LangGraph4j 1.9 Release: Skill‑Based Sub‑Agent Pattern
Implement SkilledReactSubAgent to turn markdown skills into sub‑agents exposed as tools.
Migrate to LangGraph4j 1.9 and use SkilledReactSubAgent.
Summary
LangGraph4j 1.9 introduces a new pattern that turns markdown skills into executable sub‑agents exposed as tools. The implementation parses a SKILL.md file, extracts name and description, filters allowed tools, uses the body as sub‑agent instructions, compiles the agent, and wraps it into a Spring AI function tool. This allows a parent ReAct agent to invoke a child agent as a tool while keeping each agent focused and low‑context.
The release stream 1.9 is a major version with breaking changes, making it a sprint‑backlog item for teams using LangGraph4j. The pattern improves modularity, reduces token usage, and preserves the ReAct planning model.
Key benefits include reusable skills, tool‑backed sub‑agents, and a clean hierarchy without new orchestration primitives.
Adopting this pattern requires migrating to the new release and refactoring existing skills into the SKILL.md format.
Key changes
- Parse SKILL.md
- Extract name and description
- Filter allowed‑tools
- Use markdown body as sub‑agent instructions
- Compile agent
- Wrap into function tool