The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted CSV data that is parsed and inserted into the ServiceNow database via import sets. If an attacker controls the CSV content, they could embed malicious instructions targeting the AI agent when it subsequently reads this data.
Ingestion points: CSV parsing logic in importCSVData function within SKILL.md.
Boundary markers: None observed; values are parsed directly and stored in database fields.
Capability inventory: snow_execute_script allows execution of server-side code, and snow_query_table allows database access. bulkUpdate and bulkDelete functions can modify or remove records.
Sanitization: Limited to header sanitization (replace(/[^a-z0-9]/g, "_")). Cell values are not sanitized before being stored.
[COMMAND_EXECUTION]: The skill makes use of snow_execute_script to run server-side JavaScript (ES5) within the ServiceNow environment, which is powerful but consistent with the skill's purpose as a developer tool.
[DATA_EXFILTRATION]: The exportToCSV, exportToJSON, and exportToXML functions allow for bulk extraction of data from ServiceNow tables. While intended for legitimate migration, these patterns could be used to extract sensitive organizational data if misused.

import-export