The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes access to the Bash native tool, allowing the agent to execute shell commands on the host system.
[REMOTE_CODE_EXECUTION]: The skill utilizes the SN-Execute-Background-Script tool to run dynamically generated JavaScript on the connected ServiceNow server. While this is used for legitimate data aggregation (Step 5 and Step 6), the ability to execute arbitrary server-side code is a high-privilege operation.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes free-text data from workplace incident reports.
Ingestion points: Untrusted data enters the agent context from the sn_hr_hs_incident (short_description, description) and sn_hr_hs_investigation (findings, root_cause) tables via SN-Query-Table and SN-Read-Record tools.
Boundary markers: The instructions do not define clear delimiters or use "ignore embedded instructions" warnings when the agent retrieves and processes these text fields.
Capability inventory: The agent has access to powerful capabilities including SN-Execute-Background-Script, Bash, and the ability to modify records via SN-Add-Work-Notes.
Sanitization: No sanitization, filtering, or validation of the retrieved incident data is performed before it is summarized or used in reporting logic.

health-safety-summarization