The revfactory/harness repository provides a meta-skill plugin for Claude Code that automates the design and generation of domain-specific agent teams and their associated skills. Its primary purpose is to streamline the process of decomposing complex tasks into coordinated teams of specialized agents, leveraging six pre-defined architectural patterns. By simply describing a project or domain, users can prompt Harness to generate the necessary agent definitions and skills, which are then organized within the Claude Code environment for immediate use.
Harness operates at the "L3 Meta-Factory" layer within the Claude Code ecosystem, specifically as a "Team-Architecture Factory." This distinguishes it from other tools in the ecosystem, such as Archon (which focuses on deterministic runtime configuration) or meta-harness (a Codex runtime port). Harness’s unique value lies in its ability to translate a domain description into a structured team of agents, each with specialized skills, using one of six architectural patterns: Pipeline, Fan-out/Fan-in, Expert Pool, Producer-Reviewer, Supervisor, and Hierarchical Delegation. These patterns cover a wide range of collaborative and task-decomposition strategies, allowing users to select the most appropriate structure for their needs.
Key features of Harness include automatic agent team design, skill generation with progressive disclosure for efficient context management, orchestration protocols for inter-agent communication and error handling, and built-in validation tools for trigger verification and dry-run testing. The plugin supports multiple execution modes, including default agent teams (for collaborative multi-agent tasks) and subagents (for one-off, direct tool invocations). This flexibility ensures that Harness can be adapted to a variety of workflows, from simple to highly complex.
Installation is straightforward, either via the Claude Code marketplace or as a global skill. Once installed, users can trigger Harness with natural language prompts, such as "build a harness for this project," and the plugin will generate the appropriate files and configurations. The output includes agent definitions and skill files, organized for easy integration into the Claude Code environment.
Harness is designed to coexist with other tools in the agent-framework ecosystem. For example, it can be used alongside Archon for runtime determinism or with ECC for cross-harness workflow standardization. It also integrates well with agent and skill catalogs, such as wshobson/agents, allowing users to incorporate existing components into their custom teams.
The effectiveness of Harness has been demonstrated through research and practical applications. In controlled experiments, using Harness led to a 60% average improvement in code agent output quality (from 49.5 to 79.3), a 100% win rate across 15 tasks, and a 32% reduction in output variance. These results suggest that Harness is particularly beneficial for complex tasks, where structured pre-configuration and team-based approaches yield significant gains.
Harness is suitable for a wide range of use cases, including deep research, website development, content production, code review, technical documentation, data pipeline design, and marketing campaigns. Its modular design, support for multiple languages, and integration with Claude Code’s agent team system make it a powerful tool for organizations seeking to enhance productivity and output quality through automated agent team generation.
