The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data retrieved from Palo Alto Panorama APIs, such as security rules and NAT policies. This creates a surface for indirect prompt injection where adversarial instructions could be embedded in firewall object descriptions or rule comments. \n- Ingestion points: Data returned from the PAN-OS API (Managed firewalls, security policies, NAT rules). \n- Boundary markers: None explicitly defined in the skill body. \n- Capability inventory: The skill can search and review policies but gate-writes behind external workflows. \n- Sanitization: Not explicitly implemented in the skill description, relying on the agent's internal safety filters.
[COMMAND_EXECUTION]: The skill utilizes an environment variable (PANOS_MCP_CMD) to execute tool commands via the Python runtime. This is an expected architectural pattern for Model Context Protocol (MCP) skills where the specific execution path is configured within the environment.

paloalto-panorama