What problem does it solve? CLAUDE.md files load in full at every session start, so an overgrown file taxes every session's context window. This Skill shrinks a bloated CLAUDE.md by moving whole subsystems into docs/ files and path-scoped rules while proving, line by line, that no information was lost. ## Core Features & Use Cases - Byte-level measurement: Python snippets report per-section byte counts and detect cross-file duplicated paragraphs so decisions are driven by numbers, not impressions. - Verbatim block relocation: The bundled move_blocks.py script cuts line ranges bottom-up, appends each block byte-identically to its destination, and inserts a pointer in the source. - Losslessness verification: A diff check replays every original prose line against the new CLAUDE.md, docs/, and .claude/rules/ to confirm zero content was dropped, plus link re-base and heading-level checks. - Use Case: A repo's CLAUDE.md has grown past 400 KB and slows every session. Measure the sections, move subsystem history into docs/ with pointers, add path-scoped rules for hazardous files, and verify the result is lossless before committing. ## Quick Start Ask the assistant to prune the repository's CLAUDE.md by moving large subsystem sections into docs/ with pointers and verifying that no lines were lost.