Skip to content
The panel

Specialist reviewers that stay in their lane.

General-purpose AI reviewers get noisy fast. PR Quorum gives you a catalog of specialists — three on by default, five more ready to enable, and fully custom agents you build yourself — each with its own model, focus list, confidence threshold, and posting behavior.

Specialist templates
8
On by default
3
Custom agents
Pro+
Model choice
any

Default reviewers

Correctness

Finds likely bugs, regressions, edge cases, and test gaps by arguing backward from runtime failure modes.

likely-bugregressionedge-casemissing-testnull-deref
Model
deepseek/deepseek-v4-flash
Best at
runtime bugs
Noise
balanced
Control
YAML
Security

Looks for secrets, auth bypasses, injection risks, unsafe data handling, and security-sensitive logic drift.

secret-leakauth-bypassinjectionunsafe-inputcrypto-misuse
Model
deepseek/deepseek-v4-flash
Best at
risk paths
Noise
strict
Control
YAML
Architecture

Keeps changes aligned with repo conventions, framework patterns, maintainability, and complexity budgets.

convention-driftover-abstractionpattern-violationcomplexity
Model
deepseek/deepseek-v4-flash
Best at
design drift
Noise
calm
Control
YAML

Models shown are the defaults. Route any reviewer to any OpenRouter model — your key or ours — tune its focus and confidence, or add a custom specialist, all per repo in .ai-review.yml.

More specialists, ready to enable

Flip these on per repo when the diff calls for it — same noise controls, still one clean review.

TestsEnable
missing test coverageflaky patternssnapshot driftmock misuse
PerformanceEnable
n+1 queriesunnecessary work in hot pathsmemory leaksbundle bloat
Frontend UXEnable
accessibilityloading and error statesunintended layout shiftmobile breakpoints
MigrationsEnable
destructive DDLunsafe defaultslong-running locksreversibility
AccessibilityEnable
ARIA correctnesskeyboard navigationcolor contrastsemantic HTML
Custom agents
Or build your own specialist

Create a reviewer with your own system prompt, focus list, and model — a house-style enforcer, a domain-rules checker, a flaky-test hunter. Up to 20 per panel. The default panel of three is free; enabling extra templates or custom agents is on Pro and above.

Start free on GitHub
Inside an agent

Prompts become typed findings, not vague prose

Each reviewer returns structured findings with severity, confidence, file, line, title, body, and an optional suggestion. That structure is what lets PR Quorum filter noise before it reaches the PR.

1. System prompt
Reviewer focus list, merged with your .ai-review.yml policy
2. PR diff
Sent verbatim to OpenRouter chat-completions
3. Structured output
Zod-validated JSON, parsed in the Inngest function
4. Confidence floor
Drop anything below min_confidence (default 0.75)
Example finding · structured output
{
  "severity":   "high",        // low | medium | high | critical
  "confidence": 0.87,          // 0..1, dropped below min_confidence
  "file":       "src/billing/stripe-webhooks.ts",
  "line":       142,
  "title":      "Webhook signature verified after side-effects",
  "body":       "...prose for humans...",
  "suggestion": "...diff-shaped fix..."   // optional, posted as GH suggestion
  // reviewerId is added by the runtime
}
Illustrative example — real findings reference your own code and diff lines.

Put the specialist panel on your next PR

Start with the default three, enable five more from the catalog, or build your own — then tune the model, confidence, and focus per repo.