Supported harnesses

agents syncs skills and instructions across these harnesses — the term we use for an AI coding tool that hosts skills (Claude Code, Codex, OpenCode, Gemini CLI). We avoid “agent” because it collides with each tool’s own internal agents/ subagent concept (e.g. Claude’s ~/.claude/agents/).

Each harness has a stable id used in agents.toml’s enabled_harnesses list and in lockfile entries’ harnesses field.

id	tool	skills directory	global instructions file
claude-code	Claude Code	~/.claude/skills/<name>/SKILL.md	~/.claude/CLAUDE.md
codex	OpenAI Codex CLI	~/.codex/skills/<name>/SKILL.md	~/.codex/AGENTS.md
opencode	OpenCode	~/.config/opencode/skills/<name>/SKILL.md	~/.config/opencode/AGENTS.md
gemini	Gemini CLI	~/.gemini/skills/<name>/SKILL.md	~/.gemini/GEMINI.md

All four use the same SKILL.md format (agentskills.io open standard): a directory containing a SKILL.md file with YAML frontmatter (name, description) and optional bundled scripts/, references/, assets/. agents’s symlink-from-cache install logic works identically for every harness — only the destination path differs.

Default-enabled set

By default, all four harnesses are enabled. agents’s apply will render instruction files into the agents repo and install harness symlinks for each one — even if the harness itself isn’t installed on this machine.

Managing the enabled set

Use the agents harness subcommand to toggle harnesses on and off. The commands edit enabled_harnesses in agents.toml for you, then re-render instructions and reconcile harness symlinks so the filesystem matches the new state immediately.

agents harness list                  # show every harness + enabled/disabled status
agents harness add gemini opencode   # enable one or more
agents harness remove gemini         # disable one or more

Both add and remove are idempotent — adding an already-enabled harness or removing one that isn’t enabled is a no-op with an informational message, not an error. agents refuses to remove the last enabled harness (that would disable agents itself); if you really want an empty list, edit agents.toml by hand.

Per-skill harness gating

The lockfile’s harnesses field on each skill entry restricts which harnesses that skill installs to. Use ["*"] for all enabled harnesses, or list specific ids:

[[skill]]
name = "my-skill"
harnesses = ["claude-code", "codex"]   # skip opencode + gemini for this skill

Adding support for another harness

Adding a harness is a one-row change in src/harness.rs. File an issue or open a PR with the harness’s stable id, display name, skills subdir (relative to install root), and instructions file path.

Cursor and Copilot are not currently supported because neither has a globally-syncable file surface — Cursor User Rules live in app settings and Copilot personal instructions live in VS Code settings, not files. Project-level rules for both live in the project repo, outside agents’s per-machine sync scope.