hardNewAnthropicOpenAICursor

Users approve 93% of our coding agent's permission prompts. Should we replace the prompts with an automatic classifier?

Try it in the practice workspaceDraft your answer before comparing it with the rubric.

A prompt that is almost always approved protects less than it seems, and a classifier that misses one risky action in six protects less than it claims. Count the misses both ways, then decide where neither may decide alone.

Updated Nov 2026 · Original practice questions with worked answers and teaching rubrics. Our editorial policy explains the sources, review process and limits.

Case briefFacts and constraints for your answer

All figures are an illustrative fixture. A coding agent raises 2,000 permission prompts a day and users approve 93% of them. Engineering proposes an automatic classifier instead.

Scenario inputs
Risky actionsAbout 3% of prompts, roughly 60 a day, are actions a careful reviewer would refuse; users deny 45 of them.
Published exampleOne vendor reports 0.4% false positives and 17% false negatives for its own classifier pipeline, on its own traffic.
EnvironmentAgents run in a container that can reach staging systems.

Your task. Decide whether to replace the prompts, for which actions, how you would test the classifier first, and what must hold whatever it decides.

Follow-up roundsAnswer first, then revise for each changed constraint
  1. Round 2. Shadow mode is done. Of 1,000 labelled prompts, 40 were risky, and the classifier approved 9 of them.
  2. Round 3. A month later, an auto-approved command deleted a customer's staging database.
PREMIUM / REFERENCE ANSWER

Compare your call with a worked answer.

YOUR PRACTICE GOAL

Decide whether to replace the prompts, for which actions, how you would test the classifier first, and what must hold whatever it decides.

  1. 01

    The reasoning

    A worked response, with assumptions and the conditions that change the call.

  2. 02

    The follow-ups

    Questions that challenge the answer, with reference responses to compare.

  3. 03

    The rubric

    Weak, Hire and Strong Hire signals for reviewing your own attempt.

View Premium access

All 441 answers beyond your current access, including this reference.

Try the free launch case first

One membership, the full preparation library.Every published course, answer, concept and PDF guide.

One payment · No auto-renewal · Downloaded PDFs are yours to keep

UP NEXT ON YOUR JOURNEY
NOTES & DISCUSSION · 0

No comments yet. How would you have answered?