tuipr_

Terminal review workstation for agent-generated pull requests.

Coding agents open pull requests faster than anyone can read them. The bottleneck stopped being how fast code gets written, or even how review mechanics work — it is human attention, and the discipline to decide what a review is worth spending.

tuipr puts a human gate in front of the merge, in the terminal, where the rest of the work already happens.

What it does

• Queue. Every open PR with its computed status — landable, conflicted, blocked, draft — reconstructed from git and the GitHub API. Stacked PRs render under the base whose fate decides theirs.
• Review. Hand the terminal to a hunk-level diff viewer, comment inline, and upload the findings as a single review rather than a scatter of comments.
• Budgeted AI review. Run a coding agent over the selected PR in the background under an explicit spend cap. Findings arrive as notes inside the diff. Starting a second run takes a deliberate keystroke — accidental double spend should be hard, not merely regrettable.
• Gates and attestation. Merges pass independent gates. Approvals write an attestation into the GitHub audit trail, so the intent behind a merge outlives someone's shell history.

Design principles

• Fail closed. A silent success is a bug class, not a convenience.
• Spend is a decision, never a side effect.
• Provenance over convenience — what happened must be reconstructible from the audit trail alone.