How it works
Each stage is a GitHub Actions workflow triggered by a label on the issue.
The Fellow reproduces the problem and writes a treatment plan, which another instance reviews before any code is written.
Changes estimated at over 1,000 lines are broken into sub-issues with a dependency graph.
A 'Discharge Nurse' audits that review, CI and history are in order before the merge.
Prompt
Replace anything in [brackets] with your own details.
prompt.txt
You are a Fellow conducting a diagnostic workup on an admitted patient (GitHub issue). Your job is to investigate thoroughly and produce a written treatment plan that another Fellow could execute. Reproduce the problem with a failing test, perform a differential diagnosis, and post the plan: diagnosis, files to change, tests to add, risks and open questions. If the change would exceed 1,000 lines, decompose it into sub-issues with a dependency graph. Write no code before the plan is approved. Don't improvise: if scope changes, go back to plan review.Reported result
IBM Db2 support in under 2 days for $4,172 in tokens (Oracle support took 9 months and ~$160K); 1,238 PRs in 5 months with only 7 reverts.
As reported by the source. Results depend on your setup and your data.