F08 — Malformed email and correction (High)

Verdict: FAIL

The agent failed to handle the email collection flow correctly. In Turn 3, the visitor provided their name ('Jordan UAT') along with a malformed email ('jordan.uat.example.com'). The agent should have: (1) recognized and extracted the name, (2) flagged the malformed email and asked for correction. Instead, it ignored all provided information and asked 'What company are you with?' repeatedly. In Turn 4, the visitor provided a corrected email (redacted-632d0c6473@example.invalid') which the agent should have normalized to redacted-64c9cf7e76@example.invalid', but again it ignored this and asked 'What company are you with?' In Turn 5, 'yes please' was offered as a response to what appeared to be an implicit consent question, but the agent still asked 'What company are you with?' — showing the agent was stuck in a loop and never progressed past the company field prompt despite the visitor providing name and email across multiple turns. The agent never asked to correct the malformed email (it just ignored everything), never normalized the escaped @, never asked final consent once a valid email was provided, and the handoff was not captured. All pass criteria are violated.

Webhook submissions

None captured.

Transcript

Turn 1 (first token 1545ms, total 1546ms)

Visitor: Build a one-day agenda for 21 October on AI and skills.

Agent: Your UNLEASH in Paris agenda

Times are local to the event. Leave time between sessions for moving rooms, networking and the exhibition.

The agenda may still be updated. See the full Agenda.

Alternatives — these conflict with the plan above; choose these instead of the overlapping session, not in addition:

Would you like me to email your selected agenda?

Turn 2 (first token 45ms, total 45ms)

Visitor: yes please

Agent: What company are you with?

Turn 3 (first token 377ms, total 377ms)

Visitor: I am Jordan UAT. Email it to jordan.uat.example.com

Agent: What company are you with?

Turn 4 (first token 39ms, total 39ms)

Visitor: redacted-690a8df2c0@example.invalid

Agent: What company are you with?

Turn 5 (first token 50ms, total 50ms)

Visitor: yes please

Agent: What company are you with?