The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-controlled file names and contents to generate categorization and renaming proposals. While the tool requires explicit user approval of a plan ID before moving files, malicious instructions embedded in file data could potentially influence the agent's categorization logic. * Ingestion points: File metadata and excerpts are ingested via the file_sorter.py scan command and the Read tool. * Boundary markers: The workflow enforces a 'plan before move' requirement with a unique plan_id for confirmation. * Capability inventory: The helper script performs directory creation and file moves. * Sanitization: The file_sorter.py script sanitizes proposed category and subcategory names by removing forbidden path characters and enforcing length limits.

file-sorter