service-itsm-slack-itservice-configure
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from the Salesforce organization (e.g., User and Digital Experience site records) that could potentially contain malicious natural language instructions designed to influence agent behavior.
- Ingestion points: Data enters the context via Salesforce API responses for user records (
/services/data/v67.0/query) and communities (/services/data/v67.0/connect/communities) as described inSKILL.md(Steps 3, 4, 5). - Boundary markers: The instructions do not define strict boundary markers or delimiters when interpolating these retrieved values into the conversation context.
- Capability inventory: The skill possesses state-changing capabilities via the
mcp__headless-360__dispatchtool (performing PATCH and POST operations to update permissions and org preferences) and can execute local scripts viaBash. - Sanitization: The
scripts/classify-user-access.mjsscript performs JSON parsing and deterministic field extraction, which provides a layer of structure, but the resulting strings are used by the agent without specific sanitization against prompt injection attacks.
Audit Metadata