agentforce-voice-builder
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill manages the potential attack surface for indirect prompt injection by processing caller utterances and retrieved articles (SKILL.md, references/implementation.md). It defines clear boundaries by instructing the agent to treat external content as data rather than instructions (assets/voice-instructions-template.md). It also provides specific security test cases (assets/call-tests.csv, Case V11) to verify the agent's ability to ignore embedded commands.
- [EXTERNAL_DOWNLOADS]: The skill references official technical documentation and optimization guides from Salesforce, Google, and ElevenLabs (references/source-notes.md). These external resources are used to ground agent design decisions in verified provider practices.
Audit Metadata