Code review

Code Review

Part of Agent Skills™ by googleadsagent.ai™

Description

Code Review enforces a structured pre-merge quality gate with a checklist-driven evaluation, severity-classified findings, and mandatory resolution tracking. The agent reviews every diff against a configurable set of quality dimensions before approving changes, ensuring consistent standards regardless of reviewer fatigue or time pressure.

Unlike ad-hoc review comments, this skill produces a standardized review document with findings categorized by severity: Critical (must fix before merge), High (should fix before merge), Medium (fix in follow-up), and Low (optional improvement). Each finding includes the file, line range, category, description, and a concrete suggested fix. The review document becomes part of the permanent record.

The pre-review checklist catches common oversights before deep analysis begins: missing tests, uncommitted files, linter errors, type errors, and documentation gaps. Only after the checklist passes does the agent proceed to semantic review of logic, architecture, security, and performance.

Use When

• A pull request or diff is ready for review
• The user asks for feedback on code changes
• Before merging any branch into main
• After a subagent completes a task (Stage 2 review)
• Code has been refactored and needs validation
• A new contributor's code needs onboarding-level review

How It Works

graph TD
    A[Receive Diff] --> B[Pre-Review Checklist]
    B --> C{Checklist Passes?}
    C -->|No| D[Return with Blockers]
    C -->|Yes| E[Semantic Analysis]
    E --> F[Classify Findings by Severity]
    F --> G[Generate Review Document]
    G --> H{Critical Findings?}
    H -->|Yes| I[Request Changes]
    H -->|No| J{High Findings?}
    J -->|Yes| K[Approve with Reservations]
    J -->|No| L[Approve]

The workflow gates progression: the checklist catches mechanical issues instantly, while semantic analysis evaluates design, correctness, and maintainability. The severity classification ensures critical issues block the merge while minor improvements do not.

Implementation

pre_review_checklist:
  - name: "Tests exist for changed code"
    command: "check_test_coverage_delta"
    severity: "critical"
  - name: "No linter errors introduced"
    command: "run_linter --diff-only"
    severity: "critical"
  - name: "Type checking passes"
    command: "run_typecheck"
    severity: "critical"
  - name: "No secrets in diff"
    command: "scan_secrets --diff"
    severity: "critical"
  - name: "Documentation updated"
    command: "check_doc_staleness"
    severity: "medium"

severity_levels:
  critical:
    label: "🔴 Critical"
    action: "Must fix before merge"
    examples: ["Security vulnerability", "Data loss risk", "Broken tests"]
  high:
    label: "🟠 High"
    action: "Should fix before merge"
    examples: ["Missing error handling", "Performance regression", "API contract violation"]
  medium:
    label: "🟡 Medium"
    action: "Fix in follow-up PR"
    examples: ["Code duplication", "Unclear naming", "Missing edge case test"]
  low:
    label: "🔵 Low"
    action: "Optional improvement"
    examples: ["Style preference", "Minor refactor opportunity", "Comment improvement"]

review_dimensions:
  - correctness: "Does the code do what it claims?"
  - security: "Are inputs validated? Are secrets protected?"
  - performance: "Are there N+1 queries, unnecessary re-renders, or blocking calls?"
  - maintainability: "Can another developer understand this in 6 months?"
  - testing: "Are edge cases covered? Are tests deterministic?"
  - architecture: "Does this follow established patterns? Is coupling appropriate?"

Best Practices

• Run the pre-review checklist before spending time on semantic review
• Limit findings to actionable items—avoid stylistic nitpicks unless egregious
• Always provide a suggested fix alongside each finding
• Separate critical blockers from nice-to-have improvements
• Review the test changes as carefully as the production code
• Acknowledge good patterns and design decisions, not just defects

Platform Compatibility

Platform	Support	Notes
Cursor	Full	Reads diffs via Shell tool
VS Code	Full	Git diff integration
Windsurf	Full	PR review workflow
Claude Code	Full	git diff and gh pr access
Cline	Full	Diff-based review
aider	Partial	Limited to file-level review

Related Skills

• Test-Driven Development - RED-GREEN-REFACTOR discipline that ensures reviewed code has corresponding test coverage
• Systematic Debugging - Root cause analysis methodology applied when code review uncovers defects
• Agent Security Scanning - Automated vulnerability detection that complements the security dimension of code review

Keywords

code-review quality-gate pre-merge-checklist severity-classification pull-request-review security-review performance-review review-document

---

© 2026 googleadsagent.ai™ | Agent Skills™ | MIT License