how can i force the agent not to answer account questions unitl it gets past verification

[tawny elk]

AI-generated Answer for sari_33713

Short answer: Gate “account” intents behind verification using guardrails + a verification tool + workflow routing, and test it.

How to implement (documented features):

• Add explicit guardrails in the system prompt that prohibit answering account/billing/PII questions unless a “verified” flag is true. ElevenLabs recommends a dedicated # Guardrails section for non‑negotiable rules (e.g., “Never reveal account info without proper verification.”). (elevenlabs.io)
• Create a verification step as a Tool (server tool you host) that sends/validates an OTP and returns a JSON result. Tools are the mechanism for custom logic; describe when/how the tool must be used in your prompt. (elevenlabs.io)
• Store the verification outcome in a dynamic variable (e.g., verified=true) by having the tool return JSON that updates dynamic variables. Then key your prompt logic off that variable. (elevenlabs.io)
• Enforce the sequence with Workflows:
  • Start with a Dispatch Tool node that runs the verification tool.
  • Route success to a Subagent node that enables account tools; route failure to retry or end.
  • Subagent nodes can change available tools/KB and tighten instructions after verification. (elevenlabs.io)
• Optionally restrict session access with Agent authentication (signed URLs and/or allowlists) so only your authenticated front end can even start conversations. This is separate from in‑conversation verification but adds a security layer. (elevenlabs.io)
• Write Agent Tests to assert: (1) before verified, the agent refuses account info; (2) after verified=true, it can proceed. Tool‑call testing is a documented use case for data security and authentication behaviors. (elevenlabs.io)
• For ongoing QA, add a Success Evaluation criterion like “No sensitive info disclosed prior to verification” to monitor transcripts. (elevenlabs.io)

This approach uses only built-in capabilities: prompt guardrails, Tools, Dynamic Variables, Workflows, Authentication, and Testing—all documented in ElevenLabs’ Agents Platform. (elevenlabs.io)

Sources:
Prompting guide | ElevenLabs Documentation
Tools | ElevenLabs Documentation
Dynamic variables | ElevenLabs Documentation
Workflows | ElevenLabs Documentation
Agent authentication | ElevenLabs Documentation
Agent Testing | ElevenLabs Documentation
Success Evaluation | ElevenLabs Documentation
Prompting guide | ElevenLabs Documentation

[simple stratus]

i a m in workflow