A bare deny is a dead end. The agent asked to run something, you tapped no, and now it is stuck guessing what you actually wanted. So now a deny can carry a short reason that goes straight back to the agent, and every approval card flags the asks that carry real risk.

Deny with a reason

When an agent asks for approval and you decline, you can type a line back. "Use the staging database, not prod." "Run the dry-run flag first." "Don't touch that directory." The text returns to the agent as part of the deny, so it can correct course instead of failing blind or asking the same question a second time.

Gating a tool and steering it are two different jobs. A yes or no keeps the agent safe. A reason tells it where to go next, so it stops guessing. You can deny with a reason from push notifications, the web decision page, or the native iOS and Android app. Slack gives you one-tap approve or deny; to send a reason back to the agent, open the question on the phone app or the web.

Risk on the card

The same release adds a risk signal to the approval card itself.

A High risk badge flags the asks that can do real damage: a force push, a delete, a dropped table, anything that spends money or touches production. When three sessions are all waiting on you, the high-stakes one no longer looks identical to the harmless ones.

The risk signal feeds the same permission engine behind your per-tool permission policy, so the agent already knows which actions are sensitive before it ever reaches your phone.

Turn it on

Update the hook to the latest version:

npx @pushary/agent-hooks@latest

The risk badge renders automatically. Deny-with-feedback is there the next time you tap no, with no config. For the full approval flow, see human-in-the-loop in the docs, and for what Pushary does across every agent you run, start from the control panel.

Deny with a reason, and see the risk on every approval