The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from local documents (PDFs, scans, images, and Word files) to classify them into folders.\n
Ingestion points: The scripts/scan-inventory.py script extracts text excerpts from documents, and the agent performs visual analysis on image-based files.\n
Boundary markers: The instructions require the agent to cite a specific 'basis' for every classification in the manifest, which acts as a logical check against the document content.\n
Capability inventory: The skill utilizes scripts/apply-move-plan.py to execute file moves on the local filesystem.\n
Sanitization: The apply-move-plan.py script incorporates an is_within function to validate that all source and destination paths remain within the designated root directories, preventing path traversal. It also includes collision-safe naming to prevent data loss via overwriting.

cleanup-documents