AgentEval

CX WIND TUNNEL

Handbook: agenteval.gitbook.io

🔑 Your Anthropic API Key (Bring Your Own Key)

Used only for this test run, never stored. Get one at console.anthropic.com — a full test costs roughly $1–2.

Find the failures before your customers do.

Generate adversarial synthetic customers, run them against your agent, and surface operational failures you never thought to test for.

🎯 Test Domain

What does your agent do?

Be specific — personas are tailored to this.

Number of synthetic customers

8

Turns per conversation

4

🤖 Your Agent

Live Endpoint

Prompt (Demo)

Agent endpoint URL

Reply field name

0

Failures found

0

You didn't test for

0

Passed clean

0

Total tested

Findings

Test safety boundaries before attacks succeed in production.

Generate targeted adversarial safety probes across 7 RAI categories and find out which ones break your agent before real attackers do.

🎯 Test Domain

What does your agent do?

Number of safety probes

7

Ideally 1+ probe per selected category.

Turns per conversation

4

RAI categories to test

🤖 Your Agent

Live Endpoint

Prompt (Demo)

Agent endpoint URL

Reply field name

0

Safety failures

0

Boundaries held

0

Total probes

Safety Findings