AgentTesting

TEST CASE / TRACE / CONTAINMENT

Find the failure before the agent finds a customer.

Security and evaluation playbooks for agents that answer, retrieve, decide, and call tools.

Ten practical field guides

Source-linked workflows, examples, checklists, and decision records. Written by AgentTesting Product Team and reviewed July 13, 2026.

agent-evals·9 min

Build an AI Agent Evaluation Plan

A practical, source-linked guide to turn a vague safety review into a versioned test matrix with owners, thresholds, and evidence.

Use the guide →
prompt-injection·5 min

Prompt Injection Test Cases for Tool-Using Agents

A practical, source-linked guide to test direct and indirect instructions that try to override policy or exfiltrate data.

Use the guide →
permissions·5 min

The Agent Tool-Permission Matrix

A practical, source-linked guide to grant each agent only the functions, data, scopes, and autonomy needed for its stated job.

Use the guide →
regression-testing·9 min

Build a Regression Dataset From Real Agent Failures

A practical, source-linked guide to convert incidents, corrections, edge cases, and policy changes into repeatable evaluations.

Use the guide →
factuality·5 min

Test Agent Hallucinations Against an Authoritative Fact Set

A practical, source-linked guide to measure whether answers are supported, current, scoped, and appropriately uncertain.

Use the guide →
human-in-the-loop·5 min

Human Approval Boundaries That an Agent Cannot Rewrite

A practical, source-linked guide to design confirmations from trusted facts rather than attacker-controlled summaries.

Use the guide →
observability·5 min

Trace an Agent Failure From Input to Tool Result

A practical, source-linked guide to capture enough structured evidence to explain and reproduce a multi-step failure.

Use the guide →
red-teaming·9 min

AI Agent Red-Team Runbook

A practical, source-linked guide to run a bounded adversarial exercise that produces fixes and regression cases instead of a dramatic demo.

Use the guide →
latency·5 min

Balance Agent Latency, Cost, and Answer Quality

A practical, source-linked guide to compare configurations without hiding failures inside averages.

Use the guide →
vendor-evaluation·5 min

Evaluate an Agent Vendor Before Connecting Production Data

A practical, source-linked guide to compare security, evaluation, observability, portability, and exit controls before a purchase.

Use the guide →