The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from ServiceNow incident records, including work notes and comments which are accessible to and populated by various users. This creates a surface for indirect prompt injection where malicious instructions could be embedded in these external fields to influence the agent's subsequent coordination actions.
Ingestion points: SN-Query-Table and SN-Natural-Language-Search in SKILL.md are used to retrieve content from the incident and sys_journal_field tables.
Boundary markers: The skill instructions do not implement explicit boundary markers or instructions to treat data retrieved from ServiceNow as untrusted content.
Capability inventory: The agent has extensive permissions to modify the ServiceNow environment using tools like SN-Update-Record, SN-Assign-Incident, SN-Add-Comment, and SN-Add-Work-Notes.
Sanitization: There are no instructions for the agent to sanitize, escape, or validate the information retrieved from ServiceNow before it is processed or used in communications.

major-incident