revfactory/harness-for-agy

Harness-for-agy is a meta-skill for Google's Antigravity CLI (agy) that generates an entire team of specialized AI agents for a given project domain in a single command. You describe what you need in one sentence, such as "set up a React testing automation team," and the skill produces the agent persona files, the skill definitions each agent will use, and an orchestrator skill that coordinates them, all organized into a structured directory under your project. The project is a port of a similar meta-skill originally written for Claude Code. The core concept is the same: a single command scaffolds a complete agent team. The output is adapted to fit how Antigravity registers and runs agents. Each agent gets a Markdown persona file and a flat JSON runtime profile that the CLI reads to register the agent via its define_subagent API. The orchestrator skill then uses invoke_subagent and send_message calls to coordinate work between them, with a standardized JSON message format carrying task ID, action type, target file path, and context metadata. Running /harness again in the same project directory switches to a review mode. It inspects what already exists, identifies gaps between the current state and what a well-structured team would look like, and fills in what is missing without requiring a fresh start. The meta-skill selects from six orchestration patterns (pipeline, fan-out/fan-in, expert pool, producer-reviewer, supervisor, and hierarchical delegation) based on the domain and complexity of the task. It also adjusts the configuration based on estimated user experience level, producing simpler setups for beginners and more autonomous coordination for advanced users.