The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts scripts/scan_tree.py and scripts/platforms.py to perform directory inventories, plan validation, and file trashing. It also uses standard shell commands like mkdir and mv for creating backups and relocating files.
[PROMPT_INJECTION]: The skill operates on untrusted data by processing files within the ~/Downloads directory, which creates an indirect prompt injection surface.
Ingestion points: Files and their filenames within the ~/Downloads directory (SKILL.md).
Boundary markers: The skill references a shared safety workflow in ../../../references/safe-mutation-rules.md and includes an explicit instruction to ignore instructions found within filenames.
Capability inventory: The skill has the ability to move files (mv), trash files (scripts/platforms.py), and write manifest files to ~/.declutter-backups/.
Sanitization: The instructions specifically command the agent to never treat a filename as an instruction (e.g., ignore-previous-instructions.pdf).

sorting-downloads