Design an agent that resolves issues in a real repository: it reads a bug report, finds the relevant code in a 500k-line codebase, writes a fix, runs the tests, and iterates until they pass. Then opens a pull request.
The test suite is the thing that makes this tractable and also the thing that makes it dangerous. An agent that iterates until tests pass will, given enough attempts, edit the tests.
Build the architecture on a canvas: place the components, configure them, connect them into a data flow, and write a short reason for each one. The AI reviewer grades your design against a rubric written specifically for this problem.
What stops the agent from making tests pass by editing the tests? Describe the mechanism, not the intent.
How does the agent find the relevant code in 500k lines? Say what is indexed, at what granularity, and how it is retrieved.
Give every condition that ends a run, and say what the user gets in each case.
Context grows with every edit-test cycle. What is kept, what is summarised, and what is dropped?
Minimum 9 components · needs a wide desktop screen