OpenClaw Customizer - Setup Guide¶

Prerequisites¶

Git installed
Claude Code and/or OpenAI Codex installed
Bash shell

Capabilities¶

Capability	Description
One-Command Setup	`./setup.sh` symlinks everything into place
Skill Syncing	Symlink skills from Git repo into `~/.claude/skills/` and `~/.codex/skills/`
Shared MCP Config	`.mcp.json` symlinked for consistent MCP server access
Codex Exclusion List	`codex-exclude` file filters skills not relevant to Codex
Cross-Machine Sync	Git push/pull to keep agent configs identical across machines
Auto-Sync on Session Start	Codex skills auto-synced via `sync-codex-skills.sh` on every session

Installation¶

git clone https://github.com/petekp/claude-code-setup.git
cd claude-code-setup
./setup.sh

How It Works¶

The setup script creates this symlink structure:

~/.claude/
├── skills/         → <repo>/skills/
├── scripts/        → <repo>/scripts/
├── settings.json   → <repo>/settings.json
├── statusline-command.sh → <repo>/ (copied)

~/.mcp.json         → <repo>/.mcp.json

~/.codex/skills/
├── skill-a/        → <repo>/skills/skill-a/  (symlinked individually)
├── skill-b/        → <repo>/skills/skill-b/
└── ...             (only skills not in codex-exclude)

Edit files in either location - they're the same files. Commit and push to sync across machines.

Codex Skill Syncing¶

Codex doesn't support directory-level symlinks, so skills are symlinked individually:

# Runs automatically on every Claude Code session start
scripts/sync-codex-skills.sh

Skills listed in codex-exclude are skipped - use this for skills that only make sense in Claude Code.

Configuration Files¶

settings.json¶

{
  "model": "claude-sonnet-4-20250514",
  "maxTokens": 4096,
  "temperature": 0.7,
  "permissions": {
    "allow": [
      "Bash(*)",
      "Read(*)",
      "Write(*)",
      "Edit(*)"
    ]
  }
}

.mcp.json¶

Shared MCP server configuration - add CorpusIQ MCP endpoint here:

{
  "mcpServers": {
    "corpusiq": {
      "command": "npx",
      "args": ["-y", "@corpusiq/mcp-server"],
      "env": {
        "CORPUSIQ_API_KEY": "${CORPUSIQ_API_KEY}"
      }
    }
  }
}

CorpusIQ Use Cases¶

Multi-Machine Agent Consistency: Agents running on both Linux and macOS hosts. A shared Git-backed config ensures both machines use identical skills, MCP servers, and settings.
Team Onboarding: New team members clone the config repo, run ./setup.sh, and instantly have the full CorpusIQ-connected agent toolkit - no manual skill installation needed.
Skill Versioning: Skills change rapidly. Git tracking means you can roll back to a known-good skill version if an update breaks something.
Selective Codex Deployment: Not all Claude Code skills work in Codex. The codex-exclude list prevents broken skills from being loaded.

Troubleshooting¶

Symptom	Likely Cause	Fix
Symlinks broken after OS update	macOS SIP restrictions	Re-run `./setup.sh`
Codex skills not updating	`sync-codex-skills.sh` not running	Check `~/.claude/settings.json` for `statusline-command`
Git conflicts on `settings.json`	Multiple machines editing same file	Use machine-specific settings branches or `.gitignore` patterns
"Permission denied" on setup	Script not executable	`chmod +x setup.sh`

← Skills Catalog | Marketplace → ↑ Skills Home

Curated by CorpusIQ - one MCP endpoint, all your business tools.

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects