This skill covers expert architectural patterns, routing strategies, and troubleshooting guidelines for the Google Agent Development Kit (ADK).
1. Multi-Agent Structure
Discovery Rule: Only the root_agent defined in agent.py or root_agent.yaml is automatically surfaced in the platform web UI. Sub-agents are invisible to the top-level application but are called natively via routing.
Routing Rules: Always provide the list of sub-agents directly to the parent agent via the sub_agents attribute. Do not construct wrapper functions to trigger a sub-agent manually.
2. Architectural Patterns
Relay/Router Pattern:
Design the orchestrator agent to act exclusively as a request classifier.
Direct it to output strict XML-like signals (e.g., <route>ROUTING_TO_RAG</route>).
Catch this signal in the main execution loop, running only the appropriate target agent to prevent nested agent call stacks and context pollution.
Capabilities Discovery Pattern:
When a user queries broad knowledge (e.g., "what do you know?"), trigger searches for broad index terms like "Overview" or provide a predefined hardcoded list of capabilities.
3. Routing Strategies
File Presence Pattern: Route to file-processing agents only if the user attaches a file or references "this attachment". Route to RAG/Knowledge agents for general queries and report generation.
Strict Evaluation Order: Categorize intents explicitly (e.g., [1] DIRECT, [2] RAG, [3] CHAT) and designate one specific category as the default fallback. Avoid negative constraint prompts.
4. Prompt Engineering Best Practices
No Preamble Rule: When an agent triggers a tool, harden the prompt to prohibit introductory filler (e.g., forbid saying "I can help with that"). Insist that the agent's first and only action must be to call the tool.
1---2name: adk-framework3description: Skill: ADK Framework4---5# Skill: ADK Framework67This skill covers expert architectural patterns, routing strategies, and troubleshooting guidelines for the Google Agent Development Kit (ADK).89---1011## 1. Multi-Agent Structure12* **Discovery Rule:** Only the `root_agent` defined in `agent.py` or `root_agent.yaml` is automatically surfaced in the platform web UI. Sub-agents are invisible to the top-level application but are called natively via routing.13* **Routing Rules:** Always provide the list of sub-agents directly to the parent agent via the `sub_agents` attribute. Do not construct wrapper functions to trigger a sub-agent manually.1415## 2. Architectural Patterns16* **Relay/Router Pattern:** 17 * Design the orchestrator agent to act exclusively as a request classifier.18 * Direct it to output strict XML-like signals (e.g., `<route>ROUTING_TO_RAG</route>`).19 * Catch this signal in the main execution loop, running only the appropriate target agent to prevent nested agent call stacks and context pollution.20* **Capabilities Discovery Pattern:** 21 * When a user queries broad knowledge (e.g., "what do you know?"), trigger searches for broad index terms like "Overview" or provide a predefined hardcoded list of capabilities.2223## 3. Routing Strategies24* **File Presence Pattern:** Route to file-processing agents **only** if the user attaches a file or references "this attachment". Route to RAG/Knowledge agents for general queries and report generation.25* **Strict Evaluation Order:** Categorize intents explicitly (e.g., `[1] DIRECT`, `[2] RAG`, `[3] CHAT`) and designate one specific category as the default fallback. Avoid negative constraint prompts.2627## 4. Prompt Engineering Best Practices28* **No Preamble Rule:** When an agent triggers a tool, harden the prompt to prohibit introductory filler (e.g., forbid saying "I can help with that"). Insist that the agent's first and only action must be to call the tool.
Run npx skillmds@latest add izzyfresh/adk-framework in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
Skill: ADK Framework It is listed under Coding & Dev Tools on SkillMD.
This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
IzzyFresh (@izzyfresh) published this skill. Their other Agent Skills are listed on their SkillMD profile.