adk-skill-creation

The agentskills.io format standardizes agent skill creation by enforcing rigorous frontmatter metadata, ensuring skills are discoverable and interoperable. It provides a specification-driven framework for documenting and wiring skills into ADK agents consistently.

The best way to implement progressive disclosure patterns is using a three-tier (L1/L2/L3) architecture. L1 provides metadata, L2 delivers instructions, and L3 holds reference files, optimizing token usage and agent performance by loading documentation only when required.

You generate a new domain-specific skill by loading the adk-skill-creation skill, which creates the directory structure and SKILL.md frontmatter. This includes a routing table for L3 references, ensuring the agent only loads necessary documentation dynamically.

Yes, you can use adk-skill-creation to improve agent modularity and performance. It ensures consistent skill wiring via the SkillToolset and implements progressive disclosure, which optimizes token usage by loading only necessary L3 reference files during complex agent builds.

Your ADK agent loads unnecessary documentation because the skill lacks a progressive disclosure pattern. Implementing an L1/L2/L3 architecture with a routing table for L3 references ensures the agent only loads specific documentation when required, optimizing token usage.