Module 05 Knowledge Check
5 questions. Pass mark 4 out of 5. Answer every question before checking the answer key below, then retry after reading the feedback.
1. The critical distinction for any tool is…
- A. Fast or slow
- B. Whether it reads or whether it changes something
- C. Free or paid
- D. Local or remote
2. Tool inputs produced by a model should be…
- A. Passed straight through
- B. Validated and constrained before execution
- C. Logged only
- D. Shortened
3. Least privilege for a tool node means…
- A. Fewer tools overall
- B. Each tool has only the access its job requires, nothing broader
- C. Read-only always
- D. One tool per node
4. If untrusted content (an email, a web page) can reach the model, you must assume…
- A. It is safe if short
- B. It may contain instructions attempting to trigger tool calls
- C. The model will ignore it
- D. Filtering solves it
5. Which action clearly warrants confirmation before execution?
- A. Fetching a public page
- B. Sending a message to a customer
- C. Reading a local file
- D. Counting rows
---
Answer Key and Explanations
Check these only after attempting every question.
1. B - Whether it reads or whether it changes something
Read tools are recoverable; write and send tools are not. Controls should differ accordingly.
2. B - Validated and constrained before execution
The model proposes; your code decides. Validation is where that boundary is enforced.
3. B - Each tool has only the access its job requires, nothing broader
It bounds the damage of a wrong call, a bad input, or a prompt-injection attempt.
4. B - It may contain instructions attempting to trigger tool calls
Indirect prompt injection targets exactly this path; the defence is permissions and confirmation, not trust.
5. B - Sending a message to a customer
Outbound, irreversible, externally visible effects are the canonical case for a human gate.
Practical Check
Apply this module to your own work: complete the module activity for *Tool Nodes and Controlled Action Execution*, then write one sentence naming what your result shows and one naming what it does not.
Strong Answer Pattern
A strong answer names the task, the evidence used, the check performed, and the remaining limitation. It avoids "proved", "guaranteed", or "always" unless the evidence genuinely supports it.