@lymo-inc/own-your-review
v0.2.4
Published
Stop merging things you don't understand.
Maintainers
Readme
own-your-review
Stop merging things you don't understand.
Most AI code review tools review code for you. This one makes sure you reviewed the code.
own-your-review generates comprehension questions from your PR diff that can only be answered if you actually read and understood the changes. Think of it as Gate 1 from ownyourcode — applied to every pull request.
The Problem
AI code generation has created a new failure mode: vibe merging. CI passes, the diff looks reasonable, the reviewer hits "Approve" — without genuinely understanding what changed.
Existing AI PR tools make this worse. They review code for you, further removing the human from the loop.
Quick Start
Option A: Claude Code Plugin (interactive quiz)
Install the plugin in Claude Code:
/plugin marketplace add lymo-inc/own-your-review
/plugin install own-your-review@own-your-reviewThen run on any branch with changes:
/own-your-review:quiz-meThis starts an interactive Socratic quiz — questions one at a time, with feedback on your answers and a comprehension score at the end. You can also target specific files or commits:
/own-your-review:quiz-me src/auth.ts
/own-your-review:quiz-me abc123..def456
/own-your-review:quiz-me --stagedOption B: GitHub Action (automated on PRs)
Set up with the CLI:
npx @lymo-inc/own-your-review initOr with Bun:
bunx @lymo-inc/own-your-review initOption C: Manual setup
- Create
.github/workflows/own-your-review.yml:
name: Own Your Review
on:
pull_request:
types: [opened, synchronize, ready_for_review]
pull_request_review:
types: [submitted]
concurrency:
group: own-your-review-${{ github.event.pull_request.number || github.event.issue.number }}
cancel-in-progress: false
jobs:
review:
runs-on: ubuntu-latest
permissions:
contents: read
pull-requests: write
issues: write
steps:
- uses: actions/checkout@v4
with:
fetch-depth: 0
- uses: lymo-inc/[email protected]
with:
anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}Configuration
Optionally create .github/own-your-review-config.yml:
questions:
max: 5 # Max questions per PR (1-7)
min: 2 # Min questions
reviewer_level: mid # junior | mid | senior
ignore:
paths:
- "*.lock"
- "*.generated.*"
- "migrations/"
authors:
- "dependabot[bot]"
on_unanswered:
learning_note: true # Post learning note if approved without engagingPR Labels
review:skip— Skip comprehension check for this PRreview:quick— Reduce to minimum questions
Question Categories
| Category | What it tests | When it's used | |----------|--------------|----------------| | Intent | Why this change exists | Every PR | | Mechanism | How the code works | New logic, algorithms | | Blast Radius | What else is affected | Type/API/shared code changes | | Edge Cases | Boundary thinking | New conditionals, error paths | | Trade-offs | Design decisions | Architecture choices | | Security | Trust boundaries | Auth, user input, data exposure |
License
MIT
Credits
- ownyourcode — the philosophy
- claude-code-action — the runtime
- Built by Lymo
