Quality criteria for refined_idea output. Use before RefineIdea or when ValidateRefinedIdea returns low score. Focus on concrete, decomposable, and specific output.
Install
npx skillscat add dozybot001/maars/refined-idea-quality Install via the SkillsCat registry.
SKILL.md
Refined Idea Quality Criteria
Guidelines for producing a high-quality refined idea (Markdown) that is decomposable into 3–10 tasks. Quality matters.
Quality Rules
- Concrete over vague: "Compare BERT and GPT on code completion benchmarks" not "Research NLP models".
- Combine literature insights with novel angles: Avoid mere summarization; add synthesis, comparison, or gap.
- Must be decomposable: If you cannot list 3+ distinct sub-tasks, refine further.
- Gap must be specific: Name limitations (e.g. "prior work focuses on X, neglects Y") rather than generic "more research needed".
Validation Hints (ValidateRefinedIdea)
- Score >= 4: Proceed to FinishIdea.
- Score < 4: Call RefineIdea again with improvements — tighten scope, clarify questions, or sharpen gap.
Red Flags
- Output contains "research" without scope
- Questions are yes/no or too broad
- Gap is generic ("limited studies", "need more work")
- Method or approach is a single vague step