Too many approval prompts
Reduce repeated decisions without giving up a shared policy across agents.
A policy whose default behavior is Ask sends every unmatched action to you. That can produce many prompts at the beginning, before your lists and Policy Rules reflect your normal work.
Reduce repeated prompts gradually
- Leave Remember on when reviewing an approval.
- Add recurring, low-risk actions to the Allow List.
- Use the narrowest remembered pattern that covers the action you intend.
- Review the Allow List after a few sessions and remove rules that are too broad.
The policy is shared across supported agents, so a rule does not have to be recreated for Cursor, Claude Code, Codex, and every other integration.
Check which rule asked
Open an approval's Policy Rules section to see which rule matched. If it repeatedly catches harmless actions, refine the definition or add a non-matching example. You can also raise the threshold, with the tradeoff that it may miss actions you wanted caught. See Match threshold.
A notice that rules could not be checked is different from a rule match. Open Settings > Policy Rules > Rule Evaluator and check the connection, data-sharing confirmation, and AI-credit balance. Changing a rule's threshold will not fix an unavailable evaluator.
Allow ordinary work by default
If you do not want unmatched actions to require review, duplicate a policy with a strong Deny List and change its default behavior to Allow by Default. The Restrictive Block policy is a useful starting point because it includes a broad list of high-risk command patterns.
Enabled Policy Rules run after the Deny, Allow, and Ask Lists; anything left undecided will be allowed automatically. Review the lists and rules before enabling the policy, then test it with low-risk work. See Policies and default behavior for the complete workflow.