Terminal review workstation for agent-generated pull requests.
Coding agents can open PRs faster than you can read them. The bottleneck is no longer writing code or even reviewing mechanics — it is human attention and spend discipline. tuipr is a TUI that puts a human gate in front of the merge: a PR queue with computed landability, hunk-level diff review, budgeted AI review runs, and approvals that leave an attestation in the audit trail.
Early extraction in progress. The tool exists and is in daily use in a private setting; it is being generalized and published here piece by piece. Not yet installable — watch the repo if you're interested.
- Queue — every open PR with its computed status (landable / conflicted / blocked / draft), reconstructed from git and the GitHub API, stacked PRs rendered under their base.
- Review — hand the terminal to a hunk-level diff viewer, comment inline, upload findings as a single GitHub review.
- AI review, budgeted — run a coding agent on the selected PR in the background with an explicit spend cap; findings land as notes in the diff viewer. Deliberate friction against accidental double spend.
- Gates — merge behind independent gates; approvals write an attestation body into the GitHub audit trail so intent survives outside shell history.
- Fail closed. Silent success is a bug class, not a convenience.
- Spend is a decision, not a side effect.
- Provenance over convenience: what happened must be reconstructible from the audit trail alone.
MIT