What problem does it solve? When a user doubts the accuracy of an AI assistant's recent output, the assistant tends to defend, rephrase, or rationalize instead of checking facts. This Skill forces a structured verification workflow so claims are checked against real sources rather than the same model weights that produced them. ## Core Features & Use Cases - Claim identification and prioritization: Re-reads recent output, extracts verifiable claims, and verifies at most 3 per invocation, starting with the claim closest to the user's doubt. - Two-stage source search: Searches local sources first (codebase, configuration, tests, docs, runtime commands), then external sources (official documentation, Stack Overflow, GitHub issues, blog posts) when local evidence is insufficient. - Honest synthesis: Reports results as wrong, nuanced, or confirmed with sources, without hedging or rewriting history. - Proactive trigger: Fires before asserting that a design is technically necessary, grounding the justification before complexity gets locked in. - Use Case: A user replies "dat klopt niet" to a claim about when Hash#dig was introduced in Ruby. The Skill identifies the claim, checks the local Gemfile.lock and the official Ruby changelog, and concedes with the correct version and source link. ## Quick Start Type /ground or tell the assistant that its last answer seems wrong, and it will verify the disputed claims against local and external sources.
