Discover how the ADK SkillToolset uses progressive disclosure to build self-extending AI agents that load expertise on demand. Learn four patterns to reduce token costs and enable agents to dynamically generate their own skills using the agentskills.io specification.

GoogleDevs' platform is a central hub for developers interested in Google technologies, APIs, and developer tools. Through articles, tutorials, and documentation, GoogleDevs offers insights into building applications using Google Cloud Platform, Android, Chrome, and other Google services. Developers can learn about cloud computing, machine learning, and mobile app development with Google's developer tools and platforms.

Google Developers

Google's Agent Development Kit (ADK) introduces a SkillToolset that uses progressive disclosure to reduce token costs and enable dynamic agent capabilities. Instead of loading all domain knowledge into a monolithic system prompt, skills are broken into three levels: L1 metadata (always visible), L2 instructions (loaded on demand), and L3 reference documents (loaded when needed). Four patterns are covered: inline skills defined in code, file-based skills with directory structure, external community skills, and meta skills that let agents generate new SKILL.md files at runtime. The approach can reduce baseline context usage by ~90% for agents with many skills. Generated skills follow the agentskills.io spec, making them compatible with ADK, Gemini CLI, Claude Code, Cursor, and 40+ other tools.

Developer’s Guide to Building ADK Agents with Skills

Pattern 1: Inline skills (the sticky note)

Pattern 2: File-based skills (the reference binder)

Pattern 4: The skill factory - Meta skills that write themselves