Agent Stage: Review & Validation (50)

When to use: Review your own or others' code, respond to feedback, or audit security/architecture.

Main Flow: Code Review

graph TB
    subgraph Review [👁️ Phase 1: Review]
        direction TB
        A[PR ready] --> B[code-review]
        B --> C{Security concerns?}
        C -- Yes --> D[security-check]
    end

    subgraph DeepDive [🔍 Phase 2: Deep Dive - Optional]
        direction TB
        C -- No --> E{Architecture issues?}
        E -- Yes --> F[review-architecture]
        D --> F
    end

    subgraph Respond [📝 Phase 3: Respond]
        direction TB
        E -- No --> G{Findings?}
        F --> G
        G -- Critical --> H[respond-to-review]
        H --> I[Implement fixes]
        I --> J{Critical fixed?}
        J -- No --> K[Re-review]
        J -- Yes --> L((Merge))
    end

    G -- None/minor --> L
    K --> B

    %% Styling
    style Review fill:#f5f5f5,stroke:#333,stroke-dasharray: 5 5
    style DeepDive fill:#f3e5f5,stroke:#4a148c
    style Respond fill:#e1f5fe,stroke:#01579b
    style C fill:#ffd54f
    style E fill:#ffd54f
    style G fill:#ffd54f
    style J fill:#ffd54f
    style L fill:#e8f5e9,stroke:#1b5e20,stroke-width:3px

code-review

Goal: Systematic Rails PR review.

Checklist by Area

Severity Levels

security-check

Goal: Deep security dive.

Audit Checklist

• Auth — Session management, token handling
• Authorization — IDOR, role checks, policy coverage
• Input validation — Strong params, SQL injection
• Output encoding — XSS prevention
• Redirects — Open redirect vulnerabilities
• Secrets — Never in code, logs, or VCS
• GraphQL — Introspection off in prod, depth limits

review-architecture

Goal: Structural review of boundaries and abstractions.

Review Signals

• Feature crosses multiple models without clarity
• Service creates/modifies unrelated models
• Complex callbacks calling other models
• Logic duplicated between controllers

Output

• Boundary recommendations
• Extraction suggestions
• Coupling assessment

respond-to-review

Goal: Respond to received feedback.

Process

1. Evaluate each suggestion — is it correct?
2. Push back if wrong — explain why
3. Implement accepted items — one at a time
4. Re-review mandatory if Critical findings

Anti-pattern: "LGTM! Will address in follow-up" — no performative agreement

Skills in this Stage