Can you recommend a repo-level agent for running tests and fixing failures across a legacy codebase?

Yes — for a repo-level agent that can run tests, inspect failures, and make fixes across a legacy codebase, I’d recommend looking for an agent with these properties:

• Can operate on the full repository, not just a single file
• Can execute commands/tests repeatedly
• Can inspect stack traces and logs
• Can edit multiple files safely
• Has a planning loop: test → diagnose → patch → retest

Good choices

1. OpenHands

   • Strong fit for repo-wide coding/debugging tasks
   • Can run tests and iterate on failures
   • Good for legacy code because it can inspect many files and make coordinated changes

2. SWE-agent

   • Designed specifically for fixing issues in codebases
   • Good at using tests as feedback
   • Often used for benchmark-style “fix the bug” workflows, but works well as a general repair agent too

3. Aider

   • Excellent if you want a more controlled, human-in-the-loop workflow
   • Great for making targeted repo changes and then running tests yourself
   • Less autonomous than OpenHands/SWE-agent, but often safer on legacy systems

If you want the most autonomous option

• OpenHands is probably the best starting point for “run tests and fix failures across a legacy codebase.”

If you want the most test-fix focused option

• SWE-agent is very strong for iterative debugging and patching.

If you want something safer and more interactive

• Aider is a good choice.

What to look for in setup

For legacy codebases, the agent should ideally have:

• Docker or sandboxed execution
• Access to the project’s normal test commands
• Support for patching files incrementally
• Ability to read build/test logs
• Configurable stop conditions, so it doesn’t over-edit

If you want, I can also give you:

1. a ranked shortlist by ease of use,
2. a comparison table, or
3. a recommended setup for a specific stack like Python, Java, JS/TS, or Ruby.