The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted filenames and directory metadata from the local system to generate organization plans and summaries, creating a surface where maliciously named files could influence the agent's behavior.
Ingestion points: File system metadata gathered via ls, find, and du commands in SKILL.md.
Boundary markers: The skill uses a structured Markdown template ("Organization Plan") to present its analysis, providing basic separation between data and instructions.
Capability inventory: The skill utilizes mv, mkdir, ls, find, and md5 within SKILL.md for file discovery and manipulation.
Sanitization: Instructions do not explicitly specify sanitization of file paths or names before they are interpolated into shell commands or included in generated plans.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands to perform file analysis and organization tasks.
Evidence: Execution of commands like ls -la, find, du -sh, mkdir -p, and mv is specified in SKILL.md. These commands incorporate user-supplied directory paths, which could pose a risk if the paths contain shell metacharacters and are not properly sanitized by the agent.

file-organizer