\r \r

Agentar Skill Overview\r

\r Fast CLI to distribute an AI agent.\r \r

CLI Reference\r

\r

Save (Export)\r

\r

node {baseDir}/scripts/clawctl.mjs save \x3Cnamespace/name:tag> [options]\r
```\r
\r
| Flag | Description | Default |\r
|------|-------------|---------|\r
| `-o \x3Cpath>` | Output file path | `./\x3Cname>-\x3Ctag>.claw` |\r
| `--source \x3Cpath>` | OpenClaw state directory | `~/.openclaw` |\r
| `--description \x3Ctext>` | Package description | (empty) |\r
| `--include-memory` | Include MEMORY.md | false |\r
\r
### Load (Import)\r
\r
```bash\r
node {baseDir}/scripts/clawctl.mjs load \x3Cfile> [options]\r
```\r
\r
Load **automatically creates a backup** before installing. This can be skipped with `--no-backup`.\r
\r
| Flag | Description | Default |\r
|------|-------------|---------|\r
| `--target \x3Cpath>` | Target OpenClaw state dir | `~/.openclaw` |\r
| `--agent-name \x3Cname>` | Name for the imported agent | from manifest |\r
| `--dry-run` | Preview without installing | false |\r
| `--no-backup` | Skip auto-backup | false |\r
\r
### Backup\r
\r
```bash\r
node {baseDir}/scripts/clawctl.mjs backup [--source \x3Cpath>] [--label \x3Ctext>]\r
```\r
\r
Create a snapshot of current workspace + config. Stored in `~/.openclaw/.agentar-backups/\x3Cid>/`.\r
\r
### List Backups\r
\r
```bash\r
node {baseDir}/scripts/clawctl.mjs backups [--source \x3Cpath>]\r
```\r
\r
### Rollback\r
\r
```bash\r
node {baseDir}/scripts/clawctl.mjs rollback [\x3Cid>] [--target \x3Cpath>]\r
```\r
\r
Restore from a backup. If no `\x3Cid>` given, uses the latest. Before restoring, a safety backup of the current state is created automatically.\r
\r
---\r
\r
## AI Agent Behavior Guide\r
\r
When the user triggers this skill, follow the intelligent workflows below instead of blindly running commands.\r
\r
### Export Workflow\r
\r
When the user says things like "导出 claw", "export my setup", "backup my openclaw":\r
\r
**Step 1 — Gather context automatically**\r
\r
Before asking the user anything, silently gather:\r
\r
```bash\r
# Check what's in the workspace\r
ls ~/.openclaw/workspace/\r
\r
# Read identity to infer a good package name\r
cat ~/.openclaw/workspace/IDENTITY.md 2>/dev/null | head -20\r
\r
# Check installed skills (these get recorded in the manifest)\r
clawhub list --workdir ~/.openclaw 2>/dev/null\r
\r
# Check config exists\r
test -f ~/.openclaw/openclaw.json && echo "config exists" || echo "no config"\r
```\r
\r
**Step 2 — Infer smart defaults, then confirm with user**\r
\r
Based on what you found:\r
- **ref**: Use the agent's name/identity as the package name. Example: if IDENTITY.md says the agent is called "X", use `user/x:v1`. If unclear, ask.\r
- **output**: Default to `~/Desktop/\x3Cname>.claw` (easy to find and share). On Linux, use `~/`.\r
- **description**: Summarize the agent's personality/purpose from IDENTITY.md or SOUL.md in one sentence.\r
- **include-memory**: Default NO. Only suggest YES if the user explicitly mentions wanting to transfer memories.\r
\r
Present your plan concisely:\r
\r
> I'll export your OpenClaw as `user/x:v1` to `~/Desktop/x-v1.claw`.\r
> Description: "An assistant named X with ..."\r
> Memory: not included\r
> Skills bundled: [email protected], custom-skill (2 skills from ~/.openclaw/skills/)\r
>\r
> Proceed?\r
\r
**Step 3 — Execute and report**\r
\r
```bash\r
node {baseDir}/scripts/clawctl.mjs save \x3Cref> -o \x3Cpath> --description "\x3Cdesc>"\r
```\r
\r
After success, tell the user:\r
1. The file location and size\r
2. How many files were included\r
3. How many credentials were stripped (reassure them it's safe to share)\r
