- Home
- Skills
- Code Quality & Review
- Advanced Code Review
More screenshots
Works with the AI Agents you already use
Advanced Code Review
Evidence-based code review with severity-graded findings, a health score, and a clear merge verdict — plus working tools that scope diffs, scan for risky patterns and leaked secrets, and run your repo's own checks.
$5
Advanced Code Review
Example session with this skill installed
Review this change before I merge it. It adds three export endpoints to our Node service (about 30 lines): getExport downloads a stored export by id, listExports returns all exports, and createExport writes one. No tests yet.
- Read your context and instructions
- Compiled the advanced code review
Verdict: Request changes. Health score: 49/100.
[Blocker · Confirmed] exports.js:8 — the export id is interpolated directly into the SQL query. Trigger: request the endpoint with id set to 0 OR 1=1. The handler builds the query with a template string using the raw query parameter, so the condition can be rewritten by any caller. Impact: full read of the exports table, and depending on driver settings, write access beyond it. Fix: parameterize the query and validate that id is an integer at the boundary. Test: a request with that id value must return 404 or 400, never rows.
[Critical · Confirmed] exports.js:8 — the lookup is not scoped to the caller's tenant. The query filters by export id only; nothing in the handler ties the record to the signed-in tenant. Trigger: sign in as tenant B and request tenant A's export id. Fix: add the tenant from the authenticated session (never from the request) to the query conditions. Test: a two-tenant fixture where B requesting A's id returns 404.
[Major · Confirmed] exports.js:16 — listExports loads the entire table with no limit. Every row is read and serialized on each call. Impact: memory and response size grow without bound as export history grows; the endpoint becomes a self-inflicted outage for the oldest accounts. Fix: paginate with a capped page size and add an index on tenant and creation date. Test: seed 10,000 rows and confirm the response holds at most one page within the latency budget.
[Minor · Confirmed] exports.js:24 — createExport reports success when the write failed. The error handler logs the failure and falls through to the success response, so callers believe an export exists when nothing was written and upstream retry logic never fires. Fix: return a 502 with a stable error code. Test: simulate a storage failure and expect the 502.
Verification: the file parses cleanly; the secret scan found nothing. Not verified: runtime behavior — no test suite exists for these endpoints yet, which is itself part of finding 3's fix.
Connects securely to your tools. The creator never sees your data.
About this skill
Most agent code review is a list of opinions. This skill runs a fixed protocol instead: scope the diff and triage its risk, reconstruct what the change is supposed to do, read the tests first, then work through ten axes — correctness, security, architecture, reliability, performance, concurrency, API and data compatibility, testing, readability, and delivery. Every finding follows one format: location, the concrete trigger that exposes it, evidence, impact, the fix, and the test that proves it. Findings are graded by severity (Blocker to Nit) and confidence, verdicts follow written rules instead of mood, and a weighted health score summarizes the change.
It ships with working tools, not just instructions: a diff scoper that flags auth, money, SQL, and migration surfaces plus files changed without tests; a pattern scanner (Python AST plus multi-language rules) that emits findings your agent then verifies; a secret scanner with 20+ detectors; a runner for your repo's own test and lint gates; and a report builder. The scanners are measured, not asserted: the package includes an eval harness with seeded-defect fixtures (11 of 11 planted defects detected, zero false positives on clean fixtures at Major severity or above).
Nine review modes scale the depth to the risk: Quick, Standard PR, Deep, Security-Critical, AI-Code Audit (with a dedicated catalogue of how generated code fails), Architecture, Release Gate, Hotfix, and Legacy/Refactor. Includes 18 reference guides, templates, and a worked example review.
The pattern scanner catches pattern-class defects (injection shapes, unsafe deserialization, swallowed errors, leaked secrets); it cannot see semantic bugs like a missing tenant check or wrong business logic — those come from the review protocol working on the diff and its context, so the skill still needs the change, its description, and its tests to review well. The bundled tools need Python 3 and git on the machine. The skill does not merge or modify your code; it produces the review, the findings, and the verdict.
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 any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 25 days ago
Requires