Refactor a large repo with parallel agent workers

One developer had an agent break a huge cleanup into many parallel tasks and merge a PR that cut non-test Python by 34.4%.

Done with Hermes Agentยท Advanced

What they did
A Nous Research developer set Hermes a standing objective with /goal and let it use skills it had accumulated over time. The agent divided the repository into 36 separate work areas, then sent out 1,393 subagents with their own git worktrees to make changes, verify each step, and compare interfaces with the original code. When one token expired after about fifty minutes, another session prepared the handoff and the run continued.
How it went
After about 19 hours, the merged pull request reduced non-test Python from 1,063,826 lines to 698,363, and gateway/run.py went from 34,847 lines to 5,512. Reviewers still found regressions the tests did not catch, and six files remained above 5,000 lines. The main run cost about $19,300, or roughly $25,000 including follow-up work, not counting human review.
Tips
This kind of cleanup needs tight review, because tests can miss behavior changes. It also helps to plan for long runs that may need a handoff if a session ends early.

Try it yourself with Hermes Agent

Review [repo path or name] for dead code, repeated logic, and oversized modules. Break the work into independent chunks, assign them across separate worktrees or branches, and after each change run [test command] and compare the public interfaces with the current codebase. Before any irreversible step like merging, deleting files, or submitting a pull request, stop and ask me to approve the exact change list and summary. When you finish, give me a concise report with files changed, lines removed, tests run, and any risks you found.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on nousresearch.com โ†—

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.