GitHub CLI Event Streams Change How Agents Wait for Pull Requests
A practical guide to interpreting pull-request event streams without confusing one check with release readiness.
TL;DR: GitHub CLI can now wait for pull-request events instead of repeatedly polling. A completed check-suite event is a useful wake-up, not a release verdict: a pull request can have several suites, separate commit statuses and a newer head. After the listener opens and after an event or interruption, inspect the current pull request and the checks required by your team's policy before deciding what to do.
Your agent has opened a pull request. One test suite finishes successfully, and the listener returns an event. Is the pull request ready for review or merge?
Not necessarily. The successful suite may be only one part of the required checks. A separate commit status could still be pending. A new commit may have replaced the head you were checking. The event tells you something changed; it is a reason to look, not a substitute for the current state.
On November 18, GitHub announced Event Streams support in GitHub CLI 2.105.0. The gh event listen command lets a person or agent wait for selected pull-request changes and receive events as JSON. It is a practical improvement for a waiting workflow, with important limits for anyone using its output to make a QA or release decision.
This is source-backed analysis, not an executed experiment. We did not run the command, access the preview, or test an integration with Samelogic.
What the New Command Does
According to GitHub's Event Streams documentation, gh event listen creates or reuses a subscription for the selected pull request and event types, opens the stream and prints each event as a line of JSON. By default, it exits after the first event. --count waits for a specified number, --follow continues listening and --timeout sets a waiting limit.
The supported pull-request events include submitted reviews, changes to the head or pull-request state, merge conflicts and completed check suites. A check-suite event includes that suite's conclusion. For most other event types, GitHub says the event signals a change but the client must fetch the referenced current state to understand it.
The commands are in preview. Event Streams is limited to GitHub.com, access is being rolled out through a waitlist and the API, event types and limits can change. A team should confirm its account has access and that its token has the necessary pull-request read access before depending on this workflow.
One Green Suite Is Not the Whole Pull Request
Imagine a pull request at head abc123 with two required check suites and one required commit status. The first suite reports success. The second suite is still running, and the commit status has not returned. The event is real, but the required work is not complete.
Now imagine a developer pushes a new commit while an agent is waiting. A success associated with the earlier head is not, on its own, a decision about the new head. Read the current head and inspect the applicable checks and statuses again.
GitHub explicitly documents two important boundaries: a pull request can have multiple check suites, and check-suite events do not cover commit statuses created through the REST API. There is no single Event Streams event meaning all checks have completed. Fetch the current suites after relevant events and retrieve any required commit statuses separately.
Even a complete set of passing technical checks is not automatically permission to merge. Apply your repository's actual review, branch-protection or ruleset requirements and your team's release policy. The recommended decision is about the whole current pull request, not an agent's most recent notification.
A Decision Card for a Waiting Agent
Use this card as a design aid, not as a GitHub configuration or a workflow we tested.
Attach and establish a baseline. Wait until the CLI reports that it is listening, then fetch the current pull-request state. Record the current head and the review and check requirements relevant to your repository. The default listener can receive an event and exit during this read, so reconcile the current state and check whether the listener is still active before relying on another wake-up.
Treat an event as a wake-up. Identify what changed, then fetch the current state. For a suite event, inspect the suite conclusion and the other checks required for the current head; fetch required commit statuses separately.
Recheck the head. If the head changed, evaluate the new head rather than carrying forward the old result as the final answer.
Apply the actual review policy. Check whether required reviews and any other repository rules are satisfied. A comment, approval or check event alone is not the whole decision.
Return a bounded outcome. Report ready for the specific next step only when the applicable conditions are confirmed. Otherwise report what is pending, failed or unknown and who should look next.
The agent should record the observation time and the pull request and head it inspected. If it cannot read a required signal, it should say so rather than turn missing data into a pass. Teams should also consider whether the state can change between the final read and any permitted action.
Plan for Gaps and Repeated Events
The stream only delivers events while it is open. GitHub documents a short window after subscription creation, periods before the stream opens and disconnections as potential gaps, and says there is no replay. Once the listener is open, read current pull-request state rather than assuming the stream supplied the full history.
Events can also arrive more than once or out of order. Do not infer the current state from arrival order alone. Fetch current state when processing an event; use identifiers and the referenced resource where appropriate to handle duplicates. After a disconnect, reconnect as GitHub describes and reconcile current state again. A timeout or lost connection means observation is incomplete, not that tests passed or failed.
Subscriptions expire after 24 hours. GitHub says the CLI reconnects on a dropped stream and recreates an expired subscription, but the listening command does not delete its subscription when it exits. Teams should understand ownership and cleanup for their use, especially in automated jobs. Consult the current documentation for token and error handling rather than assuming every authentication or preview-access error is transient.
Design the Agent's Next Action Around Current State
As agents wait on pull requests, an event should wake an investigation rather than stand in for the complete release decision. Teams can define the current head, required checks, commit statuses and review policy that a human or agent must recheck.
A useful result is a short, inspectable record: the head examined, the checks and reviews required by policy, the latest observed status, anything unknown and the next owner. Keep the permission to act separate from the ability to observe. For example, an agent allowed to read a pull request should not be assumed to have permission to merge it.
Where a Browser Recording Fits
A failing check may lead a person to investigate a separate visible browser issue. Samelogic's workflow lets someone deliberately start a recording of the browser steps and inspect the affected path. That context is supplementary: a recording does not establish that a GitHub check passed or that a pull request is ready. In a November 6 fake-data QA run using extension 1.10.1, the first action was incorrectly assigned the later action's failure result. Full playback was not verified, and the source build identity was unknown. Do not rely on the recording alone for a merge decision.
Explore the browser bug reproduction workflow and decide what browser context is useful alongside the current pull-request state.
Sources
Related workflows
Move from editorial context into the selector, Playwright, and bug-reproduction pages that turn exact UI evidence into action.

