cc-openclaw

内容来源：README.md（说明文档） · 原始地址 · 查看安装指南

原始内容

# cc-openclaw

Claude Code skills for managing your [OpenClaw](https://openclaw.ai) deployment. These slash commands let you scaffold agents, configure channels, set up memory systems, manage secrets, and operate your gateway — all from within Claude Code.

## What Are Claude Code Skills?

[Claude Code skills](https://docs.anthropic.com/en/docs/claude-code/skills) are reusable playbooks that Claude Code auto-discovers from `.claude/skills/` in your project. Each skill becomes a `/slash-command` you can invoke during a session. They encode best practices, configuration patterns, and operational procedures so you don't have to remember them.

## Quick Start

Clone the repo and use [GNU Stow](https://www.gnu.org/software/stow/) to symlink the skills into your openclaw-home project:

```bash
git clone https://github.com/rahulsub-be/cc-openclaw.git ~/cc-openclaw
cd ~/cc-openclaw
stow --no-folding -t ~/your-openclaw-home-repo .
```

This creates symlinks from your openclaw-home's `.claude/skills/openclaw-*/` pointing into the cc-openclaw repo. Only the `.claude/` directory is stowed — `README.md`, `LICENSE`, and other repo files are excluded via `.stow-local-ignore`.

Open Claude Code in your openclaw-home directory and type `/openclaw-` to see all nine skills in autocomplete.

### Updating

When skills get updates, pull the latest — symlinks mean changes take effect immediately:

```bash
cd ~/cc-openclaw
git pull
```

If new skills were added, re-run stow to create the new symlinks:

```bash
cd ~/cc-openclaw
stow --no-folding -t ~/your-openclaw-home-repo .
```

This is the same Git+Stow pattern used by OpenClaw itself for agent configuration.

## Available Skills

| Skill | Command | Description |
|-------|---------|-------------|
| **New Agent** | `/openclaw-new-agent <id> <name>` | Scaffold a complete agent with directory structure, 6 core .md files, and openclaw.json registration |
| **Add Channel** | `/openclaw-add-channel <agent-id> <type>` | Connect Telegram, Slack, WhatsApp, or GChat to an agent with keychain secrets |
| **Add Cron Job** | `/openclaw-add-cron <agent-id>` | Create scheduled tasks — recurring (cron), interval (every), or one-shot (at) |
| **Status Check** | `/openclaw-status` | Health dashboard: gateway, channels, agents, cron jobs, and recent errors |
| **Restart** | `/openclaw-restart` | Restart the gateway with automatic stow conflict resolution |
| **Stow Deploy** | `/openclaw-stow` | Re-deploy config via GNU Stow, resolving the known jobs.json conflict |
| **Add Secret** | `/openclaw-add-secret <service> <ENV_VAR>` | Store a secret in macOS Keychain and wire it into the secrets pipeline |
| **Add Script** | `/openclaw-add-script <agent-id> <name>` | Scaffold a deterministic shell script with JSON output and error handling |
| **Dream Setup** | `/openclaw-dream-setup <agent-id>` | Configure nightly memory distillation with cron job and archive pipeline |

## Skills in Detail

### `/openclaw-new-agent`

Creates a fully-configured agent from scratch:
- Directory structure (`memory/`, `scripts/`, `sessions/`, etc.)
- Six core directive files: `SOUL.md`, `IDENTITY.md`, `USER.md`, `AGENTS.md`, `TOOLS.md`, `SECURITY.md`
- Registration in `openclaw.json`
- Sub-agent wiring (if applicable)
- Stow deployment and gateway restart

### `/openclaw-add-channel`

Handles the full channel setup pipeline per platform:
- **Telegram**: Bot token via BotFather, keychain storage, account + binding config
- **Slack**: Bot token + app token, socket mode setup
- **WhatsApp**: Phone number allowlist
- **GChat**: Binding configuration

All secrets go to macOS Keychain — never written to files.

### `/openclaw-add-cron`

Interactive cron job builder supporting:
- **cron**: Standard cron expressions with timezone (`0 9 * * *`)
- **every**: Fixed intervals (`30m`, `1h`)
- **at**: One-shot execution at a specific time
- Agent turn payloads with configurable models and timeouts
- Isolated or main session targeting

### `/openclaw-status`

One-command health check covering:
- Gateway process (health endpoint, launchd status)
- Channel connectivity (Telegram bots, Slack socket, WhatsApp)
- Agent count and bindings
- Cron job results (last run, errors, next scheduled)
- Recent error log entries

### `/openclaw-restart`

Handles the restart correctly by:
1. Resolving the known `jobs.json` stow conflict (gateway overwrites the symlink)
2. Re-stowing config files
3. Restarting via `launchctl kickstart`
4. Verifying all channels reconnect