4. How to transfer it: "Copy this file to another machine, then tell that machine's OpenClaw '请安装这个 claw 包'"\r
\r
### Import Workflow\r
\r
When the user says things like "导入 claw", "install this claw", "load this package", or provides a `.claw` file path:\r
\r
**Step 1 — Locate the .claw file**\r
\r
If the user didn't specify a path, search for it:\r
\r
```bash\r
# Check common locations\r
ls ~/Desktop/*.claw ~/Downloads/*.claw /tmp/*.claw ~/*.claw 2>/dev/null\r
```\r
\r
If multiple `.claw` files found, list them and ask which one. If none found, ask the user for the path.\r
\r
**Step 2 — Preview first (always)**\r
\r
ALWAYS dry-run before installing:\r
\r
```bash\r
node {baseDir}/scripts/clawctl.mjs load \x3Cfile> --dry-run\r
```\r
\r
Present the preview to the user clearly:\r
- Package name and description\r
- Number of workspace files (list them)\r
- Required credentials that need to be configured\r
- Required skills that need to be installed\r
\r
**Step 3 — Analyze conflicts**\r
\r
Before installing, check what already exists:\r
\r
```bash\r
# Check if workspace files would be overwritten\r
ls ~/.openclaw/workspace/\r
```\r
\r
If existing files would be overwritten, WARN the user explicitly:\r
\r
> The following files already exist and will be overwritten:\r
> - IDENTITY.md\r
> - SOUL.md\r
>\r
> Your current agent personality will be replaced. A backup will be created automatically so you can rollback if needed. Continue?\r
\r
**Step 4 — Install (with auto-backup)**\r
\r
The load command automatically creates a backup before installing:\r
\r
```bash\r
node {baseDir}/scripts/clawctl.mjs load \x3Cfile>\r
```\r
\r
This will output something like:\r
```\r
  Creating backup before install...\r
  Backup created: 20260308-143022 (12 files)\r
  Rollback with: clawctl rollback 20260308-143022\r
```\r
\r
Tell the user the backup id and reassure them they can rollback at any time.\r
\r
**Step 5 — Post-install intelligence**\r
\r
This is where AI adds the most value. The `load` command already handles:\r
- Auto-backup before install (with rollback id)\r
- Workspace file installation\r
- **Bundled skill installation** (skills packaged in the .claw are auto-installed to `~/.openclaw/skills/`)\r
- Existing skill version comparison (skips if same or newer version already installed)\r
\r
After the load command completes, the agent should handle:\r
\r
1. **Skill gap check** — The load output shows "Additional skills referenced in config (not bundled, may be built-in)". For each:\r
   - If it's a built-in OpenClaw skill (like `coding-agent`, `github`), it's already available — no action needed.\r
   - If it's a ClawHub skill, offer to install:\r
\r
   ```bash\r
   clawhub install \x3Cskill-name> --workdir ~/.openclaw\r
   ```\r
\r
2. **Config merge** — Read both `~/.openclaw/openclaw.json` and `~/.openclaw/openclaw.imported.json`. Identify what the imported config adds (new model providers, channel settings, skill configurations) vs what the current config already has. Present a summary:\r
\r
   > The imported config includes:\r
   > - Telegram channel config (needs bot token)\r
   > - Custom model provider "kimi-code"\r
   > - Skill config for "nano-banana-pro"\r
   >\r
   > Your current config already has: [list]\r
   > New items to merge: [list]\r
   > Shall I merge the non-sensitive settings?\r
\r
   If the user agrees, read both JSON files, merge intelligently (keep current credentials, add new non-sensitive settings), and write back to `openclaw.json`.\r
\r
3. **Credential checklist** — For each `$CLAW_PLACEHOLDER` in the imported config, tell the user exactly what they need to set:\r
\r
