Storage Audit

Audits allocated disk space without changing the requested root, invoking a package manager, accessing the network, or executing a cleanup command. The bundled Bash script produces deterministic, machine-readable evidence; it renders a guarded cleanup script after the user explicitly selects current actionable candidates.

Reference Files

Workflow

1. Resolve scope. Use the user path. A request covering all projects or subfolders is workspace scope even when the root contains a Git marker or package manifest; do not let auto collapse that request into one repository. Otherwise, auto treats a Git/manifest root as one repository. Workspace discovery inventories direct children plus nested Git repositories to depth five and accepts both .git/ directories and .git files used by linked worktrees. Keep top-level multi-repository folders as repository-container rows with a nested_repo_count; list each nested repository separately. Report a discovered Git marker that Git cannot open as invalid-repository, not as an ordinary directory. Never widen the scan above the user-supplied root.
2. Run strict audit. Execute bash scripts/storage-audit.sh audit --root "<path>" --scope <resolved-scope> --strict --format json. A strict result captures candidate identity and evidence at scan time; status:"unstable" means only that aggregate allocation drifted during collection.
3. Classify evidence. Explain cleanup-eligible rebuildable entries separately from derived-review, local-data, and Git-protected entries. A-### means actionable candidate; P-### means protected or review-required candidate. Never infer that an old commit means the project checkout is disposable.
4. Report liveness. Surface project type, nested-repository count, Git cleanliness, last commit date, process count, and registered-worktree count. These are review signals, not deletion authorization.
5. Give the next action. After a strict audit with cleanup-eligible candidates, request a concise explicit selection: render all for every actionable candidate from that audit, or render A-001, A-002 for a subset. Do not paste the complete candidate list or tell the user to rerun the audit because aggregate allocation drifted.
6. Render script after selection. For render all, run plan --root <path> --scope <scope> --all. For a subset, bind every requested ID to its canonical path from the preceding report: --candidate A-001 --expected-path <canonical-path>. plan collects a fresh strict candidate snapshot and rejects an ID whose current path differs from the selected path, then prints one self-contained selected-target script. Do not save or invoke it.

Supported Operations

Error Handling

Output

Return: root, snapshot status, allocated KiB, project type and nested-repository count, project liveness rows, classified artifact rows, and exact evidence for every recommendation. A strict report with cleanup-eligible candidates ends with concise render all and subset-selection instructions. State No command executed. whenever rendering a cleanup script.