burnt storm
Issue: I've noticed that as the number of skills and stored memories increases, the Hermes Agent tends to become "less focused" and prone to logic errors.
Observations：
Instruction Overload: Core tasks are sometimes ignored when the system prompt is saturated with skill descriptions.
Memory Interference: Irrelevant past data is being retrieved, causing hallucinations in current tasks (especially during Python/scraping tasks).
Request: Are there plans to implement a memory pruning mechanism or a more advanced RAG filtering system to maintain performance as the agent's database grows?
Environment: WSL2 / SiliconFlow API / DeepSeek models.