SIGMA: Skill-Incidence Graphs for Compositional Multi-Agent Design

2026-06-19 · Source: cs.MA updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

SIGMA, a novel skill-incidence graph framework, addresses the generalization limitations of existing multi-agent system (MAS) designers by constructing agents as task-conditioned bundles of reusable skills. Unlike methods optimizing communication topologies over predefined agents, SIGMA predicts a skill-agent incidence matrix, composes agent node embeddings from selected skills, and decodes a communication topology. During execution, skill-specific mailboxes route messages to relevant assigned capabilities, making the incidence structure operational. Evaluated across six reasoning and coding benchmarks with three base LLMs, SIGMA achieved the best average performance, improving over the CARD baseline by 2.06, 2.36, and 1.75 points. It also demonstrated stronger robustness to unseen skill libraries, with an average performance drop of only 0.96 points.

Key takeaway

For AI engineers designing multi-agent systems that require adaptability to diverse or evolving tasks, SIGMA offers a robust approach. By enabling the dynamic composition of agents from a library of reusable skills, your systems can achieve superior generalization and performance compared to fixed-topology designs. Consider integrating skill-incidence graph frameworks to enhance system robustness and efficiency across various reasoning and coding applications.

Key insights

SIGMA enables multi-agent systems to generalize by dynamically composing agents from reusable skills, improving robustness and performance.

Principles

• Agent composition from skills enhances MAS generalization.
• Skill-incidence graphs enable dynamic agent formation.
• Compositional design complements topology optimization.

Method

SIGMA predicts a skill-agent incidence matrix, composes agent node embeddings from selected skills, decodes a communication topology, and routes messages via skill-specific mailboxes.

In practice

• Design agents from a reusable skill library.
• Implement skill-specific message routing.
• Apply to reasoning and coding tasks.

Topics

• Multi-Agent Systems
• Skill-Incidence Graphs
• Agent Composition
• Large Language Models
• Communication Topology
• Task Generalization

Best for: Research Scientist, NLP Engineer, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• How AI Agents Navigate Massive Skill Graphs — Skill-DAG enables LLMs to dynamically interpret and evolve a typed skill graph for scalable, context-aware skill selection, moving beyond semantic similarity.
• A Few AI Agents End Up Doing 80% of the Work - That Is a Design Problem — AI agent networks naturally form scale-free topologies with concentrated hubs, not flat distributions.
• The Illusion of Multi-Agent Advantage — Automated Multi-Agent Systems often incur high costs for superficial complexity, failing to outperform simpler Single-Agent Systems.
• SkillCAT: Contrastive Assessment and Topology-Aware Skill Self-Evolution for LLM Agents — SkillCAT enables LLM agents to self-evolve skills by contrastive assessment and topology-aware execution, boosting performance without training.
• OpenClaw-Skill: Collective Skill Tree Search for Agentic Large Language Models — Collective Skill Tree Search (CSTS) uses multi-model collaboration to build diverse, generalizable skill trees for LLM agents.
• Formal Skill: Programmable Runtime Skills for Efficient and Accurate LLM Agents — Formal Skill transforms LLM agent capabilities from informal text prompts into token-efficient, enforceable runtime-native protocols.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.MA updates on arXiv.org.