Runabout

Runabout API Settings
Runabout in Jira 1
Runabout in Jira 2

Inspiration

I got frustrated when spending lots of time reproducing bugs at work, so I made a tool to do it for me!

What it does

Runabout analyzes the information in a bug and generates code behind the scenes to reproduce the bug locally on your computer. It uses AI to determine if any classes or methods from your code are referenced in either a stack trace or error message. It then builds a context in natural language around the issue. Then when you use the Runabout plugin in the code Editor IntelliJ, Runabout will add real code examples to the context and use GPT-3.5 to generate code that will be used to run the specific method that caused the issue.

How we built it

The tools is made up of a Jira Forge app, a server, and an intellij plugin. The Jira app first takes information about the issue and sends it to the server for analysis. The server saves the generated context and passes an ID back to Jira. This ID can then be used to reference the issue within your IDE. The IDE plugin can then use the information from the server to actually run your code.

Challenges we ran into

Getting the AI to generate usable code was more difficult than I thought it would be! To generate valid code that references and calls into other methods from a project, the LLM needs a lot of context about the existing source. Managing this while being limited in how much context the LLMs allow makes for interesting problems w/ respect to prompt building.