fact-checker_skill

Overview

This skill verifies factual claims in documents using web search and authoritative sources, then proposes corrections for user approval. It is built for technical docs, AI model specs, statistics, and general factual statements, focusing on accuracy and traceable citations.

How this skill works

I scan the document to identify verifiable statements (specs, versions, dates, metrics). For each claim I search official sources and cross-check multiple authoritative references. I generate a structured correction report with status codes (accurate, incorrect, outdated, unverifiable) and proposed edits, then apply changes only after explicit user confirmation.

When to use it

• When you need to fact-check a document or section for accuracy
• When verifying AI model specifications, context windows, or release notes
• When updating outdated statistics, version numbers, or API capabilities
• When validating benchmark scores or performance claims
• Before publishing technical documentation that cites external specs

Best practices

• Focus checks on verifiable claims: specs, dates, numbers, versions
• Use official product pages, API docs, release notes, and registries as primary sources
• Include temporal context in corrections (e.g., "as of [date]")
• Cross-reference at least two authoritative sources when possible
• Present corrections and sources to the user and wait for explicit approval before editing

Example use cases

• Fact-check model context windows and update references to current model names
• Verify benchmark numbers in a research chapter and replace incorrect figures with sourced values
• Confirm library version numbers in a dependency list and recommend upgrades with changelog links
• Audit API capability claims in a developer guide and annotate unverifiable statements
• Prepare a correction report before committing changes to documentation