ai-disk-doctor

AI Disk Doctor

Use the AI Disk Doctor MCP tools as the workspace storage governance layer. The local AI Disk Doctor Core is the only source of truth for scanning, model inventory, history, diff, rules, policy, risk, and future cleanup planning.

When To Use

Use this Skill for questions such as:

• Why is my disk suddenly full?
• Which AI tools, IDEs, or model caches are consuming storage?
• What did Claude, OpenCode, Cursor, Ollama, or other tools leave locally?
• What changed since the previous scan?
• Which findings need review, and what evidence supports that classification?

I3 Workflow

1. Complete the package setup and verification path in docs/I3_SELF_SERVICE.md before asking for a natural storage diagnosis. Keep Claude Code in the safe-alpha-v1 profile.
2. For a natural storage diagnosis such as why the computer is getting full, call aidisk_workspace_explain as the primary entry point with {} and no category field. Its handler performs the capability compatibility check internally.
3. Supply aidisk_workspace_explain.category only when the user explicitly requests a focused, named Core category. Preserve the user's requested category without guessing a different value.
4. Use aidisk_capabilities for explicit capability inspection or compatibility troubleshooting, such as what AI Disk Doctor supports or whether the Core contract/schema is available. It is not required before aidisk_workspace_explain.
5. Call core_status when the Core may be missing or the environment is unfamiliar. If workspace explain reports an unavailable or incompatible Core, report that diagnostic state and stop.
6. Use scan_summary for explicit scanner-level findings or low-level scan inspection. Prefer aidisk_workspace_explain for general questions about why storage is growing or the workspace is full. Use its category only when the user asks for a focused area; do not provide rules, policy, root, reports, or executable paths.
7. Call ai_model_inventory only for explicit model-asset or model-inventory questions, or inspection of known AI model files. Only use the optional tool selector. Treat unknown, incomplete, custom, shared, or credential-adjacent assets conservatively.
8. Call latest_diff when the user asks what changed recently. Core owns snapshot discovery and diff semantics.
9. Explain findings with Core evidence. Preserve partial status, warning text, and bounded-output markers; do not turn handling or risk evidence into cleanup authorization.