Testing

Free Canadian small business workbook

Agent Testing Scenario Pack

A worksheet for testing agent behavior across normal cases, edge cases, restricted requests, tool failures, and escalation.

7 pages · PDF workbook · No email required

Actual working-record worksheet from the Agent Testing Scenario Pack, page 3 of 7.
Preview all 7 pages Read the complete workbook before downloading.
Preview the complete PDF All 7 pages · Free to explore

Scroll through the full workbook below. Download the PDF for selectable text, printing and your own working copy.

Workbook page 1 of 7. For selectable text, use the free PDF download.
Page 1 of 7
Workbook page 2 of 7. For selectable text, use the free PDF download.
Page 2 of 7
Workbook page 3 of 7. For selectable text, use the free PDF download.
Page 3 of 7
Workbook page 4 of 7. For selectable text, use the free PDF download.
Page 4 of 7
Workbook page 5 of 7. For selectable text, use the free PDF download.
Page 5 of 7
Workbook page 6 of 7. For selectable text, use the free PDF download.
Page 6 of 7
Workbook page 7 of 7. For selectable text, use the free PDF download.
Page 7 of 7

What’s inside

A working guide
for a real decision.

Build a repeatable scenario suite for agentic workflows before launch and use it as a regression backlog as tools, prompts, and policies change. Use this guide to make the operating choice explicit, test it with evidence, and record who owns the next decision.

Who this is for

Agentic systems using tools, approvals, or multi-step workflows.

Bring the people who own the work

Include the workflow owner, a person who performs the work, a reviewer with authority to approve or stop it, and the relevant data or tool owner.

Includes an explainer, working worksheet, three illustrative examples and practical considerations. This is a planning PDF, not prebuilt automation software.

Use it with your own work

Five questions
to work through.

Choose one workflow. Record enough context to support a decision, then use the worksheet to name the owner, evidence and next review.

  1. Normal case

    What should the agent do when context and tools are healthy?

  2. Missing data

    How should the agent behave when required state is unavailable?

  3. Conflict

    What should happen when sources or user instructions disagree?

  4. Restricted request

    How should the agent handle unsafe, unauthorized, or policy-breaking requests?

  5. Tool failure

    What retry, fallback, or escalation path is expected?

Illustrative examples from the workbook

Where this becomes practical.

Consulting Practice

Test missing scope details

A proposal agent should ask for clarification rather than invent a deadline.

Independent Retailer

Test a failed supplier lookup

An ordering agent must not claim an order was placed when the tool failed.

Accounting Practice

Test restricted client access

An agent must refuse retrieval from another client’s file and produce an auditable error.

AI-supported role
Run approved scenarios and record tool calls, refusals, errors, escalation, and final outputs.
Human authority
Test owners define expected behavior, inspect failures, and approve release criteria.
What to measure
Track scenario coverage, pass rate, severe failures, regression rate, and unresolved defects.

Illustrative scenarios, not client results or a promise of performance. The workbook includes additional case-specific prompts.

A useful first step

Use the workbook.
Bring us the next question.

Download and use it without booking a call. If you want help applying it, email Chris with the workflow you’re considering and what you want to improve.

Discuss your workflow

Further reading in the library

General operational guidance only. These resources are not legal, privacy, cybersecurity or regulatory advice.