The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from ServiceNow records to generate summaries and recommendations.
Ingestion points: The skill reads case descriptions from the sn_customerservice_case table and journal entries from the sys_journal_field table, which are typically populated by external customers or other users.
Boundary markers: While the summary follows a scannable format, there are no explicit boundary markers or instructions to the model to ignore instructions embedded within the record data.
Capability inventory: The agent has the ability to execute Bash commands and perform queries through ServiceNow-specific MCP tools.
Sanitization: The procedure does not mention any sanitization or validation of the retrieved text before it is presented in the sidebar summary.
[DATA_EXFILTRATION]: The skill retrieves sensitive customer information (PII) from ServiceNow to build context for agents.
Evidence: The skill performs queries on the csm_consumer and customer_account tables to retrieve names, email addresses, phone numbers, and physical location data.
Context: This data retrieval is a core function of the skill and is performed within the user's ServiceNow environment to assist with legitimate customer service tasks.

csm-sidebar-summarization