Attention map & reviewer routingLive
Three hunks that need human eyes, and who should look.
AI review should make human review sharper, not replace it. Forseti marks up to three hunks where risk, uncertainty and new code meet, says why, and suggests two people to read them.
PR #518 · summary
Example
Needs human eyes
- 1src/billing/refund.ts L41–58New retry path around a money movement; no test touches it.
- 2src/auth/session.ts L12–19Changes token expiry; the finder was unsure of intent.
- 3db/migrations/0042.sql L1–9Adds an index without CONCURRENTLY on a large table.
Suggested reviewers
PSPriya ShahCODEOWNERS · src/billing/
TRTom ReidLast 3 commits to session.ts
Request review
How it works
Step by step.
The hunks that need human eyes, and who should look.
- 01Score hunksHigh risk × uncertainty × novel code. The top three, at most, are marked with a reason.
- 02Find ownersCODEOWNERS on the base branch, plus the most recent authors of the touched files.
- 03FilterThe PR author and bots are excluded. Two people are suggested.
- 04RequestOne click in the dashboard asks them on GitHub. The request is audited.
In detail
What it does, exactly.
- It points humans at the riskiest few minutes of reading, instead of adding more comments.
- Suggestions come from your own ownership rules and history.
- On GitHub
- Sticky summary: attention map + suggested reviewers
- In the dashboard
- PR detail, request review
In the summary
The attention map and suggested reviewers sit in the sticky PR summary on GitHub.
In the dashboard
The same map on PR detail, with a request-review button that uses GitHub's own reviewers API.
A reason for each hunk
Never just a highlight: each mark says what makes it worth a human look.
Uses what you have
No new ownership file. CODEOWNERS and git history are enough.
Put Forseti on your next pull request.
Connect GitHub in four clicks. Your first review starts by itself on your latest small open PR, priced before it runs.