The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/process-watch.py script contains a kill command that allows the termination of system processes using psutil.Process.send_signal. This high-privileged capability could be misused to cause system instability or denial of service if the agent is instructed to target critical processes.
[DATA_EXFILTRATION]: The skill facilitates the exposure of sensitive system data. Specifically, the info command in scripts/process-watch.py displays process command lines and lists of open file paths. Command-line arguments frequently contain sensitive credentials, API tokens, or secrets, while file handles can reveal the locations of private user data (e.g., SSH keys). Although the skill instructions suggest redacting this data, the provided implementation lacks automated sanitization.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting untrusted data from the system environment.
Ingestion points: Process names and command-line strings are retrieved via psutil in scripts/process-watch.py and presented to the agent.
Boundary markers: The script output does not use delimiters to wrap the untrusted system data, increasing the risk of the agent interpreting metadata as instructions.
Capability inventory: The skill has capabilities for process termination and network connection monitoring.
Sanitization: The script performs no escaping or validation of process-related strings before they are processed by the agent.

process-watch