### `/openclaw-stow`

Deploys config from your openclaw-home repo to `$HOME` via GNU Stow:
- Detects and resolves known conflicts (jobs.json)
- Verifies key symlinks after deployment
- Reports any unresolved conflicts

### `/openclaw-add-secret`

Full secrets pipeline:
1. Stores value in macOS Keychain (`openclaw.keychain-db`)
2. Updates `openclaw-secrets.sh` (for launchd/gateway)
3. Updates `openclaw-env.sh` (for shell sessions)
4. Updates `secrets.sh` (for provisioning on new machines)

### `/openclaw-add-script`

Scaffolds deterministic shell scripts following production patterns:
- `set -euo pipefail` with structured error handling
- `json-response.sh` shared library (log, json_success, json_error)
- stdout is sacred: only JSON output
- stderr for logging via `log()` function
- Automatic TOOLS.md documentation

### `/openclaw-dream-setup`

Configures nightly memory distillation:
- `DREAM-ROUTINE.md` with distillation spec and token budgets
- Cron job for nightly execution
- QMD memory index paths in openclaw.json
- Session startup sequence in AGENTS.md
- Archive pipeline for old distillations

Token budgets: 2,500/day per distillation, 7,500 rolling 3-day digest.

## How It Works

All skills use automatic repo detection — no hardcoded paths:

```bash
OPENCLAW_REPO=$(readlink ~/.openclaw/openclaw.json 2>/dev/null | sed 's|/.openclaw/openclaw.json||')
```

This resolves your openclaw-home repo location by following the `openclaw.json` symlink that GNU Stow creates. Skills work regardless of where you've cloned your repo.

## Prerequisites

- [OpenClaw](https://openclaw.ai) installed and configured
- [Claude Code](https://docs.anthropic.com/en/docs/claude-code) CLI
- [GNU Stow](https://www.gnu.org/software/stow/) (`brew install stow`)
- macOS (skills use `security` CLI for Keychain, `launchctl` for service management)

## Companion Playbooks

These skills are designed to work alongside the [OpenClaw documentation playbooks](https://openclaw.ai/docs) covering:

1. First Agent Setup
2. Memory System (QMD + Dream Routines)
3. Deterministic Scripts
4. Sub-Agent Architecture
5. Cron System
6. Security Model
7. Channel Configuration
8. Stow-Based Deployment
9. Monitor-DevBot Pattern

Each skill references its corresponding playbook for deeper context.

## Contributing

PRs welcome. When adding a new skill:

1. Create `.claude/skills/<skill-name>/SKILL.md`
2. Use YAML frontmatter: `name`, `description`, `argument-hint`, `disable-model-invocation: true`
3. Use `$OPENCLAW_REPO` detection — no hardcoded paths
4. Follow the step-by-step format with bash code blocks
5. Include verification steps
6. Update this README

## License

MIT

cc-openclaw

Claude Code skills for managing your OpenClaw deployment. These slash commands let you scaffold agents, configure channels, set up memory systems, manage secrets, and operate your gateway — all from within Claude Code.

What Are Claude Code Skills?

Claude Code skills are reusable playbooks that Claude Code auto-discovers from .claude/skills/ in your project. Each skill becomes a /slash-command you can invoke during a session. They encode best practices, configuration patterns, and operational procedures so you don't have to remember them.

Quick Start

Clone the repo and use GNU Stow to symlink the skills into your openclaw-home project:

git clone https://github.com/rahulsub-be/cc-openclaw.git ~/cc-openclaw
cd ~/cc-openclaw
stow --no-folding -t ~/your-openclaw-home-repo .

This creates symlinks from your openclaw-home's .claude/skills/openclaw-*/ pointing into the cc-openclaw repo. Only the .claude/ directory is stowed — README.md, LICENSE, and other repo files are excluded via .stow-local-ignore.

Open Claude Code in your openclaw-home directory and type /openclaw- to see all nine skills in autocomplete.

Updating

When skills get updates, pull the latest — symlinks mean changes take effect immediately:

cd ~/cc-openclaw
git pull

If new skills were added, re-run stow to create the new symlinks:

cd ~/cc-openclaw
stow --no-folding -t ~/your-openclaw-home-repo .

This is the same Git+Stow pattern used by OpenClaw itself for agent configuration.

