The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing local scripts via environment variables (e.g., $NETBOX_MCP_SCRIPT, $PYATS_MCP_SCRIPT, $SERVICENOW_MCP_SCRIPT) using a wrapper command $MCP_CALL. While these are intended tool invocations for the platform, they involve shell execution and script parameters.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection as it ingests and processes data from external systems (NetBox) and live network hardware.
Ingestion points: External data enters the context through netbox_get_objects and pyats_run_show_command in SKILL.md.
Boundary markers: The skill lacks explicit delimiters or instructions to treat data from network devices (such as interface descriptions or hostname strings) as untrusted, which could potentially contain malicious instructions aimed at the agent.
Capability inventory: The agent has capabilities to execute shell commands, create ServiceNow incidents, and record turns in GAIT.
Sanitization: There is no documented evidence of sanitization or validation of the data retrieved from device command outputs before it is interpolated into incident descriptions or reports.

netbox-reconcile