building-voice-asa-agent
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill instructs the implementation of flows using
SystemModeWithoutSharing. In the Salesforce ecosystem, this allows the process to run in a system context, bypassing the sharing rules and object permissions of the calling user. While this is a standard design pattern for Agentforce to ensure the agent can access necessary data without having overly broad direct permissions, it constitutes a controlled privilege escalation surface that requires careful logic within the flows. - [INDIRECT_PROMPT_INJECTION]: The agent is designed to process untrusted data in the form of user utterances via voice and messaging channels. It possesses functional capabilities to perform write operations (creating/updating records). The skill mitigates this by recommending a 'verification gate' and specific subagents for off-topic handling, but the fundamental surface for indirect injection exists wherever user input influences automated actions.
- [COMMAND_EXECUTION]: The instructions involve the use of the Salesforce CLI (
sf agent) for generating, validating, and publishing agent bundles. These are standard development and administrative operations within the intended platform environment.
Audit Metadata