claw-drive_skill

Overview

This skill is an AI-managed personal drive called Claw Drive that auto-categorizes, tags, deduplicates, and retrieves files using natural-language commands. It stores files locally under ~/claw-drive, optionally syncs selected files to Google Drive, and provides a searchable index to find items quickly. Privacy-first rules ensure file contents are never read without explicit consent.

How this skill works

When storing a file, the skill runs a privacy check, asks for consent to read contents, extracts text or image descriptions (only with permission), classifies into a category, chooses a destination path, creates a descriptive filename and rich description, tags the file, and stores it via the claw-drive CLI which handles copying, hashing, and deduplication. For retrieval, the skill spawns a lightweight sub-agent to search INDEX.jsonl and returns the best matches without loading the full index into the main session.

When to use it

• When you receive attachments (email, chat, upload) that need safe, indexed storage
• When you want to archive or back up files with deduplication and descriptive metadata
• When you need to find a previously stored file using natural language
• When preparing a bulk import or migration of many files into an organized archive
• When you want optional Google Drive sync while keeping local control of data

Best practices

• Always use the claw-drive CLI to store/update/delete files—do not write directly to ~/claw-drive/
• Ask the user for permission before reading file contents; default to sensitive if no reply
• Keep descriptions concise but include key searchable details (dates, amounts, IDs) without full sensitive data
• Use 1–5 lowercase tags, include the category as a tag, and reuse existing tags when possible
• For searches, let the sub-agent read INDEX.jsonl and return only parsed matches to the main session

Example use cases

• Store an emailed W-2: extract employer and wages, tag tax-year and finance, name it w2-2025.pdf
• Archive a vet invoice: extract clinic, pet name, amount; store under medical with tags like max and invoice
• Save a passport scan as sensitive: place in identity/ and store only filename and user-provided description
• Migrate a messy folder: run migrate scan, review the classification plan, and apply with deduplication
• Find a document by query: spawn search sub-agent to return top matches from INDEX.jsonl, then deliver the chosen file