WhatProblem WhatProblem
Home
Who it's for
For Maintainers For Product Teams For QA Teams For Managers For Developers For Every Issue
Guides How it works Blog

Commands

You have run into the bot on a GitHub issue and you want to tell it what to do. Two commands cover it.

The two commands

Comment either of these on the issue. They work anywhere in the comment, on a line of their own or in the middle of a sentence, and the case does not matter.

  • !whatproblem stop

    Leave this issue alone. The bot replies once to confirm it heard you, then posts nothing further on that issue.

  • !whatproblem analyze

    Read this issue now and ask what is missing. This also brings the bot back to an issue it has stopped on, or has already finished with.

If one comment contains both, stopping wins. Asking the bot to leave an issue alone should never be the thing that starts it up.

The match is a plain text search of the whole comment, so a command still counts when it sits inside something else you have written, including a quote of somebody else's comment.

Stopping applies to one issue

Stopping is per issue. It is not per repository and not per account, so !whatproblem stop on issue #12 changes nothing about #13. You have to say it on each issue you want it off.

To stop it everywhere, remove repositories from the installation, or uninstall it from Settings, Applications, WhatProblem AI. Both of those are done in GitHub's own settings by whoever administers the account, not from a comment.

Bringing it back

!whatproblem analyze on an issue the bot has stopped on, or has already summarised, starts a fresh pass. The round count goes back to zero and it picks the issue up as if for the first time, so a summary it posted too early can be corrected rather than being the last word.

The first time an issue is picked up, it costs one against the account's monthly allowance. Every pick-up after that on the same issue is free, however many times you bring the bot back and however many rounds of conversation follow. Replies and the closing summary are not charged either.

What it does without being asked

  • Reads a new issue as soon as it is opened, and asks what is missing.
  • If the issue already says enough, its first comment restates the problem back to you instead of asking questions.
  • Carries on for at most five rounds on one issue. After the fifth, the next comment gets the closing summary rather than another question.
  • Ends by posting a summary of the clarified problem. On an issue that was clear from the start, it ends with a short note instead of restating everything a second time.
  • Holds a thread open when a reply left the issue less clear than it was and questions are still outstanding, rather than declaring it done.

It only ever posts comments. It does not close your issue, label it, edit anything you wrote, or touch pull requests.

After a week of quiet

Nothing happens at the one week mark itself. There is no timer running against your issue and nothing gets labelled stale.

The next comment on the issue is what checks. If the thread has been quiet for more than a week by then, the bot treats the conversation as finished: it replies once to say so and to give you the way back, and then stays out of the thread.

A week is deliberately longer than it takes to answer a question properly. Fetching a log file, reproducing something, or simply coming back on Monday should never be the thing that ends the conversation.

The way back is the same as for any finished issue. A comment containing !whatproblem analyze is read as a request to carry on, so it revives the thread instead of being swallowed by the gap, and the issue is picked up fresh.

Where commands are not read

  • Pull requests. The bot never comments on them. A command in a pull request thread is dropped before anything looks at it.
  • Comments written by bots. If a CI robot or a coding agent posts one of these commands, it is not read.

Issues filed by bots are a separate case. The bot will not start on one by itself, because CI robots, linters and dependency bots are skipped so the allowance goes on issues people wrote, and the bot leaves a note explaining why, at most once a month per repository. A person can still call it onto one of those issues with !whatproblem analyze, and it will answer.

Older forms that still work

These were documented earlier and still work, because comments outlive documentation.

  • To pick an issue up: !whatproblem, !analyze, @whatproblem analyze, @whatproblem-ai analyze
  • To stop: @whatproblem stop, @bot stop, stop analyzing this issue

Use the ! forms. @whatproblem is an unrelated GitHub user's account. The bot reads the literal text of your comment rather than receiving a GitHub mention, so the @ forms reach it while also notifying a stranger who has nothing to do with any of this.

If this does not match what you are seeing

Tell us on the public tracker, which is the right place for anything about how the bot behaves. For anything tied to a specific account or installation, email support@whatproblem.dev instead, on any plan.

Product

For Maintainers For Product Teams For QA Teams For Engineering Managers For Developers For Every Issue

Resources

Guides Blog How it works Commands Plans and limits

Legal

Privacy Policy Terms of Service

Support

Give Feedback Contact Us

Copyright © 2025-2026 WhatProblem, a trading name of Williams Initiatives Ltd. All rights reserved. UK company number 14929985.