skill-code-review

The skill-code-review agent provides an advanced, multi-perspective code auditing environment designed to eliminate the blind spots inherent in single-model LLM analysis. By orchestrating a pipeline of up to eight disparate AI providers including Codex, Gemini, Copilot, and Claude, the agent synthesizes diverse architectural and security viewpoints. It is specifically built for high-stakes software development environments where autonomous code generation, complex architectural patterns, or rigorous security compliance are required. The process operates through a structured, multi-phase methodology that performs consensus scoring to validate findings before they reach production.

• Executes a mandatory multi-LLM review pipeline ensuring diverse adversarial perspectives rather than a single-model assessment.
• Automates security vulnerability detection mapped against OWASP standards and best practices.
• Conducts autonomous codegen risk assessment to identify placeholder logic, dead branches, and speculative abstractions.
• Verifies test-driven development (TDD) compliance by analyzing test history and provenance for production code changes.
• Performs stub detection and implementation completeness verification to identify empty functions or incomplete logic before merges.
• Includes a scope-drift check that compares pull request diffs against stated project intent, surface-level requirements, and commit history.
• Implements a "Quick Mode" for rapid sanity checks and pre-commit reviews, alongside a full deep-audit mode for architectural impact reviews.

Usage involves triggering the code-reviewer persona through the orchestrate.sh CLI or via auto-routing triggers when review intent is detected. Users should provide specific context such as PR bodies, architectural design documents, or specific security concerns. The system is designed to be persistent, integrating with memory-persistence layers to recall past project decisions and context across sessions. It is recommended for developers managing large-scale repositories, CI/CD integrations for autonomous agent output, or security-sensitive codebases requiring multi-model validation. The agent outputs structured review reports containing consensus results, TDD compliance assessments, and autonomous risk scores, enabling teams to catch critical logic errors that single-agent setups might overlook.