The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a potential surface for indirect prompt injection due to its handling of external data.
Ingestion points: Untrusted data enters the agent context from network devices via pyATS commands and from NetBox object queries as described in SKILL.md.
Boundary markers: The reconciliation workflow lacks explicit delimiters or markers to distinguish between system instructions and the data being processed from devices.
Capability inventory: The skill has the ability to execute multiple automation scripts through the $MCP_CALL pattern and interact with external systems like ServiceNow.
Sanitization: There is no evidence of data sanitization or validation logic applied to external device content before it is interpolated into generated reconciliation reports or ServiceNow incident descriptions.

netbox-reconcile