The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to run pip install commands to resolve missing dependency errors identified in logs. This poses a risk if an attacker provides a malicious package name in a workflow or log file that the agent then attempts to install.
[EXTERNAL_DOWNLOADS]: The guide recommends using a download_model tool with arbitrary URLs to replace corrupted model files. If the URL source is not validated, the agent could be tricked into downloading and placing malicious files on the system.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes **/*.json files (ComfyUI workflows) which are untrusted data sources.
Ingestion points: ComfyUI workflow files (**/*.json).
Boundary markers: None identified; the agent is not instructed to ignore instructions embedded within the workflow data.
Capability inventory: The skill uses tools for command execution (pip install), remote downloads (download_model), and system information gathering (get_logs, get_system_stats).
Sanitization: No validation or sanitization of the content within the JSON workflows is mentioned before the agent acts upon the troubleshooting advice.

troubleshooting