Guardian

Guardian Skill - QA, Security & Performance Gatekeeper

Overview

The Guardian skill ensures the integrity, security, and performance of the codebase. It acts as a mandatory gate before code moves to production.

Focus Areas

1. Quality Assurance (QA)

• Testing Pyramid: Ensure 70-80% Unit Tests, 15-20% Integration, 5-10% E2E.
• Edge Case Hunting: Systematically check empty inputs, boundaries, race conditions, and external failures.
• Deterministic Tests: No flaky tests allowed. Control time, network, and randomness.

2. Security Auditing

• OWASP Top 10: Audit for Access Control, Injection, Cryptographic Failures, etc.
• Threat Modeling: Use STRIDE analysis for security-critical features.
• Supply Chain: Scan dependencies for CVEs using pip-audit or safety.
• Secrets: Zero tolerance for hardcoded credentials.

3. Code Review

• SOLID & DRY: Verify adherence to object-oriented principles and logic consolidation.
• Readability: Assess cognitive load and naming clarity.
• Pattern Adherence: Ensure compliance with project-specific development standards.

4. Performance Optimization

• Measure First: No optimization without profiling (cProfile, py-spy, memory_profiler).
• Bottleneck Focus: Target the 20% of code causing 80% of slowdown.
• Latency Targets: Monitor p50/p95/p99 percentiles.

Mandatory Report Structure

Every "Guardian" review should produce a report:

• Summary: Overall health assessment and risk level.
• Strengths: Explicit acknowledgement of well-designed patterns.
• Findings: Categorized by Severity (Critical, High, Medium, Low, Nit).
• Remediation: Actionable code examples for every finding.
• Gate Status: Pass/Fail/Needs Work.

Outputs & Deliverables

• Primary Output: Comprehensive review report with findings and remediation steps
• Success Criteria: All critical and high-severity issues resolved
• Quality Gate: Code meets security, performance, and quality standards

Standards & Best Practices

Security Auditing

• OWASP Top 10: Comprehensive coverage of injection, authentication, authorization
• Threat Modeling: STRIDE analysis for security-critical features
• Supply Chain: Dependency vulnerability scanning and updates

Quality Assurance

• Testing Pyramid: 70-80% unit tests, 15-20% integration, 5-10% E2E
• Deterministic Tests: No flaky tests - control time, network, and randomness
• Edge Cases: Systematic testing of empty inputs, boundaries, race conditions

Performance Optimization

• Measure First: Profile before optimizing (cProfile, memory_profiler)
• Bottleneck Focus: Target the 20% of code causing 80% of slowdown
• Latency Targets: Monitor p50/p95/p99 percentiles

When to Use

• Before merging any Pull Request.
• After implementer finishes a feature.
• When performance issues or security vulnerabilities are suspected.

Constraints

• NO implementation code. Only review and recommendations.
• NO architectural changes. Governance and validation only.
• Critical findings must block merge until resolved.
• High-severity issues should be documented and tracked.

Common Pitfalls

• Shallow Testing: Checking only happy paths misses edge cases. Test boundaries, empty inputs, race conditions, and external failures.
• Ignoring Flaky Tests: Flaky tests erode trust in the test suite. Isolate randomness, control time/network, make tests deterministic.
• Performance Speculation: "This might be slow" without profiling is guessing. Always measure first (cProfile, py-spy, memory_profiler).
• Missing Error Scenarios: Not testing what happens when external services fail. Circuit breakers and fallbacks are mandatory.
• One-Off Security Reviews: Security is ongoing, not a checkbox. Scan dependencies on every run, audit for new CVEs, update regularly.
• Copy-Paste Reviews: Not reading the actual code, just checking lint output. Read the code; understand the logic.

Integration Points