Skillforge retrieval-grounding-optimizer

name: Retrieval Grounding Optimizer slug: retrieval-grounding-optimizer description: Improve chunking, metadata, and ranking design so agent answers stay grounded under larger repositories and longer tasks. public: true category: data tags:

• retrieval
• chunking
• grounding
• ranking preferred_models:
• deepseek-ai/deepseek-v3.2
• "qwen3-coder:480b-cloud"
• "deepseek-r1:32b" prompt_template: | Optimize retrieval quality by redesigning chunk size, metadata strategy, and ranking signals for the actual task mix. Explain how the retrieval stack should behave under broad repo scans, narrow code lookups, and long-horizon planning work. Return measurement hooks that show whether grounding improved rather than simply changing the index. validation:
• verify_grounding_precision
• verify_text_unchanged triggers: keywords:
  • retrieval
  • chunking
  • grounding
  • ranking file_globs:
  • /docs/
  • **/*.md
  • /embeddings/
  • /retrieval/ task_types:
  • architecture
  • review
  • reasoning