An approval gate pauses the agent before an action that needs human permission. The permission prompt should show the exact command or operation, why it is needed and what resources it can affect. The person can allow it once, allow a narrow class of similar actions, or reject it.
Gates reduce the chance that a mistaken or malicious instruction causes damage. Use them for:
file writes outside the task's scope
package installation
network access
credentials
production systems
destructive commands
Make low-risk actions safe by default and keep prompts specific, so users are not trained to approve everything. Approval is not proof that the action is correct. It is a decision to permit a known impact within the larger review process.
Rewriting in plainer words…
This answer doesn't lend itself to a diagram - it reads best . No credits were charged.
Saved in this browser - sign in to keep your review list.
How should your speech become text?
Listening… your words appear above as you speak - tap Stop when you're done.
Recording · cr - tap Stop & transcribe when you're done.
Transcribing with AI…
Voice:
Keep going - a few more words and AI can grade it.
Interview lens
How interviewers actually use this question
Likely follow-ups
How do you keep users from approving every prompt without reading it?Make low-risk actions safe by default, keep prompts specific, and use narrow allow rules so prompts are rare enough that people read them.
What goes wrong if you turn all approvals off?Nothing stops a mistaken or injected instruction from acting. If you must, run it in an isolated sandbox with no credentials or production access.
How much does an approval gate protect you against prompt injection?It only helps if the prompt shows the real action and the person reads it. Pair it with a sandbox and least-privilege credentials.
The user approved the command and it ran. Why is the change still not trusted?Approval permits a known impact. It doesn't check correctness, so tests, diff review and the normal review process still have to check the result.
What you can say
An approval gate pauses the agent before an action that needs human permission, so a mistaken or malicious instruction is less likely to cause damage.
The prompt should show the exact command or operation, why the agent needs it, and what resources it can affect.
The person can allow it once, allow a narrow class of similar actions, or reject it.
I'd gate file writes outside the task's scope, package installs, network access, credentials, production systems and destructive commands.
Low-risk actions should be safe by default and the prompts specific, or people get trained to approve everything without reading.
One limit is that approval isn't proof the action is correct. It permits a known impact, and the wider review process still has to catch mistakes.
Weak answers to avoid
Treats approval as proof the action is rightApproval only permits a known impact. Say that tests, diff review and the normal review process still decide whether the change is correct.
Wants a prompt on every actionConstant prompts train people to approve without reading, which removes the protection. Make low-risk actions safe by default and gate the risky classes.
Can't say what the prompt should showA bare 'Allow this command?' gives no basis for a decision. A good prompt shows the exact command or operation, why it's needed and what it can affect.
Only offers allow or denyWithout a narrow 'allow similar actions' option, users either approve the same thing over and over or grant broad access. Mention allow once, allow a narrow class, reject.
Interview lens
Likely follow-ups, what you can say, and the weak answers to avoid.
