Give code review
a repeatable
finish line.
easy-code-review turns coverage goals and review rules into focused code improvements. It runs your coding agent through a repeatable pipeline that measures results, feeds failures back into the next attempt and saves a detailed report of every run.
Currently a workspace package in the react-global-state-hooks repository. Try the beta from source.
A result you can inspect.
Coverage from your test runner
Configured executable validation
Agent findings and rubric
Per-run report and file changes
.review/runs/Measure the starting point.
Your test runner measures coverage before the agent changes the target. A baseline makes the next result comparable, and the built-in test-coverage rule targets the files below your goal.
Measure the result.
Keep the judgment separate.
A concrete target
A file, folder, glob, Nx project, commit or your current changes. Give the agent a defined review task.
Bounded iteration
Use coverage goals and retry limits. Concrete failures can trigger another pass; the loop stops on repeated failures, stalled progress or your attempt limit.
A traceable report
Inspect the run output and the diff under .review/runs/. Test coverage and an agent’s rubric score are different kinds of evidence.
Make the checks executable.
01. Prepare the project
Node.js, installed workspace dependencies and a configured test runner (this repository uses Vitest). A review.config.json at the workspace root points at a folder with settings.ts and optional rules/. Build the code-review package before the first run.
02. Choose your agent
One installed, signed-in agent CLI: Claude, Codex, Kiro or Copilot. The pipeline uses a fast model for scoring and a capable model for edits. Runs use that provider’s own quota or billing.
yarn install
yarn workspace code-review build
# Review the playground with a coverage goal.
yarn review playground --goal 85
# Create a review rule.
yarn review rule createProject settings and rules.
Use the existing review.config.json and settings module. The built-in test-coverage rule measures coverage and scores test quality; prompt-based rules add your own criteria. Check the package README for configuration and CLI updates when upgrading a beta release.
One record per run.
Inspect reports under .review/runs/, along with the changed files: per-file results, an event log, usage and every prompt and response. Keep actual measurements separate from agent-generated assessments.
An agent is still writing to your workspace.
Clean baseline
Start from a clean working tree so new changes are attributable, and inspect them with git diff before keeping them.
Explicit permissions
Choose the agent’s edit permissions. Kiro and Copilot adapters use workspace-wide write access; folder-level restrictions are unavailable through those adapters. Claude also reports blocked edits directly to the pipeline.
Human inspection
Validate the output and examine the complete diff before accepting the result. Model scores complement your final code review; they do not replace it.
Share a review workflow, request a rule or report an issue in the repository’s issue tracker.
Repeat the process.
Inspect the evidence.
Build a review loop around checks your project can actually run.