AI QA for the code your agents ship. Four signals run on every pull request and press one verdict onto the diff — izri/quality. Pass, and it merges. Fail, and your eye lands on exactly what stopped it.
Izri reads every file the diff touches and lines them up against what the PR set out to do. Changes that drift outside the stated intent get flagged before anything else runs.
Unit, integration, end-to-end — Izri runs the tests that matter to this diff and reads the results against it. Green only counts if the right tests actually ran.
A passing test means nothing if it never touches what changed. Izri traces coverage from each test back to the diff — and catches tests that look green but skip the new lines entirely.
For changes that touch the UI, Izri renders the routes this PR affects and diffs them against main — surfacing layout shifts the tests would never catch. A soft signal: it informs the verdict without hard-blocking.
izri/quality takes the worst hard rule and a composite of the soft signals, then presses a single verdict — recomputed on every push, large diffs and one-liners alike.
Pin izri/quality to branch protection and the merge button waits on it. Each signal posts its own child check — keep them informational, or pin them for surgical gates.
A single comment, edited in place on every push — hard findings, contributing signals, and AI insight in one block. It updates as you fix; it never stacks duplicate noise on the thread.
Every delta gets a report — line-level annotations, a visual-diff drill-down, and project quality over time. Watch the AI-generated code in your repo trend up, push by push.
Drop the GitHub Action in, point the CLI at your project, or mount the MCP server in your agent. The four signals score the next PR you ship.