ChatGPT in Slack: turn a bug thread into a pull request
This research library uses AI-assisted source research and drafting. Linked sources support product claims; analysis and proposed exercises are our interpretation. Unless an article documents a test and its results, do not read it as a hands-on review or an independently verified benchmark.
ChatGPT in Slack lets anyone in an enabled channel mention @ChatGPT to investigate a bug, and in workspaces with Cloud delegation it hands the repository work to a separate Codex Cloud task. OpenAI launched the new @ChatGPT app for Slack and Microsoft Teams on 29 September 2026 for Business and Enterprise workspaces. The Use ChatGPT in Slack guide describes the flow this article walks through: bug report in a thread, Codex task, reviewed pull request.
What ChatGPT in Slack does with repository work
A normal @ChatGPT request answers in the thread using the tools and connections your admin configured. Repository work is different. When Cloud delegation is enabled, ChatGPT sends it to a separate Codex Cloud task and posts the result back to the same thread.
Two accounts are involved. The Slack surface has a service account that finds eligible environments. The coding task itself runs as the person who asked. The docs state that the service account does not need GitHub access. Your own access to the environment and its repositories decides whether the task can start.
Teammates can add context in the thread without an individual ChatGPT license. That is useful for bug reports, because the person who saw the failure is often not the person who will fix it.
Microsoft Teams gets the same @ChatGPT app, but the Codex Cloud delegation steps in the docs are written for Slack. Check the Teams setup guide before you plan a Teams-first pilot for coding work.
What your admin must enable first
Your workspace admin handles most of this once. Use it as a checklist when the first request fails:
- @ChatGPT is installed in your Slack workspace and enabled for the channel.
- Codex Cloud is enabled for the ChatGPT workspace.
- Where Codex Cloud role-based access control applies, both the Slack surface's service account and your account have Codex Cloud access.
- A cloud environment for the repository is published and shared with the workspace.
- You can open that environment and any repositories that require authentication.
Personal cloud environments do not appear in this flow. If your team only has private environments today, publish and share one first. The Codex Cloud docs cover creating an environment: choose repositories, let Codex prepare dependencies and tools, review the checks, then select Publish.
Walk a bug report from thread to pull request
- Start in the thread where the bug was reported. Mention @ChatGPT, name the repository, and say what result you want. The docs use this example:
@ChatGPT investigate the failing checkout tests in acme/storefront and propose a fix. - Handle the prompts. If ChatGPT asks you to connect your account, use the link, connect Slack for the same workspace, then return to the thread and mention @ChatGPT again. If it shows a private approval card, review the assignment and environment choice, then choose Allow or Deny.
- Follow the task. ChatGPT can post when the task starts. Where available, Follow along opens the task so you can inspect its work, and Cancel stops it. Both are meant for the requester.
- Ask for the missing proof. Continue in the same thread, for example:
@ChatGPT add a regression test for the fix and summarize the checks you ran.Follow-ups from the same account continue the same coding task and environment. - Review before you merge. ChatGPT returns the result to the thread. Read the changes and the reported checks, then commit or open a pull request from the task when you are satisfied.
We teach the same rule for every coding agent, and it applies here. A thread message that says "fixed" is not evidence. The diff, the regression test and the command output are.
Write the request so Codex picks the right environment
ChatGPT chooses from accessible workspace-shared environments. Naming the repository or the shared environment in your message helps it choose. If you need a different environment, start a new conversation instead of redirecting an existing one.
Long bug threads also cause trouble. The docs suggest summarizing the relevant issue, the expected outcome and the repository in your latest request. A good request reads like a small ticket:
@ChatGPT in acme/storefront, checkout fails for carts with a discount code.
Expected: order total applies the discount once. Actual: applied twice.
Repro is in the first message of this thread. Find the cause, fix it,
add a regression test, and report the test command and its result.
Keep secrets out of Slack prompts. Repository work should use the environment's credential controls, not a token pasted into a channel. Anyone with access to the channel can read the reply, and a private approval card does not make the channel reply private.
When the coding task does not start
| Symptom | Likely cause | What to check |
|---|---|---|
| No response | App not enabled for the channel, or a Slack Connect channel | Mention @ChatGPT explicitly in an internal channel |
| Asked to connect again | Account connected to a different Slack workspace | Connect for the workspace where you asked, then retry in the thread |
| No eligible Cloud environment | Environment not published or not shared | Publish and share it; personal environments never appear |
| Coding task will not start | Cloud delegation or Codex Cloud access missing | Ask the admin to confirm access for you and the service account |
| Follow along and Cancel missing | Deployment configuration | These controls depend on setup and are intended for the requester |
Moving from @Codex to @ChatGPT
If your team used @Codex in Slack, start new requests with @ChatGPT after your admin enables the ChatGPT app. Existing Slack installations need a Slack admin to approve extra app permissions for the new experience. OpenAI says existing installations will not stop working immediately at launch. At launch, @ChatGPT does not use memory, so put repository conventions in the repo's AGENTS.md rather than expecting the Slack agent to remember them.
Our Delegate, Review, Own method fits this flow directly: Slack is where you delegate, the thread is where you review, and the engineer who merges owns the change. Try it on one small, reproducible bug in a repository that already has a published shared environment.