How to Register Agent Reach as an Agent Skill for OpenClaw: Complete Setup Guide

Register Agent Reach as an agent skill for OpenClaw by running agent-reach skill --install, which copies the bundled SKILL.md file and reference documents into ~/.openclaw/skills/agent-reach/.

The Panniantong/Agent-Reach repository provides a ready-to-use skill package that enables OpenClaw (and compatible agent frameworks) to discover and invoke Agent Reach commands automatically. This registration process requires no manual file manipulation; the built-in CLI handles detection, locale selection, and installation across multiple standard directories.

What Is the Agent Reach Skill Package?

Agent Reach ships with a self-contained skill definition located in agent_reach/skill/. This package contains:

• SKILL.md — The core skill definition that OpenClaw parses to understand available commands (e.g., "search Twitter" or "watch YouTube video")
• references/ — A subdirectory of auxiliary markdown files referenced by SKILL.md for detailed usage instructions
• Locale variants — Support for localized versions such as SKILL_en.md when available

According to the source code in agent_reach/cli.py, the private helper _install_skill() manages the entire registration workflow by loading these resources via importlib.resources and writing them to the appropriate filesystem location.

Installation Methods

You can register the skill either automatically during the main Agent Reach installation or manually at any time afterward.

Method 1: Automatic Installation During Setup

When you run the environment-aware installer, it executes the skill registration as the final step:

agent-reach install --env=auto

This command completes the full Agent Reach setup and then internally triggers _install_skill() to copy the skill files into the first writable directory it discovers.

Method 2: Manual Skill Registration

If you already have Agent Reach installed but skipped the skill step, register it explicitly:

agent-reach skill --install

This invokes _install_skill() directly, which performs four critical operations:

1. Locale Detection — Identifies the appropriate skill file (SKILL.md or locale-specific variants like SKILL_en.md) at line 333 of agent_reach/cli.py
2. Resource Loading — Uses importlib.resources to access the package's internal skill directory (line 359)
3. File Deployment — Writes the selected SKILL.md into the target folder and recursively copies the references/ subdirectory (line 368)
4. Directory Resolution — Locates the first writable path from the prioritized list

Target Directory Resolution

As implemented in agent_reach/cli.py around line 389, the CLI searches for skill directories in this strict priority order:

• ~/.agents/skills (generic agents folder)
• ~/.openclaw/skills (OpenClaw specific)
• ~/.claude/skills (Claude Code)
• $OPENCLAW_HOME/.openclaw/skills (if the OPENCLAW_HOME environment variable is set)

If none of these directories exist, the CLI creates ~/.agents/skills/agent-reach as a fallback location (line 1110). After successful installation, the CLI prints a confirmation message indicating the skill is installed for OpenClaw.

Verifying the Installation

Confirm that OpenClaw can discover the skill by checking for the presence of the core definition file:

ls ~/.openclaw/skills/agent-reach/SKILL.md

You should also verify the references/ subdirectory exists alongside it:

ls ~/.openclaw/skills/agent-reach/references/

When these files are present, OpenClaw automatically parses SKILL.md on startup and registers Agent Reach's command suite without requiring additional configuration.

Removing the Skill

To unregister Agent Reach from OpenClaw and remove all associated files from every known skill directory, run:

agent-reach skill --uninstall

This executes _uninstall_skill() (defined at line 421 of agent_reach/cli.py), which deletes the agent-reach folder from ~/.agents/skills, ~/.openclaw/skills, ~/.claude/skills, and any other detected locations.

Summary

• Single Command Registration: Use agent-reach skill --install to register Agent Reach as an OpenClaw skill instantly.
• Automatic Discovery: The CLI detects the correct locale, loads resources via importlib.resources, and writes to ~/.openclaw/skills/ or the first available agent directory.
• No Manual Configuration: Once SKILL.md is copied to the target folder, OpenClaw automatically recognizes the available commands.
• Clean Removal: The skill --uninstall command reverses the process by deleting the agent-reach folder from all standard skill locations.

Frequently Asked Questions

Where does Agent Reach install the skill files by default?

The CLI prioritizes OpenClaw-specific directories, installing to ~/.openclaw/skills/agent-reach/ when available. If that folder does not exist, it falls back to ~/.agents/skills/agent-reach/ or creates this directory if no standard locations are found. You can override the root path by setting the OPENCLAW_HOME environment variable before running the install command.

Can I install the skill without installing the full Agent Reach package?

No, the agent-reach skill --install command requires the Agent Reach CLI to be installed first, as it relies on the bundled skill resources inside the package. However, you can install Agent Reach minimally and then immediately run the skill command to complete only the OpenClaw registration step.

What happens if I have multiple agent frameworks installed?

The _install_skill() function writes to the first writable directory in its search order, meaning OpenClaw's folder (~/.openclaw/skills) takes precedence over generic or Claude Code locations. If you need the skill available in multiple frameworks simultaneously, you can manually copy the agent-reach folder from the first installation target to the others.

How do I update the skill to a new version?

Run agent-reach skill --uninstall first to remove the old version from all directories, then run agent-reach skill --install again. This ensures that _uninstall_skill() (line 421) cleans up legacy references before the new SKILL.md and documentation are written to the target folder.

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