Module 03 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. Actions should be sorted by…
- A. how often they run
- B. what undoing them involves
- C. how much they cost
- D. how consequential they feel
2. A broad query tool cannot be made safe by instructions because…
- A. instructions are too long
- B. the tool executes whatever arrives, whatever the instruction said
- C. models ignore instructions
- D. it is slower
3. A 500-unit refund ceiling belongs…
- A. in the agent's instructions
- B. as a written rule, where an auditor can read it and a check can enforce it
- C. in the tool's docstring
- D. in the review card
4. An idempotency key should be derived from…
- A. a timestamp
- B. the run id and the step name
- C. a random uuid per call
- D. the action's arguments
5. An entry on the never list is only real if…
- A. it is written in bold
- B. something other than an instruction enforces it
- C. it appears in the prompt
- D. the reviewer has read it
---
Answer Key and Explanations
Check these only after attempting every question.
1. B - what undoing them involves
Deleting a test record feels minor and is irreversible; a large internal note feels significant and is one click to remove.
2. B - the tool executes whatever arrives, whatever the instruction said
Every restriction written in prose is a sentence that may or may not be honoured. A narrow tool cannot do the thing at all.
3. B - as a written rule, where an auditor can read it and a check can enforce it
It is a policy decision rather than a formatting rule, and a policy written into a prompt is a request to something that already has the amount.
4. B - the run id and the step name
A key generated per call makes every retry look like new work. Arguments cannot distinguish two real requests from two retries.
5. B - something other than an instruction enforces it
'Never read salary data' holds if no tool reads salary data. It does not hold if a general query tool exists and a sentence asks it not to.
Practical Check
Apply this module to your own work: complete the module activity for *Tools, Permissions and Safe Action Boundaries*, 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.