Agents made code cheap to write. Not cheap to maintain.
Agent code passes tests and looks fine in review. Weeks later it gets patched, rewritten, or reverted. You pay for the work twice, and nothing in your pipeline warned you. As coding agents produce more than your team can review, every change can't get the same attention. Rework Risk tells you where to spend it.
Tests
tell you whether code passes now.
Review
tells you whether code looks right now.
Rework Risk
predicts whether code will hold up.
How it works
Connect a repo
Read-only. Rework Risk finds merged changes that later needed corrective work, whoever or whatever wrote them.
See your backtest
The PRs you had to fix, which ones Rework Risk would have flagged using only what was known at merge time, and what that rework cost.
Score new changes
Every new PR gets a 30-day rework risk score. Your pipeline decides what to do with it: route to a senior reviewer, ask for more tests, or merge.
Outcomes flow back on their own
Reverts and fix PRs already live in your git history, so there is nothing extra to report. Each outcome makes the shared model sharper.
Rework Risk is an advisor, not a gate. It never blocks a merge and never sits in your model traffic. It works with Claude Code, Codex, Cursor, or the custom agents your team builds, and it scores human-written changes too.
What counts as rework
Rework Risk predicts unplanned corrective work attributable to a change within 30 days of merge. Churn alone isn't failure.
Counts
- Reverts
- Fix PRs or commits that repair the original change
- Incidents linked to the change
Doesn't count
- Planned iteration and follow-up work
- Refactors and requirement changes
- Unrelated edits to the same lines
An outcome is what happened (this PR was reverted within 30 days). A prediction is what Rework Risk expects (23% chance of corrective rework within 30 days). Every score is stated against that fixed window.
Why not build it yourself?
You can, from your own history. But corrective rework is rare and slow to observe, so no single team sees much of it. Rework Risk learns from outcomes across many organizations, codebases, and agent setups, and you keep your own tools.
Rework Risk is built on Hokusai, so it is owned by the people who improve it. When your outcome data measurably improves the model, Hokusai gives you a stake in it. How ownership works