increment-quality-judge-v2

What This Skill Does

The increment-quality-judge-v2 is a sophisticated, AI-driven assessment tool designed to enforce rigorous quality standards in your development workflow. It leverages the 'LLM-as-Judge' pattern, a gold-standard approach in machine learning evaluation that mimics expert peer review through chain-of-thought processing. By systematically breaking down your specifications, task plans, and implementation artifacts, this skill identifies structural, logical, and technical risks using a BMAD (Basis, Magnitude, Assessment, Direction) risk-scoring framework. It evaluates your work against seven critical quality dimensions, providing a granular verdict that culminates in a formal quality gate decision (PASS, CONCERNS, or FAIL). Unlike manual review, which is prone to human fatigue and subjective bias, this skill delivers consistent, scalable, and evidence-based critiques in near real-time, ensuring that only high-quality increments proceed through your pipeline.

Installation

To integrate this skill into your local environment, ensure you have the OpenClaw SDK installed. Execute the following command in your terminal:

clawhub install openclaw/skills/skills/anton-abyzov/sw-increment-quality-judge-v2

Once installed, the skill becomes available via your CLI and internal slash command interfaces. No further agent configuration is required.

Use Cases

• Spec Validation: Ensure new increment specifications are exhaustive, unambiguous, and technically feasible before coding begins.
• Task Completeness: Audit task lists to ensure all requirements are mapped to specific, actionable, and testable units of work.
• Quality Gate Enforcement: Automate the assessment of pre-commit or pre-merge artifacts to prevent technical debt from entering the codebase.
• Technical Due Diligence: Use the risk-scoring mechanism to evaluate the complexity and potential failure modes of an proposed architectural change.

Example Prompts

1. "Run a quality check on the current spec for increment 0042 to ensure it meets our documentation standards."
2. "/sw:qa 0089 --pre - Perform a deep dive evaluation on the task list and provide a risk report."
3. "Evaluate the proposed implementation plan for the new auth module; look specifically for edge-case coverage and security gaps."

Tips & Limitations

• Context is Key: Always provide the full scope of your increment; the skill's accuracy improves with the richness of your input metadata.
• Not an Agent: Do not attempt to spawn this as an autonomous agent. It is a utility skill. Using it via the provided CLI commands ensures the highest performance and stability.
• Iterative Refinement: When the skill returns a CONCERNS verdict, treat the actionable recommendations as a checklist. You can re-run the assessment after implementing changes to verify the fix.
• Human-in-the-loop: While highly accurate, the judge acts as an assistant. High-criticality decisions should still be reviewed by senior personnel where organizational policy dictates.