What problem does it solve? Agent-generated artifacts often reference context the reader never sees, such as "as discussed", "the old handler", or "we decided", leaving documents confusing when read outside the original session. ## Core Features & Use Cases - Reference Detection: Identifies conversational references, process narration, diff talk, ghost alternatives, and unanchored temporal words that fail the standalone test. - Three-Way Fixing: Classifies each failed reference as decoration (delete it), essential (rewrite to state what the thing does), or unrecoverable (preserve and report what is missing). - Scope Awareness: Applies to prose files and code comments while deliberately excluding commit messages, PR descriptions, changelogs, ADRs, and migration guides where change references belong. - Use Case: After an AI assistant drafts a README during a long session, run this skill to strip phrases like "as we agreed earlier" so the document reads correctly for anyone who never saw the conversation. ## Quick Start Review this document and remove every reference that a reader holding only this file could not resolve.