voice-profile
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from sent emails to generate a writing profile meant for inclusion in future system instructions.
- Ingestion points: Sent email messages from the user's connected mail account.
- Boundary markers: None present to delimit email content during processing.
- Capability inventory: Salesforce MCP tools (
dispatch_readonly,dispatch) and email reading capabilities. - Sanitization: Basic cleaning is performed (stripping signatures and quoted text), but no safety-specific sanitization of email content is applied.
- [DATA_EXFILTRATION]: The skill accesses sensitive personal communication (sent emails). This is the primary purpose of the skill and the results are returned directly to the user without being sent to unauthorized third-party domains.
- [COMMAND_EXECUTION]: The skill uses the
dispatch_readonlytool to execute GraphQL queries against the Salesforce API to verify server connectivity and retrieve user identification. This is a standard use of the authorized vendor tools.
Audit Metadata