DeepYard
S

SkillGate

Training framework for skill selection in long-horizon autonomous agents

Open SourceFree

About

Research framework addressing the credit assignment problem in hierarchical agent systems. SkillGate introduces specialized training signals to teach agents when to invoke procedural skills during long-horizon task execution, solving the 'selector credit starvation' problem where skill selection policies receive insufficient feedback. Relevant for researchers building autonomous agents that need to choose between multiple capabilities dynamically.

Details

Language
Patterns

Tags

autonomousframeworkopen-sourcemulti-agentorchestration