Problem
Currently, if an agent loads all skills into its context window simultaneously, it consumes excessive tokens and can confuse the LLM about which workflow to follow (the "worst practice" of prompt stuffing). The docs/getting-started.md recommends "Context-Aware Loading", but there is no structured skill that teaches the agent how to route itself dynamically.
Proposed Solution
I'd like to contribute a skill-router meta-skill. Instead of loading everything upfront, the agent first loads only the skill-router. The router analyzes the user's intent and dynamically injects only the specific skill required (e.g., TDD, debugging, or the new resilience/fuzzing skill) into the context.
This approach formalizes the "Context-Aware Loading" guideline into a callable, structured workflow. I will also include a helper script that generates a quick JSON index of all skills, which the router can reference to save even more tokens.
Would this be a welcome addition to the skills/ directory?
Problem
Currently, if an agent loads all skills into its context window simultaneously, it consumes excessive tokens and can confuse the LLM about which workflow to follow (the "worst practice" of prompt stuffing). The docs/getting-started.md recommends "Context-Aware Loading", but there is no structured skill that teaches the agent how to route itself dynamically.
Proposed Solution
I'd like to contribute a skill-router meta-skill. Instead of loading everything upfront, the agent first loads only the skill-router. The router analyzes the user's intent and dynamically injects only the specific skill required (e.g., TDD, debugging, or the new resilience/fuzzing skill) into the context.
This approach formalizes the "Context-Aware Loading" guideline into a callable, structured workflow. I will also include a helper script that generates a quick JSON index of all skills, which the router can reference to save even more tokens.
Would this be a welcome addition to the skills/ directory?