Use scratch to create, audit, migrate, or
retire project scratch data under a visible <repository>.scratch sibling. The
skill keeps linked Git worktrees under <repository>.worktrees, classifies
scratch data by recovery value, and requires an inventory and exact review
before cleanup. Before deleting anything, verify sidecar ownership and process
state, inspect for unpublished or unique work, produce an exact dry-run
inventory, and obtain explicit authorization for each recovery item to delete.
See the cleanup procedure and the
retention and migration contract.
Its layout and manifest contract
defines the standard categories, task-level scratch.toml metadata, retention
rules, and migration procedure for existing flat project directories.