   > To complete the setup, you need to configure:\r
   > - Telegram bot token: `openclaw config set channels.telegram.botToken \x3Cyour-token>`\r
   > - Gateway auth token: `openclaw config set gateway.auth.token \x3Cyour-token>`\r
\r
4. **Verification** — After everything is done, do a quick health check:\r
\r
   ```bash\r
   # Verify workspace files are in place\r
   ls ~/.openclaw/workspace/\r
   # Verify skills are installed\r
   ls ~/.openclaw/skills/\r
   # Verify config is valid JSON\r
   node -e "JSON.parse(require('fs').readFileSync(require('os').homedir()+'/.openclaw/openclaw.json','utf8')); console.log('Config OK')"\r
   ```\r
\r
### Rollback Workflow\r
\r
When the user says "回滚", "rollback", "undo import", "restore my old config":\r
\r
**Step 1 — List available backups**\r
\r
```bash\r
node {baseDir}/scripts/clawctl.mjs backups\r
```\r
\r
Show the user what's available with timestamps and labels.\r
\r
**Step 2 — Confirm which backup to restore**\r
\r
If only one backup exists, confirm:\r
\r
> Found 1 backup: `20260308-143022` (before-load:my-bot.claw, 12 files)\r
> Restore this? A safety backup of your current state will be created first.\r
\r
If multiple, ask user to pick.\r
\r
**Step 3 — Execute rollback**\r
\r
```bash\r
node {baseDir}/scripts/clawctl.mjs rollback \x3Cid>\r
```\r
\r
The rollback command automatically creates a safety backup of the current state before restoring, so the user can undo the undo if needed.\r
\r
**Step 4 — Verify**\r
\r
```bash\r
ls ~/.openclaw/workspace/\r
```\r
\r
Confirm the restored files match expectations.\r
\r
### Manual Backup Workflow\r
\r
When the user says "备份一下当前状态", "save a checkpoint", "backup before I try something":\r
\r
```bash\r
node {baseDir}/scripts/clawctl.mjs backup --label "user description here"\r
```\r
\r
Report the backup id so they can reference it later.\r
\r
### Edge Cases\r
\r
- **User says "导出" without any context**: Gather info first, then propose defaults. Never ask more than 2 questions.\r
- **User provides a URL instead of a file path**: Download the file first with `curl -o /tmp/downloaded.claw \x3Curl>`, then proceed with import.\r
- **Import on a fresh OpenClaw with no existing config**: Skip conflict checking, just install everything directly. Auto-backup still runs (backs up the empty state — gives a clean rollback target).\r
- **User wants to export + share in one step**: After export, suggest easy transfer methods (AirDrop, scp, cloud drive).\r
- **Multiple .claw files**: If the user says "install all claw packages", iterate through them one by one with preview for each.\r
- **Rollback chain**: Each rollback creates a safety backup, so the user can always undo a rollback. Explain this when asked.\r
- **Disk space**: If the user has many backups, suggest cleaning old ones: `rm -rf ~/.openclaw/.agentar-backups/\x3Cold-id>`.\r
\r
## What Gets Exported\r
\r
- Workspace files (system prompt, agent personality, custom instructions)\r
- Sanitized configuration (credentials replaced with `$CLAW_PLACEHOLDER`)\r
- **All user-installed skills** from `~/.openclaw/skills/` (full files, auto-installed on import)\r
- Skill list and channel list (metadata for dependency tracking)\r
\r
## What Does NOT Get Exported\r
\r
- API keys, tokens, passwords (automatically stripped)\r
- Session logs and history\r
- Memory (unless `--include-memory` is specified)\r
- Credentials directory\r
- Binary files (images, keys, certificates)\r
\r
## What Gets Backed Up (on load/rollback)\r
\r
- All workspace files (`~/.openclaw/workspace/`)\r
- Configuration (`~/.openclaw/openclaw.json`)\r
- Stored in `~/.openclaw/.agentar-backups/\x3Ctimestamp>/`