SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale
SkillDAG introduces a way for LLM agents to select skills by exposing their relationships as a typed graph. Instead of fixed similarity matching, it gives the agent search results including matches, neighbors, and conflicts. An agent-driven 'propose-then-commit' loop allows the agent to build and evolve this skill graph based on actual task execution evidence.