Available Skills

Skill	Command	Description
New Agent	`/openclaw-new-agent <id> <name>`	Scaffold a complete agent with directory structure, 6 core .md files, and openclaw.json registration
Add Channel	`/openclaw-add-channel <agent-id> <type>`	Connect Telegram, Slack, WhatsApp, or GChat to an agent with keychain secrets
Add Cron Job	`/openclaw-add-cron <agent-id>`	Create scheduled tasks — recurring (cron), interval (every), or one-shot (at)
Status Check	`/openclaw-status`	Health dashboard: gateway, channels, agents, cron jobs, and recent errors
Restart	`/openclaw-restart`	Restart the gateway with automatic stow conflict resolution
Stow Deploy	`/openclaw-stow`	Re-deploy config via GNU Stow, resolving the known jobs.json conflict
Add Secret	`/openclaw-add-secret <service> <ENV_VAR>`	Store a secret in macOS Keychain and wire it into the secrets pipeline
Add Script	`/openclaw-add-script <agent-id> <name>`	Scaffold a deterministic shell script with JSON output and error handling
Dream Setup	`/openclaw-dream-setup <agent-id>`	Configure nightly memory distillation with cron job and archive pipeline

Skills in Detail

`/openclaw-new-agent`

Creates a fully-configured agent from scratch:

Directory structure (memory/, scripts/, sessions/, etc.)
Six core directive files: SOUL.md, IDENTITY.md, USER.md, AGENTS.md, TOOLS.md, SECURITY.md
Registration in openclaw.json
Sub-agent wiring (if applicable)
Stow deployment and gateway restart

`/openclaw-add-channel`

Handles the full channel setup pipeline per platform:

Telegram: Bot token via BotFather, keychain storage, account + binding config
Slack: Bot token + app token, socket mode setup
WhatsApp: Phone number allowlist
GChat: Binding configuration

All secrets go to macOS Keychain — never written to files.

`/openclaw-add-cron`

Interactive cron job builder supporting:

cron: Standard cron expressions with timezone (0 9 * * *)
every: Fixed intervals (30m, 1h)
at: One-shot execution at a specific time
Agent turn payloads with configurable models and timeouts
Isolated or main session targeting

`/openclaw-status`

One-command health check covering:

Gateway process (health endpoint, launchd status)
Channel connectivity (Telegram bots, Slack socket, WhatsApp)
Agent count and bindings
Cron job results (last run, errors, next scheduled)
Recent error log entries

`/openclaw-restart`

Handles the restart correctly by:

Resolving the known jobs.json stow conflict (gateway overwrites the symlink)
Re-stowing config files
Restarting via launchctl kickstart
Verifying all channels reconnect

`/openclaw-stow`

Deploys config from your openclaw-home repo to $HOME via GNU Stow:

Detects and resolves known conflicts (jobs.json)
Verifies key symlinks after deployment
Reports any unresolved conflicts

`/openclaw-add-secret`

Full secrets pipeline:

Stores value in macOS Keychain (openclaw.keychain-db)
Updates openclaw-secrets.sh (for launchd/gateway)
Updates openclaw-env.sh (for shell sessions)
Updates secrets.sh (for provisioning on new machines)

`/openclaw-add-script`

Scaffolds deterministic shell scripts following production patterns:

set -euo pipefail with structured error handling
json-response.sh shared library (log, json_success, json_error)
stdout is sacred: only JSON output
stderr for logging via log() function
Automatic TOOLS.md documentation

`/openclaw-dream-setup`

Configures nightly memory distillation:

DREAM-ROUTINE.md with distillation spec and token budgets
Cron job for nightly execution
QMD memory index paths in openclaw.json
Session startup sequence in AGENTS.md
Archive pipeline for old distillations

Token budgets: 2,500/day per distillation, 7,500 rolling 3-day digest.

How It Works

All skills use automatic repo detection — no hardcoded paths:

OPENCLAW_REPO=$(readlink ~/.openclaw/openclaw.json 2>/dev/null | sed 's|/.openclaw/openclaw.json||')

This resolves your openclaw-home repo location by following the openclaw.json symlink that GNU Stow creates. Skills work regardless of where you've cloned your repo.

Prerequisites

OpenClaw installed and configured
Claude Code CLI
GNU Stow (brew install stow)
macOS (skills use security CLI for Keychain, launchctl for service management)

Companion Playbooks

These skills are designed to work alongside the OpenClaw documentation playbooks covering:

First Agent Setup
Memory System (QMD + Dream Routines)
Deterministic Scripts
Sub-Agent Architecture
Cron System
Security Model
Channel Configuration
Stow-Based Deployment
Monitor-DevBot Pattern

Each skill references its corresponding playbook for deeper context.

Contributing

PRs welcome. When adding a new skill:

Create .claude/skills/<skill-name>/SKILL.md
Use YAML frontmatter: name, description, argument-hint, disable-model-invocation: true
Use $OPENCLAW_REPO detection — no hardcoded paths
Follow the step-by-step format with bash code blocks
Include verification steps
Update this README

License

MIT