Paper summary: Raven: The Harness of Harnesses for Composable Agentic Intelligence

Problem

As agents tackle long, cross-domain workflows, harnesses get too complex to design by hand, and a harness tuned for one domain doesn’t generalise. The question shifts from “build a better harness” to “build, improve and orchestrate many specialised harnesses automatically”.

Method

Raven (“The Harness of Harnesses”) is an open-source multi-agent ecosystem that:

• automatically constructs and evolves modular harnesses for specific models and domains, treating each executable model–harness pair as a composable unit
• uses a Host Agent to decompose goals, match subtasks to specialised agents, coordinate dependencies and integrate results
• preserves experience across tasks (host archive, EverOS) and turns it into reusable procedures (Skill Forge)

The authors give sufficient conditions under which such composition expands reliable task coverage beyond any individual agent, under a shared budget.

Key results

From the abstract: on complex, long-horizon tasks, Raven significantly outperforms state-of-the-art agent systems.

Limitations

There’s no dedicated limitations section in the extracted text, and the abstract doesn’t give specific numbers.