The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the use of start_process and interact_with_process to execute arbitrary shell commands, providing persistent terminal access to the host system.
[DATA_EXFILTRATION]: The instructions guide the agent to access sensitive file paths, specifically system logs located at /var/log/app.log. Combined with the documented curl and ssh tool usage, this establishes a functional chain for data exfiltration.
[REMOTE_CODE_EXECUTION]: The skill provides detailed workflows for executing dynamic code strings in Python (python3 -i) and Node.js (node:local) REPL environments.
[COMMAND_EXECUTION]: The agent is instructed to use kill_process to terminate arbitrary OS processes via their PID, presenting a risk of system disruption or denial of service.
[PROMPT_INJECTION]: The skill exposes a significant surface for indirect prompt injection by ingesting untrusted data from the filesystem and providing high-impact capabilities like shell access.
Ingestion points: read_file, read_multiple_files, start_search, read_process_output.
Boundary markers: None identified in the instructional content.
Capability inventory: start_process, write_file, move_file, edit_block, kill_process.
Sanitization: None specified; instructions encourage direct interpolation of content into tool arguments.

desktop-commander-overview