The Agent Skills Directory

[SAFE]: The skill is a well-documented process for factual verification. It defines clear phases for claim extraction, evidence gathering, adjudication, and reporting.
[SAFE]: The referenced files in references/ contain methodology guidelines (e.g., lateral reading, source-tier climbing) and configuration for data freshness windows. These are static markdown files providing instructions rather than executable code.
[SAFE]: The evals/ directory contains a comprehensive fixture corpus (articles and source documents) used for testing the fact-checker's performance against 'seeded errors' like misquotes and stale statistics. These fixtures are fictional data sets for benchmarking.
[SAFE]: While the skill's frontmatter allows tools like Bash, WebFetch, and WebSearch, these are standard for research-oriented skills. The instructions guide the agent to use them for legitimate verification tasks (e.g., checking primary sources or news archives).
[SAFE]: The instruction set includes specific 'Error Handling' and 'Burden of Proof' rules that encourage transparency (e.g., labeling claims as 'Missing-source' or 'Unverifiable' rather than guessing).

fact-check