AI agents for bug fixing

Reproduce a reported bug, fix it and open a pull request.

13 agents, each with its inputs, outputs and review level checked by a person.

Bug fixing agents take a reported issue, reproduce it, find the cause, write a fix and tests, and open a pull request. They are coding agents pointed at the backlog rather than at new features, and they work best on issues with clear reproduction steps. The listings record what each agent needs (an issue tracker or a description, a repository), what it returns and whether a person reviews the fix before it is merged.

Choose by how your issues arrive. Agents that connect to an issue tracker can pick up tickets as they are labelled; agents in the terminal take a description from you. In both cases, the review level shows whether the fix waits for a person.

The agents

Groups that share agents with this one.

Questions

Can the agent fix a bug it cannot reproduce?

It will try from the description and the code, but the better agents say when they could not reproduce the issue. The acceptance section of each listing shows what is checked.

Does it write a regression test?

Several agents add a test that fails before the fix and passes after; the workflow section of each listing says whether the vendor documents that.