Omni-IO Skills: Harnessing Your Agent Omni-Native
The paper argues agents can become multimodal through an external skill harness, without retraining their reasoning core.
Omni-IO Skills wraps existing agents with hierarchical skills, a multimodal execution interface, dependency-aware scheduling, and a persistent asset registry. Its workflows use Declare Execution Graphs so independent operations can run in parallel and successful outputs can be reused later. The authors report 27 skills across 38 tasks in seven artifact modalities, covering understanding, generation, reasoning, and retrieval. On UniM-90, the harness raised input-support rates for GPT-5.6 Sol and Claude Sonnet 5 to 100% and sharply improved the reported semantic-quality scores. HF Daily Papers' note
Omni-IO Skills wraps existing agents with hierarchical skills, a multimodal execution interface, dependency-aware scheduling, and a persistent asset registry. Its workflows use Declare Execution Graphs so independent operations can run in parallel and successful outputs can be reused later. The authors report 27 skills across 38 tasks in seven artifact modalities, covering understanding, generation, reasoning, and retrieval. On UniM-90, the harness raised input-support rates for GPT-5.6 Sol and Claude Sonnet 5 to 100% and sharply improved the reported semantic-quality scores. HF Daily Papers' note
score 5