Show HN: SkillMesh (role-based tool routing for Claude/Codex) (github.com)

SkillMesh, a new tool aimed at optimizing the interaction between large language models (LLMs) like Claude and Codex, has been introduced to address the challenges of tool overload in model prompts. By employing a retrieval-based routing system, SkillMesh intelligently selects only the most relevant tools based on user queries, drastically reducing the number of prompt tokens from an unwieldy 50,000+ to around 3,000. This selective approach not only enhances accuracy but also lowers operational costs by avoiding the need to process a bloated list of tools with each request. For the AI/ML community, SkillMesh represents a significant advancement in managing tool complexity and improving model performance. The core functionality relies on BM25 and dense retrieval techniques to score and filter expert cards, which encompass execution behavior, decision trees, and output contracts necessary for effective model operation. By focusing on relevant context, SkillMesh ensures that multi-domain tasks remain precise and manageable, thereby optimizing the deployment and efficiency of models in practical applications. Users can easily implement SkillMesh with straightforward configuration setups, enabling teams to enhance their LLM capabilities without the overhead of excessive tool integration.