citing 40+ papers and benchmark figures, run the pipeline to confirm every arXiv ID resolves, every GitHub repo exists, and every performance number has independent evidence—then get a report with suggested fixes. ## Quick Start Ask the agent to fact-check your report file, for example: run /fact-check on docs/llm-survey.md and show me the verdict summary.

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 technical document with an AI agent?▼

Run the fact-check pipeline on your Markdown file; it extracts verbatim claims, classifies them by type, and verifies each against rule engines or web search. The output is a report with per-claim verdicts, evidence tiers, and suggested fixes.

What types of claims can be verified automatically?▼

Authority claims like arXiv IDs, DOIs, GitHub PRs, npm/PyPI packages, URLs, and RFC numbers are verified deterministically via HTTP checks. Numerical, temporal, and comparative claims go through web search with evidence grading, while pure opinions are marked REFUSED.

Does fact-check support incremental re-verification of edited documents?▼

Yes, it stores per-document ledgers keyed by claim content hash. On re-runs, git diff hunk analysis identifies changed regions so only modified claims are re-extracted and re-verified, while unchanged claims carry forward.

How are compound claims with multiple facts handled?▼

Compound claims are decomposed into atomic sub-claims using a fixed catalog of 7 patterns (and-enumeration, parenthetical, from-to, clause-embed, and others). Claims matching no pattern are flagged as compound_embedded rather than force-split.

Why do some claims come back as UNVERIFIABLE?▼

A claim is UNVERIFIABLE when web search finds no evidence or only low-tier sources, or when a rule-engine check fails after retries. The verdict reflects insufficient current evidence, not that the claim is false.

Related Skills

Explore More Education & Research Skills→
10553731651055373165

translate-article

Translates English technical articles to Chinese Markdown using Playwright, Readability, and custom algorithms.

Community
Advanced
0xtaoist0xtaoist

ln-614-docs-fact-checker

Extract verifiable claims from Markdown docs and validate them against the codebase.

Community
Advanced
Azeem-2Azeem-2
1

content-editing

Run a six-pass editorial workflow covering grammar, citations, consistency, formatting, readability, and factual accuracy.

Community
Advanced
motoki317motoki317
4

Fact Check

Verify user claims against Context7 and web sources, producing discrepancy reports.

Community
Advanced