What problem does it solve? AI assistants often guess when information is missing, producing hallucinated answers about versions, APIs, or code behavior. This Skill acts as a reliability gate that forces evidence-based reasoning and returns "insufficient information" instead of fabricated answers. ## Core Features & Use Cases - Confidence Gate: Computes a 0-100 confidence score and only delivers a final answer at 100/100, when every claim is backed by repo files, command outputs, or provided docs. - Request Classification: Sorts requests into repo-grounded, doc-grounded, or external-facts categories, with mandatory verification for changeable facts like "latest" versions or current pricing. - Structured Output: Produces a strict answer format with evidence citations and verification checks, or a "what's missing" checklist telling you exactly what to provide to reach full confidence. - Use Case: Before applying a code change in an unfamiliar repo, ask the assistant to confirm the approach only if fully verified; it will either deliver an evidence-backed patch or list the exact commands and files needed to proceed safely. ## Quick Start Ask the assistant to answer your question using the grounded skill only if it is 100% sure based on verifiable evidence.
