HeyBobHand a bug to Cursor without leaving the conversation.
Connect Cursor and Bob turns a reported bug into a running Cursor agent, checks on it, steers it when it drifts, and brings the pull request back to where the bug was raised. In the HeyBob app on Mac, Windows and Linux, or in Slack.
- Asks before every launch
- Nothing merges on its own
- A receipt on every job
What Bob can do with Cursor.
Cursor’s cloud agents are coding agents that work on your repositories in Cursor’s own environment, and Cursor names fixing bugs as one of their main jobs. Bob is the one who briefs them and follows up.
Check on agents
List your Cursor agents, see whether one is running, stuck, failed or done, and read what it tried.
Launch an agent
He writes the brief from the bug, picks the repository and branch, and can ask Cursor to open a pull request.
Send a follow-up
When the first attempt goes the wrong way, he tells the agent what it missed instead of starting over.
Five things to hand him.
Give him the bug, the place to work and the test that proves the fix.
“Use the steps and error in this thread to launch a Cursor agent on acme/web from main. Fix the checkout total bug, add a test for it, and open a pull request. Leave the payment code alone.”
He shows you the launch first. After your yes, Cursor does the work.
“Launch two Cursor agents on this timing bug with different approaches. Each adds a test and opens its own pull request so we can compare.”
Two tries cost more Cursor usage, so each launch is its own yes.
“Check the Cursor agent for ENG-482. If it changed how other code calls this, tell it to keep that the same and fix the bug underneath.”
He reads what it did before he tells it anything.
“Find the three Linear bugs tagged agent-ready. Gather what done looks like for each and prepare a Cursor launch. Ask me before launching any.”
Linear has the problem, GitHub has the code, Cursor does the fix.
“If the agent on the checkout bug is done, sum up the files it changed, the tests it ran, what could still go wrong, and link the pull request here.”
The reviewer gets the evidence, not “the agent finished.”
Connected in a few minutes.
Make a key in your Cursor dashboard, under integrations.
Tell Bob to connect Cursor.
Open the Connect Cursor button he posts and paste the key on that secure page.
Come back and ask again. The job picks up from there.
The key never goes in a chat message. Cursor is one connection for the whole workspace today, so use a key from a team account, and give Cursor access only to the repositories it should change.
Bob briefs the agent. He does not drive your editor.
- No remote control. Bob launches a separate cloud agent. He cannot see or steer a Cursor window open on someone’s laptop.
- No automatic merge. An agent can open a pull request. Your reviews and checks still decide what goes in.
- Checks are free to run. Listing agents and reading status happen on their own. Launches and follow-ups ask first.
- Or he fixes it himself. With GitHub connected, Bob can write and test the fix himself and open the pull request for review, as in this bug pass.
Usage reports too.
The same connection can read Cursor’s team numbers: members, daily usage, spending and how much AI-written code was accepted. Ask for a weekly adoption brief or a spending alert, and join it with GitHub delivery. Treat the numbers as signals about the team, not a ranking of people. More requests do not mean better engineering.
Common questions.
What can Bob do with Cursor?
Launch Cursor cloud agents, list them, check their status, read what they did, send follow-up instructions, and report on Cursor usage and spending for the team.
Can Bob use a Cursor agent to fix a bug?
Yes. Give him the bug, the repository and the branch to start from. He launches a Cursor agent with that brief and can ask it to open a pull request. He asks before every launch.
How does Bob connect to Cursor?
You make a key in the Cursor dashboard and paste it on the secure page Bob links. It never goes in a chat message.
Does Bob take over my open Cursor editor?
No. He launches a separate cloud agent on a connected GitHub repository. He does not see or steer anyone’s editor.
Does Bob change anything in Cursor without asking?
Checking agents and reading reports happen on their own. Launching an agent, sending a follow-up or changing a setting asks first unless your team decides otherwise.
Sources.
What Cursor’s cloud agents are for comes from Cursor’s own page.