buddy-agent-harness
Initialize or update a consumer repository's standards-based agent configuration so its instructions, skills, and compatible tool settings work across multiple agent harnesses.
Install the plugin
In Claude Code, add the cyberplace marketplace and install the plugin and its init-buddy-agent-harness skill:
/plugin marketplace add cyberuni/cyberplace
/plugin install buddy-agent-harness@cyberplace
The plugin depends on the cyber-agentkit plugin, whose reference skill loads the references these skills name. Claude Code installs it with this one. Cursor, Codex, and GitHub Copilot CLI read no plugin dependencies, so install cyber-agentkit there yourself.
CLI
doctor is the read-only half. It runs on any clone, with the plugin installed or not:
npx -y buddy-agent-harness doctor
init runs behind the init-buddy-agent-harness skill, which sorts the configuration you already wrote before the command links anything.
dep-plugins derives a marketplace catalog for plugins shipped by this repository's own dependencies, so a harness can install them:
npx -y buddy-agent-harness dep-plugins
It writes the catalog and reports what each detected harness needs in order to match it, with the command that does it — reported rather than run, because registering and installing write into a developer's own harness state.
To mount these commands on buddy, install the package alongside repobuddy and declare its plugin module in the consumer repository's .repobuddy.json:
{
"plugins": ["buddy-agent-harness"]
}
Then run buddy agent-harness doctor. repobuddy deliberately loads plugins declared in this configuration; it does not scan installed dependencies.
The canonical configuration is the repository root's AGENTS.md and .agents/ tree: .agents/AGENTS.md holds shared behavior, .agents/skills/**/SKILL.md holds capabilities, and separately named files hold tool settings.
Codex, Cursor, GitHub Copilot CLI, Gemini CLI, and Devin Desktop read .agents/skills/ natively, so nothing is written for them. Claude Code gets a directory-level symlink to it. Claude Code and Cursor are always enabled; a detected harness directory or --harness codex,gemini-cli adds others. The command preserves user-authored configuration and reports TOON by default; it records nothing on disk.
init takes --root, --harness, --copy, --force, and --format; --help lists them.
buddy-agent-harness doctor is the read-only check that those bridges still resolve after a clone. It reports every bridge with a status, names the init command that repairs each finding, and always exits 0 — the diagnosis succeeding is not the same as the repository being healthy. See skills/doctor-buddy-agent-harness/SKILL.md.
The init-buddy-agent-harness skill goes further than the CLI: it surveys agent configuration the repository already has, consolidates it into the canonical source with your approval, and then links. See skills/init-buddy-agent-harness/SKILL.md.
