hv-maintenance

Inventory & cleanup

Preflight

Requires the desktop-hypervisor MCP server: confirm mcp__hypervisor__vm_info is available. Missing → stop; register with claude mcp add --scope user hypervisor -- <binary> (MCP Registry: io.github.bryanjbelanger/desktop-hypervisor-mcp) and restart the session. Run the inventory per ready provider (provider action=list, then pass provider= to scope each pass).

This skill is read-heavy and delete-careful: enumerate fully, propose, get explicit confirmation, then delete.

Inventory pass (always safe)

• vm_info action=list vs action=running → the stopped set
• vm_info action=show vm=NAME per stopped VM → specs and disk footprint
• snapshot action=list vm=NAME tree=true → deep or stale snapshot chains (deltas are often where the disk actually went)
• Image cache: ~/.hypervisor-images (or HV_IMAGE_DIR) grows by design — idempotent downloads; report its size and note it's safe to prune old images.
• Provider-native orphans (unregistered disks, stale media entries) are outside the neutral surface: use execute_command with the provider's own argv (VBoxManage list hdds / closemedium; vmrun listSnapshots etc.) and say which provider you're driving.

Cleanup rules

• NEVER delete without listing exactly what will be removed and getting a yes — a "stale" VM may be the user's long-term environment.
• VM removal: vm_lifecycle action=stop vm=NAME hard=true (if running) → vm_lifecycle action=delete vm=NAME (removes the VM and its files).
• Confirmed-good changes: delete their safety snapshots (see hv-snapshot).

Report

Table: item, provider, size/age, reason flagged, action taken or proposed. Totals for space reclaimed vs. reclaimable-pending-approval.