What problem does it solve? When the same skill name exists in many projects, copies drift apart: hardlinked files get rewritten into new inodes, one-time copies never sync, and priority shadowing hides stale versions. This Skill provides the audit commands and decision tree to find that drift across repos and decide what to do with each diverged copy. ## Core Features & Use Cases - Cross-repo drift detection: Group every .claude/skills/*/SKILL.md by name, inode, and content hash using find, stat, diff, and a Perl whole-tree sweep to locate identical-but-unlinked or genuinely diverged copies. - Triage decision tree: Distinguish harmless inode splits, pure staleness, upstream-worthy improvements, project-specific customizations, and shadowed sources, with the correct fix for each (relink, fold upstream, rename to a project-prefixed skill, or reorder sources). - Systemic gap repair: Register missing source directories with manage-skills sources add and verify winners with manage-skills locations so drift does not reappear. - Use Case: After noticing a project's skill copy never updates, run the inode/hash audit across ~/dev, discover a one-time copy that diverged, diff it against the master, and relink it after confirming the differences are pure staleness. ## Quick Start Ask the AI to audit all projects under ~/dev for drifted copies of a given skill name and triage any differing versions using the decision tree.
