Four AI engineers argue about your code

Senior, security, performance, and product agents — each with a conflicting objective — debate your GitHub repo through structured proposal → critique → revision → consensus rounds, then hand you a prioritized engineering report.

See sample report

Public GitHub repos work instantly. Private repo support is coming later.

What do you want checked?

Choose a review type. You can run multiple later.

What you'll get

A structured report with severity, evidence, and actionable fixes.

Readiness

Medium risk

Findings

7

Critical

1

High

2

Medium

4

Criticalsrc/app/api/sessions/[sessionId]/advance/route.ts

Long-running review blocks request lifecycle

Why it matters: The orchestrator runs synchronously inside the HTTP handler. If it takes >30s the request times out, leaving the review in a broken state.

Suggested fix: Move orchestration to a background job queue and return 202 Accepted immediately.

Highsrc/app/api/gate/route.ts

Demo password gate uses weak access control

Why it matters: A single plaintext password in an env var protects the entire app. No rate limiting, no lockout, no session expiry.

Suggested fix: Replace with proper auth (OAuth or magic link) or add rate limiting + constant-time comparison at minimum.

Medium/ (root)

Missing CI build/test pipeline

Why it matters: No GitHub Actions or CI config found. Tests exist but never run automatically, so regressions ship unchecked.

Suggested fix: Add a CI workflow that runs lint, type-check, and vitest on every push and PR.

How it works

1

Paste repo

Drop a GitHub URL. Public repos work instantly.

2

Four engineers debate

Senior, security, performance, and product agents propose, critique each other by maximum objective conflict, then revise until consensus.

3

Get a prioritized fix plan

Receive a structured report with severity, evidence, and fixes.

Why not just ask a chatbot?

A single prompt gives you a wall of text. This gives you an engineering report.

Adversarial by design

Critiques are routed by maximum objective conflict — Senior ↔ Performance, Security ↔ Product. Agents can't rubber-stamp each other.

Structured findings

Findings are categorized by severity with file paths, not buried in paragraphs.

File-level evidence

Every finding points to the exact file and line, so you know where to look.

Copy-paste fixes

Suggested fixes are concrete enough to paste into your editor or coding tool.

Analyze your repo

Paste a GitHub URL and get your report in minutes.