The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill enables arbitrary Bash command execution for file operations, including destructive actions like rm -rf in the user's trash and potential for accidental bulk deletion through tools like find and mv.
[DATA_EXFILTRATION] (HIGH): The skill explicitly targets sensitive user directories such as ~/Documents and ~/Desktop. Access to these paths constitutes high data exposure risk even in the absence of network commands.
[PROMPT_INJECTION] (HIGH): High vulnerability to Indirect Prompt Injection (Category 8). The skill processes untrusted external data (filenames and directory structures) and possesses significant write/delete capabilities. 1. Ingestion points: Filenames and metadata retrieved via find and ls. 2. Boundary markers: Absent. 3. Capability inventory: mv, mkdir, rm, zip, tar, and the Write tool. 4. Sanitization: Absent. A malicious file could be named to manipulate agent logic or escape shell command contexts.

file-manager