<img src="assets/banner.svg" alt="chutes" width="100%" /
chutes
Per-file triage for large refactors. A four-hundred-file migration arrives as one number. chutes turns it into three piles you can actually staff.
1,008 files classified in 20 seconds for 4.2 cents. Measured, not estimated — the run is written up in measurements/, including the part where it found something wrong with the tool.
HERO DEMO — not yet recorded. Capture it, drop the files in assets/, and replace this comment with:
<a href="assets/demo.mp4"<img src="assets/demo.gif" alt="chutes demo" width="100%" /</a
See assets/README.md for exactly what to record and the commands to do it. --
Pre-alpha, and not yet calibrated
Two warnings come before anything else, because they decide whether this tool is useful to you today.
This is pre-alpha. Commands, configuration keys and output formats will change without notice. There is no stable release.
The classification is not calibrated. chutes reports a confidence number with every decision, but nobody has yet measured whether that number is honest — whether the files it calls 90% confident are right about 90% of the time. Treat every result as a suggestion to a human reviewer. Do not wire it into anything that acts unsupervised.
The number that isn't a plan
A large refactor arrives as one number: four hundred files. That number tells you nothing you can plan with. Some of those files are a find-and-replace. Some need somebody who remembers why the code is shaped that way. A few should not be migrated at all — they should be redesigned.
Reading four hundred files to find out which is which costs more than doing some of the work. So the usual answer is to not find out: assign the whole thing to whoever is free, and discover the hard files in code review, weeks later, when the plan has already been committed to.
Three Lanes
| Lane | Meaning | Who does it | | --- | --- | --- | | Mechanical | The change is deterministic. | A codemod, or a junior engineer with the pattern in front of them. | | Judgment | The change is clear, the context is not. | Someone who knows why this code exists. | | Redesign | The file should not be migrated as it stands. | A designer of the new thing. Migrating it faithfully preserves the problem. |
Files that match no rule are Untouched — the Migration does not reach them.
The output is a Plan: plan.jsonl for machines, PLAN.md for people. Both are committed to version control, so the triage is reviewable in a pull request and diffable as the work lands.