Scan Organizer

What it does

Extracts text from scanned PDFs (Docling + vision OCR), classifies documents into categories using a language model, and moves files into <scans_dir>/<category>/ with .md and .meta.json sidecars. Provides auditability via a manifest and supports dry-run, reclass, undo, and single-file processing.

When to use it

Use when you have a directory of scanned PDFs that need automated text extraction, classification (medical, financial, insurance, tax, legal, personal, household, other), and structured output for downstream workflows (search, ingestion, compliance). Good for medical records, invoices, insurance paperwork, and bulk archive processing.

What's included

• Scripts: project exposes uv commands to process, status, undo, reclass and dry-run (see Commands).
• References: No external references directory detected; the repo contains project files and metadata.
• Instructions: Clear CLI runbook for processing new scans, dry-run previews, single-file processing, force re-process, status checks, undo, and reclassification. Outputs JSON and writes manifest for audit.

Compatible agents

Best used with agents that can call shell tooling and file I/O (Copilot/Codex/CLI-capable agents). The skill assumes an OpenAI-compatible model endpoint for classification (OpenAI, Ollama, OpenRouter).