How to Install Agent Reach Skill Files for OpenClaw and Claude Code

Run agent-reach skill --install to automatically detect OpenClaw, Claude Code, or generic agent directories and deploy the appropriate skill documentation files.

The Panniantong/Agent-Reach repository provides a CLI tool that automates the deployment of structured documentation—known as skills—to AI agent environments. When you need to install Agent Reach skill files for specific integrations like OpenClaw or Claude Code, the tool handles directory detection, locale selection, and file copying automatically. This guide explains the installation mechanics based on the actual source code implementation.

How the Skill Installer Works

The installation process centers on the _install_skill() helper function defined in agent_reach/cli.py. When you execute the install command, the CLI performs a multi-step detection and copy operation without modifying any upstream tool source code.

Detecting Target Skill Directories

The installer first identifies which agent environments exist on your system by checking three standard locations (lines 12-16):

• ~/.agents/skills – Generic agents (highest priority)
• ~/.openclaw/skills – OpenClaw CLI integration
• ~/.claude/skills – Claude Code integration

If you have set the OPENCLAW_HOME environment variable, the installer prepends that custom path to the search list (lines 19-22).

Selecting the Correct Locale

Before copying files, the installer checks your locale settings. It examines the environment variables AGENT_REACH_LANG, LC_ALL, LC_MESSAGES, and LANG in sequence (lines 49-62). If any value starts with en or english, the installer loads SKILL_en.md; otherwise it falls back to the default SKILL.md.

Copying and Installing Files

The _copy_skill_dir() function (lines 71-107) handles the actual file operations:

1. Removes any previous installation at the target location.
2. Creates the target directory structure.
3. Writes the selected skill markdown file.
4. Copies the references/ subdirectory containing supplemental documentation.

The main installation loop (lines 124-132) iterates through the candidate directories and installs the skill into the first existing directory it finds, printing a confirmation message like:

Skill installed for OpenClaw: /home/you/.openclaw/skills/agent-reach

If none of the agent directories exist, the installer creates ~/.agents/skills/agent-reach as a fallback (lines 134-140).

Installing Skills for Specific Integrations

Use the following commands to manage skill files for your specific agent setup.

Install for All Detected Agents

Run the default install command to deploy skills wherever Agent Reach finds existing agent directories:

agent-reach skill --install

Install During Full Setup

The full installation command automatically includes the skill step:

agent-reach install --env=auto

Targeting OpenClaw with Custom Paths

If you only use OpenClaw, the same command works—the installer automatically detects ~/.openclaw/skills and copies the files there. To ensure OpenClaw uses a custom location, set the environment variable before running:

export OPENCLAW_HOME=/custom/path
agent-reach skill --install

Removing Skill Files

To uninstall skills from all locations, run:

agent-reach skill --uninstall

This executes the _uninstall_skill() function (lines 444-452 in agent_reach/cli.py), which deletes the agent-reach folder from each known skill directory without affecting other agent configurations.

Summary

• The _install_skill() function in agent_reach/cli.py automatically detects OpenClaw, Claude Code, and generic agent directories at ~/.openclaw/skills, ~/.claude/skills, and ~/.agents/skills.
• Locale detection via AGENT_REACH_LANG, LC_ALL, LC_MESSAGES, and LANG determines whether to install SKILL_en.md or the default SKILL.md.
• The installer copies files from the package's agent_reach/skill/ directory, including the references/ subdirectory, into the first existing agent directory or falls back to ~/.agents/skills/agent-reach.
• Use agent-reach skill --install to deploy and agent-reach skill --uninstall to remove skills without modifying upstream tool code.

Frequently Asked Questions

What files does Agent Reach actually install?

Agent Reach installs markdown documentation files that teach AI agents how to invoke the tool. Specifically, it copies either SKILL_en.md (for English locales) or SKILL.md (default) from the agent_reach/skill/ package directory, along with any files in the references/ subdirectory. These are documentation files, not executable code, and reside in the agent's skills directory so the AI can read them.

Can I install skills to a custom OpenClaw directory?

Yes. While the installer automatically checks ~/.openclaw/skills, you can override this by setting the OPENCLAW_HOME environment variable. When this variable is set, the installer uses $OPENCLAW_HOME/skills as the target location, inserting it at the front of the candidate directory list before running agent-reach skill --install.

Why does the installer choose ~/.agents/skills when I have multiple agents installed?

The installer iterates through the list of candidate directories and installs into the first existing directory it finds. If you have multiple agent directories (e.g., both OpenClaw and Claude Code), it installs to whichever directory appears first in its internal priority list. If you need to install to a specific directory that isn't being selected, ensure that directory exists before running the command, or temporarily remove other agent directories to force the fallback behavior.

How do I verify the skill files were installed correctly?

After running agent-reach skill --install, check for the confirmation message indicating the target directory. For OpenClaw, verify that ~/.openclaw/skills/agent-reach/SKILL.md (or SKILL_en.md) exists. For Claude Code, check ~/.claude/skills/agent-reach/. The installer also copies the references/ folder, so ensure that subdirectory exists alongside the main skill file.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}