Implement skill-based knowledge retrieval layer #22
Description
Agents should retrieve knowledge based on:
- selected skills
- relevant document versions
- filtered chunks
Goals:
- prevent irrelevant knowledge injection
- maintain token efficiency
- improve reasoning accuracy