- Home
- Skills
- Code Quality & Review
- PR Review Pro: Verified, Severity-Ranked Code Review
Works with the AI Agents you already use
PR Review Pro: Verified, Severity-Ranked Code Review
Point it at a PR number, URL, branch or your uncommitted changes and get a merge verdict in minutes.
Free
PR Review Pro: Verified, Severity-Ranked Code Review
Example session with this skill installed
Review PR #482 before I merge it. Focus on real bugs and security, skip style nitpicks.
- Read your context and instructions
- Compiled the pr review pro
Review: #482 feat/checkout-v2 — 3 findings (1 blocking)
Read 9 changed files (+412 / −88), their callers and 4 related test files. PR goal: "support retrying failed payments without losing the cart". 7 candidate issues found, 4 dropped after verification.
Verdict: ❌ Request changes
🔴 Blocking
- Discount applied twice on payment retry —
src/checkout/apply.ts:88(CONFIRMED)
Scenario: payment fails →retryPayment()callsfinalizeOrder()again →applyDiscount()runs on the already discounted total. Reproduced with the existingcheckout.test.tsfixture: 10% coupon yields −19%.
Fix
- total = applyDiscount(total, order.couponCode)
+ if (!order.discountApplied) {
+ total = applyDiscount(total, order.couponCode)
+ order.discountApplied = true
+ }
🟡 Should fix
- N+1 query in order history —
src/orders/list.ts:31(PLAUSIBLE)
Scenario:itemsloaded per order insidemap(); a customer with 200 orders triggers 201 queries.
Fix:prisma.order.findMany({ where, include: { items: true } })
🔵 Consider
- No test for expired coupon on retry —
src/checkout/apply.test.ts
The new retry path is untested when the coupon expires between attempts. Add a case asserting the full price is charged.
✔️ Looks good
- Idempotency key on the payment provider call prevents double charges.
- Clear separation between
retryPaymentandcreatePayment.
Next step: fix #1 (want me to apply it and re-run the checkout tests?).
Connects securely to your tools. The creator never sees your data.
About this skill
Most AI code reviews drown you in style nitpicks and false positives. This skill reviews like a careful senior engineer: few findings, each one verified, each one actionable.
Point it at a PR number, URL, branch or your uncommitted changes and get a merge verdict in minutes.
What it does
- Collects the right diff automatically:
gh pr diff, branch vs. default branch (merge-base), or staged/unstaged work. Reads the PR description and commit log to check the code against its stated goal. - Reads real context: the full surrounding functions, callers of changed signatures and the tests that cover the code, not just the hunk.
- Six focused review passes: correctness, security, data/migrations, API contracts, tests and performance.
- Verifies before reporting: every candidate finding is challenged (guards, callers, running tests or a quick script). Anything without a concrete failure scenario is dropped. Survivors are tagged CONFIRMED or PLAUSIBLE.
- Actionable output: severity-ranked (🔴 Blocking / 🟡 Should fix / 🔵 Consider), each with
file:line, failure scenario and a diff-style fix, plus an overall verdict. - Safe by default: read-only. It only edits files if you ask it to apply fixes, then re-runs tests.
What it catches
| Pass | Examples |
|---|---|
| Correctness | off-by-one, null paths, async/await misuse, race conditions |
| Security | SQL/shell/HTML injection, missing authz, secrets, SSRF, path traversal |
| Data | destructive migrations, N+1 queries, missing indexes, timezone bugs |
| API | breaking signatures, response shape or env var changes |
| Tests | changed behavior without tests, tests that cannot fail |
| Performance | work in hot loops, unbounded memory, sync I/O on request path |
Best for
- Self-review before opening a PR
- Second reviewer on teammates' PRs
- Solo developers without a reviewer
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Code Quality & Review.
- Very large diffs (thousands of lines) are reviewed by priority files first; the report says what was not covered.
- Findings marked PLAUSIBLE were not executed (e.g. no runnable tests or services); verify them before acting.
- Focuses on correctness, security and risk; it intentionally skips formatting and style unless you ask.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
This skill has been used 10 times.
Trust & safety
Security scan
Security scan passed 11 days ago
- Free with an Agensi account
- Software
Git
- Optional
GitHub CLI
GitHub CLI is for reviewing PRs by number or URL.
Permissions
Read Files: reads the diff, surrounding code, callers and tests. Terminal: runs git/gh to collect the diff and optionally your test suite to verify findings. Write Files: only when you explicitly ask it to apply fixes. No network calls beyond gh fetching your own PR.