Toknary
Old code is where an agent saves the most time and where it can do the most harm. The difference is how you structure the work.
Where should I start?
With tests. If the code has none, ask the agent to write characterization tests: tests that record what the code does today, even when that behaviour looks odd. They are your safety net.
How big should each step be?
Small enough to review in a few minutes. Rename one module. Extract one function. Replace one deprecated call across the repository. Ask for a diff, not a rewrite, as in change the lines, do not reprint the file.
A safe loop
- Run the tests. They should pass before you start.
- Ask for one refactor and a short plan.
- Apply it, run the tests again.
- Commit. Then repeat.
Commit after every green step, as described in a safe Git workflow with an AI agent. If a step goes wrong, you lose minutes, not an afternoon.
What should I not hand over?
Decisions about the architecture itself. Let the agent do the mechanical changes while you decide the direction.
Install Toknary and try it on one small module first.