Agent Integration Guide

grimoire works across eight AI agents. Each has its own plugin format and installation method. This guide covers installation, usage, and tool mapping for each.

Supported agents

Claude Code

Install:

/plugin marketplace add jeffreytse/grimoire-core
/plugin install grimoire@grimoire-core

Plugin file: .claude-plugin/plugin.json

Loads all skill paths listed in the skills array. Skills are namespaced by plugin name, e.g. /grimoire-engineering:propose-conventional-commit.

Using skills:

/suggest-best-practice
/write-best-practice-skill
/review-best-practice-skill

Invoke any meta skill by name with a leading slash. For domain skills, describe your situation and suggest-best-practice routes automatically.

Tool names: Use native Claude Code tool names (Read, Write, Edit, Bash, Skill, TodoWrite).

GitHub Copilot CLI

Install:

copilot plugin marketplace add jeffreytse/grimoire-core
copilot plugin install grimoire@grimoire-core

Plugin files: .claude-plugin/plugin.json and .claude-plugin/marketplace.json (Copilot CLI checks .claude-plugin/ automatically — shared with Claude Code)

Skills are namespaced by plugin name, e.g. grimoire-engineering:propose-conventional-commit.

Using skills: Describe your task. Copilot CLI activates the matching skill based on its description triggering conditions.

Codex CLI

Install:

grimoire install --target codex

Codex CLI supports AGENTS.md natively — grimoire’s AGENTS.md is auto-loaded at session start, providing skill path conventions and domain routing. For the full skill library, run grimoire to copy skills to the Codex skills directory.

Using skills: Codex CLI has an interactive plugin browser: run codex then type /plugins to browse and install marketplace plugins. For grimoire skills installed via grimoire, describe your task and Codex routes to the matching skill via AGENTS.md context.

Cursor

Install:

grimoire install --target cursor

No official Cursor CLI plugin install command exists. Skills installed via grimoire are available through AGENTS.md context injection. Cursor reads AGENTS.md from the project root automatically in Agent mode.

Using skills: Describe your situation in the Cursor chat or Composer panel. Skills activate when the task matches a skill’s triggering conditions.

OpenCode

Install: Add to your opencode.json (global: ~/.config/opencode/opencode.json, or project-level):

{
  "plugin": ["grimoire@git+https://github.com/jeffreytse/grimoire-core.git"]
}

Restart OpenCode. The plugin registers all grimoire skills automatically.

Plugin file: .opencode/plugins/grimoire.js (ESM module)

The plugin:

1. Scans ./skills/ recursively for SKILL.md files
2. Injects AGENTS.md as bootstrap context
3. Maps Claude Code tool references to OpenCode native equivalents

Using skills:

use skill tool to load engineering/development/propose-conventional-commit
use skill tool to list skills

Tool mapping:

Pinning a version:

{
  "plugin": [
    "grimoire@git+https://github.com/jeffreytse/grimoire-core.git#v1.0.0"
  ]
}

Troubleshooting — skills not found:

opencode run --print-logs "hello" 2>&1 | grep -i grimoire

Gemini CLI

Note: As of June 18, 2026, free/individual Gemini CLI access (Google AI Pro/Ultra, free Gemini Code Assist for individuals) was retired in favor of Antigravity CLI — extensions carry over as Antigravity plugins, without full feature parity yet. Paid Gemini Code Assist Standard/Enterprise and Google Cloud users are unaffected and keep Gemini CLI access via paid API keys. See the migration announcement for details.

Install:

gemini extensions install https://github.com/jeffreytse/grimoire-core

Plugin files: gemini-extension.json (extension metadata) + GEMINI.md (context injected at session start)

gemini-extension.json points to GEMINI.md via the contextFileName field. Gemini CLI loads GEMINI.md into context at startup, giving the agent the skill path convention and domain list.

Using skills: Gemini CLI loads skill metadata at session start. Activate a skill by describing your task — the agent matches to the closest skill. For explicit activation:

activate skill: engineering/development/propose-conventional-commit

Key difference from Claude Code: Gemini CLI uses activate_skill rather than slash commands. GEMINI.md provides the routing bootstrap; individual skills are activated on demand from their SKILL.md files.

Tool mapping: See GEMINI.md in the repo root for Gemini-specific tool references.

Antigravity CLI

Google’s successor to Gemini CLI (see the note above) — grimoire treats it as its own target, binary name agy.

Install:

grimoire install --target antigravity

Plugin file: ~/.gemini/AGENTS.md — same bootstrap-context mechanism as other AGENTS.md-reading agents.

Skills install to ~/.gemini/antigravity-cli/skills (global) or .gemini/antigravity-cli/skills (project-scoped) — auto-detected by grimoire install like any other agent.

OpenClaw

Install:

grimoire install --target openclaw

See .openclaw/INSTALL.md for plugin configuration.

Other agents

Any agent that reads AGENTS.md from the project root gets grimoire’s bootstrap context automatically — skill path convention, domain list, and routing instructions.

Install:

grimoire install --target all

Adding grimoire to a new agent

If you’re integrating grimoire with an agent not listed here:

1. Check for a plugin format. Most agents use either a JSON manifest (plugin.json) or a context file (markdown loaded at startup).

2. Point to ./skills/ — all SKILL.md files live under this path in a consistent structure (<domain>/<subdomain>/skills/<name>/SKILL.md).

3. Inject bootstrap context — copy AGENTS.md or GEMINI.md content into whatever context injection mechanism your agent supports. This gives the agent the skill path convention and domain list.

4. Map tool names — SKILL.md files reference Claude Code tool names (Read, Write, Edit, Bash, Skill, TodoWrite). Document the equivalents for your agent so skills can be adapted.

5. Open a PR — add the new agent’s plugin file(s) to the repo and update the supported agents table in this document.

Keeping plugins in sync

When a new domain is added to grimoire (via design-best-practice-domain), the following files must be updated:

.opencode/plugins/grimoire.js discovers skills from ./skills/ recursively — no update needed when adding domains.