Herding CATs: ALARA for Agent Harness Engineering in Portable Composable Multi-Agent Teams

TL;DR

This paper introduces a plain-text data layer called CAT that improves management, sharing, and coordination of multi-agent systems by allowing direct control over agent tools and behaviors.

Contribution

It applies the ALARA principle to agent context management, enabling scalable, transparent, and collaborative control of agent tool access and configurations.

Findings

CAT enables real agentic usage with a command-line shell.

Evaluated 22 models across 115 tasks, identifying success and failure patterns.

Characterized model performance and breakdowns across tasks and model sizes.

Abstract

Industry practitioners and academic researchers regularly use multi-agent systems to accelerate their work, but the applications through which users operate these systems do not provide a simple, unified mechanism for scalably managing critical components of the agent harness. This lack of control adversely impacts both the quality of individual human-agent interactions and reduces the capacity for practitioners to coordinate context engineering efforts. The behavioral specifications that define what agents in such systems can do remain fragmented across prose instruction files -- for which compliance cannot be guaranteed -- or framework-internal configurations, making these specifications difficult to share, version, or collaboratively maintain across teams and projects. Applying the ALARA principle from radiation safety (exposures kept as low as reasonably achievable) to context, we introduce a context-agent-tool (CAT) data layer expressed through interrelated plain-text files, allowing users to directly declare tool access for each agent and to modify the tools themselves that are used by the agents when processing. We demonstrate capability of this CAT data layer to enable real agentic usage by using a command-line shell that loads the team and executes agent runs -- \texttt{npcsh} -- and evaluating 22 locally-hosted models from 0.6B to 35B parameters across 115 practical tasks spanning file operations, web search, multi-step scripting, tool chaining, and multi-agent delegation. We characterize which model families succeed in certain task categories and where they break down across $\sim$2500 total executions.