The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because it ingests and processes unverified metadata from network-connected IP cameras.
Ingestion points: The scripts/onvifctl.py script extracts various strings (e.g., Manufacturer, Model, Scopes) from XML responses returned by cameras during discovery and information gathering.
Boundary markers: The skill lacks explicit delimiters or instructions to the agent to ignore or isolate instructions that might be embedded in the metadata returned by a camera.
Capability inventory: The skill allows the agent to perform network requests (SOAP over HTTP) and write local files (snapshots), which could be misused if the agent is tricked into executing unauthorized commands by malicious metadata.
Sanitization: Camera-provided data is returned to the agent without filtering or sanitization to remove potential prompt injection payloads.
[SAFE]: The skill implements standard ONVIF protocols (SOAP/XML and WS-Security) and avoids hardcoded credentials by supporting environment variables and command-line arguments.
[SAFE]: The script implementation is restricted to the Python standard library and does not utilize risky functions such as eval(), exec(), or subprocess spawning.

ptz-onvif