What problem does it solve?
When you ask an AI to "eval" or "review" current work, it often confuses the evaluation request with the task being judged, invents scores nobody asked for, or applies generic quality checklists. This Skill resolves the actual work product and original goal from the conversation, then delivers a focused, evidence-backed review.
Core Features & Use Cases
- Qualitative Review by Default: Inspects files, diffs, plans, or conversation artifacts and reports findings by impact, strengths, prioritized fixes, and a plain-language completion verdict—no numbers unless requested.
- Checklist and Rubric Modes: Builds task-specific definition-of-done checklists or unscored must/should/could rubrics on request.
- Scored Evaluation (Opt-In): Applies the Qworld Recursive Expansion Tree (RET) method with weighted binary criteria only when the user explicitly asks for scores, grades, or LLM-as-judge evaluation.
- Use Case: After implementing a feature, ask "eval current work" and receive a verdict like "partially complete" with cited evidence of unmet requirements and the smallest fixes that close the gaps.
Quick Start
Ask the assistant to review the current work against the original request and tell you what is missing.
