General-purpose agents such as Claude and Codex are strong at planning and reasoning over long horizons, but their production capabilities remain fragmented across text, images, audio, video, documents, 3D assets, and code. A new research paper, Omni-IO Skills: Harnessing Your Agent Omni-Native, proposes a way to close that gap without touching the agent's reasoning core.
The authors frame the problem as a choice between two costly paths: extending a foundation model to new modalities, which ties capability growth to expensive model updates, or assembling specialist models and tools, which leaves unclear how procedures, dependencies, intermediate assets, and cross-turn revisions should be coordinated.
Their answer is Omni-IO Skills, a plug-and-play Agent Harness that wraps around an existing agent rather than replacing it. It combines hierarchical Skills, a standardized multimodal execution interface, dependency-aware orchestration, and a persistent Asset Registry. Multi-asset workflows are represented as Declare Execution Graphs, which schedule independent operations concurrently and register successful outputs for downstream and cross-turn reuse across replaceable execution backends.
The harness ships with 27 Skills covering 38 representative tasks across seven artifact modalities and four capability families: understanding, generation, reasoning, and retrieval. The authors note that real tasks rarely stay within one modality — for example, a request to watch a video, compose background music, and turn it into a slide deck with a cover chains understanding, generation, and conversion while reusing intermediate assets across turns.
On the UniM-90 benchmark, the harness reportedly raises the input-support rates of GPT-5.6 Sol and Claude Sonnet 5 from 40.00% and 38.89% to 100%, while increasing the relative Semantic–Quality Coupled Score from 26.99 to 74.94 and from 27.82 to 77.78 respectively. Strict Structure Score reaches 100.00 and 99.78. The authors position these results as evidence that harness-level capability composition is a practical route to broad, evolvable omni systems without changing the host agent's reasoning core.
Code and 20 example workflows have been open-sourced alongside the paper.