ai-code-reviewer_skill

Overview

This skill performs deep, multi-model AI-powered code reviews using external CLIs (OpenAI Codex and Google Gemini). It returns severity-based findings, deduplicates results, and detects secrets to produce actionable, confidence-scored recommendations. Use it when you need high-assurance reviews beyond simple linting or formatting.

How this skill works

The skill calls one or both external model CLIs to analyze provided code snippets or projects, then aggregates findings with a similarity-based deduplication and consensus scoring. It also runs a local secret scanner for hardcoded credentials and persists results in a short-term cache for repeated queries. Async operations expose status information so long-running audits can be polled and resumed.

When to use it

• High-confidence code reviews that need multiple AI perspectives
• Deep security checks including secret and credential detection
• Performance and architectural analysis requiring Gemini-style insights
• Large-codebase audits where caching and parallel execution matter
• Full audits combining functional review and secret scanning

Best practices

• Choose 'Quick Review' for fast single-model checks and 'Deep Review' for combined consensus
• Provide project context (framework, API type) to improve relevance
• Exclude test and mock files to reduce noise in secret scanning
• Run combined analysis for high-stakes changes and keep cache TTL in mind
• Handle missing CLI or auth errors by following the provided remediation guidance

Example use cases

• Pre-merge deep review of a critical backend change using combined models
• Pre-commit secret scan to catch hardcoded keys before pushing
• Performance audit of a hotspot function with Gemini-focused analysis
• Incident triage where rapid, aggregated findings speed remediation
• Periodic large-repo audits that leverage cache to reduce costs