Skip to content
Back to skills

Openclaw Native

ASecurity

Native integration with OpenClaw platform — auto-detection, skill sync, bidirectional hooks, agent profiles, memory sync

  • 6 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added May 27, 2026
data-aibashtestingcode-reviewfrontend

Works with

  • claude code
  • cli

Security analysis

A100/100

Scanned May 27, 2026

npx -y skills add KevinZai/commander --skill openclaw-native --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Openclaw Native?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Openclaw Native
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/kevinzai-openclaw-native/badge)](https://www.skillsdirectory.com/skills/kevinzai-openclaw-native)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: openclaw-native
description: Native integration with OpenClaw platform — auto-detection, skill sync, bidirectional hooks, agent profiles, memory sync
triggers:
  - "openclaw"
  - "sync with openclaw"
  - "openclaw integration"
  - "connect openclaw"
  - "openclaw status"
  - "register with openclaw"
model: sonnet
disable-model-invocation: false
---

# OpenClaw Native Integration

First-class integration between CC Commander and the OpenClaw AI agent platform. Automatically syncs skills, hooks, and agent profiles between both systems.

## Overview

OpenClaw is a personal AI assistant platform that orchestrates multiple agents across channels (Discord, Slack, Telegram, etc.) from a single gateway. This integration makes CC Commander a native extension of OpenClaw — not just a bridge.

## Auto-Detection

The `openclaw-sync.js` SessionStart hook automatically detects OpenClaw:

1. Check `CC_OPENCLAW_ENABLED=1` env var
2. Probe gateway at `http://localhost:18789/health`
3. If healthy: sync skills, register hooks, report session
4. If not found: exit silently (no errors, no delays)

```bash
# Enable OpenClaw integration
export CC_OPENCLAW_ENABLED=1

# Custom gateway URL (if non-default)
export CC_OPENCLAW_URL=http://localhost:18789
```

## Skill Sync

On session start, all Kit skills are registered with OpenClaw's skill registry:

```
Syncing: 450+ skills → OpenClaw registry
  ├── Core workflow skills (confidence-check, four-question-validation, etc.)
  ├── 11 CCC domains (ccc-seo, ccc-design, ccc-testing, etc.)
  ├── Integration skills (openclaw-native, status-updates, etc.)
  └── Mode switcher (9 workflow modes)
```

OpenClaw agents can then invoke Kit skills by name, enabling:
- Alfred using `/code-review` from Discord
- Morpheus running `/plan` for architecture decisions
- Neo orchestrating `/tdd` across worker agents

## Hook Integration

Bidirectional event forwarding between Kit hooks and OpenClaw webhooks:

### Kit → OpenClaw Events
| Kit Hook | OpenClaw Event | Data |
|----------|---------------|------|
| PostToolUse | `cc_kit_tool_use` | tool name, duration, result |
| Stop | `cc_kit_session_end` | session summary, cost, files changed |
| cost-alert | `cc_kit_cost_alert` | current cost, ceiling, % used |
| context-guard | `cc_kit_context_warn` | context %, tokens remaining |

### OpenClaw → Kit Events
| OpenClaw Event | Kit Action | Description |
|---------------|------------|-------------|
| `agent_task_assigned` | Load context | Pre-load relevant files |
| `agent_mode_change` | Switch mode | Apply workflow mode |
| `channel_message` | Status update | Forward to status-updates skill |

## Agent Profiles

Generate OpenClaw agent profiles from Kit workflow modes:

```bash
# From Kit mode → OpenClaw agent profile
/openclaw agents generate

# Creates profiles for:
# - normal_mode_agent (balanced, standard workflow)
# - design_mode_agent (frontend-focused, design system aware)
# - saas_mode_agent (full-stack, billing/auth/deploy)
# - research_mode_agent (deep analysis, multi-source)
# - yolo_mode_agent (fast, minimal verification)
```

## Memory Sync

Bridge Kit session memory with OpenClaw agent memory:

- Kit sessions saved to `~/.claude/sessions/` sync to OpenClaw's `~/.openclaw/memory/`
- OpenClaw agent learnings feed back into Kit's `tasks/lessons.md`
- Shared memory enables cross-agent knowledge transfer

## Health Monitoring

The `/openclaw health` command checks:
- Gateway status (up/down, uptime, version)
- Channel connectivity (Discord, Slack, Telegram, etc.)
- Agent status (active, idle, error)
- Skill registry (synced count, stale entries)
- Memory DB size and health
- Recent error log

## CLI Commands

See `/openclaw` command for:
- `/openclaw status` — Gateway and agent overview
- `/openclaw sync` — Force skill + hook sync
- `/openclaw configure` — Set up integration
- `/openclaw health` — Detailed health check
- `/openclaw agents` — List and manage agents

## Configuration

| Setting | Env Var | Default | Description |
|---------|---------|---------|-------------|
| Enable | `CC_OPENCLAW_ENABLED` | `0` | Set to `1` to enable |
| Gateway URL | `CC_OPENCLAW_URL` | `http://localhost:18789` | Gateway endpoint |
| Timeout | `CC_OPENCLAW_TIMEOUT` | `5000` | Request timeout (ms) |
| Auto-sync | `CC_OPENCLAW_AUTOSYNC` | `1` | Sync on session start |
| Debug | `CC_OPENCLAW_DEBUG` | `0` | Verbose logging |

## Architecture

```
CC Commander                    OpenClaw Gateway
┌──────────────────┐              ┌──────────────────┐
│  Skills (280+)   │──── sync ───→│  Skill Registry  │
│  Hooks (18)      │←── events ──→│  Webhook System  │
│  Modes (9)       │──── sync ───→│  Agent Profiles  │
│  Sessions        │←── sync ───→│  Memory DBs      │
└──────────────────┘              └──────────────────┘
         ↕                                 ↕
    Claude Code CLI              38 AI Agents (12 workspaces)
```

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.