trim

What problem does it solve? Documentation and code files accumulate bloat over time — restated facts, embedded changelogs, dead references, and comments that earn nothing — making them hard to maintain and easy to misread. This Skill distills a named file to its smallest, highest-signal form while guaranteeing zero information loss and identical observable behavior. ## Core Features & Use Cases - Eleven bloat patterns with detection commands: Detects wall-of-text paragraphs, restated facts, decision archaeology, embedded changelogs, dead §refs, redundant CSS inheritance, Figma node ids, and finished todos using concrete grep/awk commands. - Zero-information-loss guarantee: Lists load-bearing facts before editing and verifies each one survives after rewriting, with mandatory link/anchor resolution checks. - Multi-file group trims: Builds a trim map across multiple targets so each fact gets exactly one home with one-line pointers elsewhere. - Use Case: A 400-line conventions doc has grown contradictory and repetitive. Run the trim command on it to get a proposed restructure, then a verified rewrite that cuts 40% of the lines while keeping every contract, guard, and file:line reference intact. ## Quick Start Ask the agent to trim references/development/conventions.md to its smallest form without losing any actionable facts.