DIY AI code review pipeline. Front loaded static analysis feeding multi-lens review w/ depth routing & BYO configuration.

Warning: this might be slop.

Install

bun add -g @noslop/cli

noslop runs on Bun and needs a signed-in Codex or Claude Code CLI on your PATH.

Supported model CLIs

• Codex: the default; each call runs read-only.
• Claude Code: each call runs with the read tools alone.

Features

• A static stage before any model: one parse finds what code can decide and routes each lens to its slice. See the static stage.
• Lenses by trigger: correctness on every change, every other lens only when the change calls for it. See routing and depth.
• Depth by risk: a small change gets a light review, and a change to a risk class a deep one. See routing and depth.
• Findings you can check: each finding anchors to the diff and carries a severity and a verdict. See findings.
• Your own rules: the taste lenses read your rules files and cite each finding to a quote from them. See your own rules.
• A check of open findings: the GitHub App posts inline comments and a check run that passes once no finding is open, so a relay or a branch rule can gate the merge on it. See the GitHub App.
• Measurement: a store of every finding and its answer, and an eval harness that replays a labelled corpus. See the findings store.

Configuration

A repo works with no config. A root review.json sets your rules files, stack patterns and risk classes, and can extend a shared config; noslop config prints the merged result. The reviewer config covers the merge, and the JSON Schema lists every key.

Documentation

The docs index lists the guides, the architecture docs and the decision records.