Buddy Agent Harness

What it does

Buddy Agent Harness treats the repository root’s AGENTS.md and .agents/ tree as canonical agent configuration: .agents/AGENTS.md holds shared behavior, .agents/skills/**/SKILL.md holds capabilities, and separately named files hold tool settings.

It consolidates the configuration you already have into that source, then polyfills the harnesses that cannot read it. Like any polyfill, it writes nothing where support is already native: Codex, Cursor, Copilot CLI, and Devin Desktop read .agents/skills/ directly. Only Claude Code and Gemini CLI need a link.

It never invents policy, never rewrites your instructions, and leaves anything without a documented mapping canonical rather than converting it.

Install

In Claude Code, add the cyberplace marketplace and install the plugin:

/plugin marketplace add cyberuni/cyberplace
/plugin install buddy-agent-harness@cyberplace

Install from the marketplace rather than a git checkout: the skills run a CLI that has dependencies. See Introduction.

Use it

Ask your agent to initialize the repository, and it runs the init skill: survey, classify, confirm, apply, verify. Nothing you authored changes before you approve the plan.

When a harness later loads no project skills, the doctor skill says which bridge stopped resolving and names the command that repairs it.

Behind each skill is a command, and the commands are an npm package the plugin is not a prerequisite for. doctor runs on any clone at all:

npx -y buddy-agent-harness doctor --format text

It prints TOON by default, for an agent to parse. --format text prints the same report for a person.

Where to go next

• Skills: init, doctor and enhance, and when to run the CLI instead.
• Skill: init: the five phases, and what you approve.
• Migrating Existing Configuration: CLAUDE.md, .cursor/rules/, existing skills.
• Open Standards: AGENTS.md, Agent Skills, and how much authority .agents/skills/ has.
• Harness Differences: who reads what, and who needs a bridge.