Short summary
Agent merge stops when CI is still running outside the merge UX
Affected version or release
Current app version / production build on macOS
Installation context
N/A
What happened?
When I ask the agent to merge a PR from chat, it uses the agent merge skill. If required checks are still running, the skill observes that CI is pending and then stops instead of continuing to wait/poll.
This makes the flow unreliable outside of the dedicated merge UX, because there is no ongoing “tick” to keep the merge attempt alive. The user has to notice that the agent stopped and manually prompt it to continue.
I saw this pattern in my recent session history multiple times: agent merge was invoked, checks were still running, and follow-up “continue” prompts were needed to make progress. I’m intentionally not including org, repo, PR, or transcript details here.
Steps to reproduce
- Open a PR with required checks that are still running.
- Ask the agent from chat to merge the PR.
- Let the agent invoke the agent merge skill.
- Observe that the agent sees CI/checks are still running and stops.
- Prompt the agent to continue.
- Observe that the merge flow can proceed only after the manual continuation.
Expected behavior
Agent merge should keep waiting/polling until the PR becomes mergeable, fails definitively, or needs explicit user action.
If the chat-based agent merge path cannot support that lifecycle, it should hand off to the same underlying UX-backed merge behavior that has a tick/polling mechanism, rather than stopping after a single pending-check observation.
Short summary
Agent merge stops when CI is still running outside the merge UX
Affected version or release
Current app version / production build on macOS
Installation context
N/A
What happened?
When I ask the agent to merge a PR from chat, it uses the agent merge skill. If required checks are still running, the skill observes that CI is pending and then stops instead of continuing to wait/poll.
This makes the flow unreliable outside of the dedicated merge UX, because there is no ongoing “tick” to keep the merge attempt alive. The user has to notice that the agent stopped and manually prompt it to continue.
I saw this pattern in my recent session history multiple times: agent merge was invoked, checks were still running, and follow-up “continue” prompts were needed to make progress. I’m intentionally not including org, repo, PR, or transcript details here.
Steps to reproduce
Expected behavior
Agent merge should keep waiting/polling until the PR becomes mergeable, fails definitively, or needs explicit user action.
If the chat-based agent merge path cannot support that lifecycle, it should hand off to the same underlying UX-backed merge behavior that has a tick/polling mechanism, rather than stopping after a single pending-check observation.