call-follow-up
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill follows professional standards for CRM data management and meeting follow-ups.
- [DATA_EXPOSURE]: The skill uses Salesforce-authorized GraphQL endpoints (
/services/data/v66.0/graphql) for reading account and opportunity information. All data access is scoped to the user's active session and relies on legitimatedispatch_readonlytools, which is the standard mechanism for secure agent interaction with Salesforce. - [COMMAND_EXECUTION]: No dangerous shell commands or system-level scripts are present. The skill uses structured tool calls for internal logic.
- [PROMPT_INJECTION]: The instructions include clear 'global-rules-bootstrap' to manage tool connectivity and silent execution, which are standard operational guidelines for AI agents rather than malicious overrides.
- [INDIRECT_PROMPT_INJECTION]: The skill processes meeting transcripts which are external data. However, it follows best practices by extracting specific structures (decisions, commitments) and grounding them in Salesforce data before use. The impact is restricted to drafting emails and Slack messages, with the skill explicitly stating 'Do not send', maintaining a human-in-the-loop requirement for all external communication.
- [REMOTE_CODE_EXECUTION]: There are no remote downloads, package installations, or external script executions. All functionality resides within the platform's provided tools.
Audit Metadata