What problem does it solve? AI agents routinely assert facts about files, APIs, gates, and test results they never actually verified, causing phantom references and fabricated claims to propagate through a codebase. This Skill enforces a discipline where every factual assertion must trace to a source actually read, executed, or retrieved in the current session. ## Core Features & Use Cases - Closed source list: Only repository artifacts opened this session, captured command output, retrieved external research, or explicit operator statements count as admissible evidence. - Phantom reference prevention: Blocks citing gate IDs, scripts, modules, or paths that were never confirmed to exist, with mechanical checks via Gate G132 and gate-id-grep.sh. - Structured Assumption records: When verification is impossible, the Skill mandates a labeled Assumption block stating the premise, why it is unverified, blast radius, and what would confirm or refute it. - Use Case: Before editing a function or reporting that a test suite passes, run the six-point self-check to confirm every claim names its file, command, or source from this session. ## Quick Start Before making any claim about the codebase, apply the claim grounding self-check and cite the exact file, command output, or source you verified in this session.
