Feature request for product/service
– Other –
Describe the request
Feature request for: Other — Grok Bot
Related (not the same request):
That thread is about waking a Bot from outside. This request is about what Auto Review does after that wake.
Problem
A Grok Bot routine can start from a signed webhook. That delivery is the user’s request. After the Bot composes a result, it often needs one known follow-up, such as an HTTP POST to an automation URL.
Auto Review still judges that follow-up against recent typed chat in the Bot thread. If the last typed turns were about a different topic, the outbound call is blocked and an approval card appears. Standing routine instructions do not count as intent. The webhook that started the run also does not count.
That makes unattended “external event → Bot works → one declared side effect” unreliable. Email in / email out is one case. Slack, GitHub, and other webhook wakes have the same shape.
Request
On a routine whose trigger is a webhook:
- Let me declare an allowlist of follow-up actions for runs started by that trigger only (exact URL and method, or exact tool name).
- When that routine is started by a verified webhook, treat the delivery as user intent for those declared actions. Do not require a recent typed chat turn that mentions the target or subject.
- Auto Review may still block anything not on that list. Require Approval rules still win.
This is not “never ask before sending.” It is “the signed trigger I configured is the approval for the actions I attached to that trigger.”
Success test
Start the routine only by webhook. Do not type in the chat. The Bot performs only a predeclared follow-up. No approval card. A different URL or tool still requires approval.
Operating System (if it applies)
Windows 10/11