manage-assets

Repos don't get slow from code. They get slow from binaries — a PDF committed last year, a 400 MB SQLite file a junior engineer checked in, a node_modules/ that snuck past .gitignore, a dist/ directory nobody bothered to exclude. A single 200 MB blob in git history turns git clone into a coffee break for every new collaborator, forever.

This skill surfaces that bloat. It is diagnosis-only — it never deletes a file, never rewrites history, never runs git filter-repo, never migrates to LFS. When the operator approves a finding, the skill hands off: refactor-verify for delete-from-history operations (it owns the verification discipline), manage-secrets-env if a leaked credential turns up inside a blob, fight-repo-rot if the asset is unused.

What this skill is: a sorted list of what's making the repo heavy, with provenance and a proposed fix owner.

What this skill is not: a history-rewriting tool, an LFS migration executor, or a dead-code detector (that's fight-repo-rot). It surfaces bloat; it does not remove bloat.

State assumptions — before acting

Before starting the procedure, write an explicit Assumptions block. Don't pick silently between interpretations; surface the choice. If any assumption is wrong or ambiguous, pause and ask — do not proceed on a guess.

Required block: