Agentar Skill Overview

Fast CLI to distribute an AI agent.

CLI Reference

Save (Export)

CODEBLOCK0

Load (Import)

CODEBLOCK1

Load automatically creates a backup before installing. This can be skipped with --no-backup.

Backup

CODEBLOCK2

Create a snapshot of current workspace + config. Stored in ~/.openclaw/.agentar-backups/<id>/.

List Backups

CODEBLOCK3

Rollback

CODEBLOCK4

Restore from a backup. If no <id> given, uses the latest. Before restoring, a safety backup of the current state is created automatically.

AI Agent Behavior Guide

When the user triggers this skill, follow the intelligent workflows below instead of blindly running commands.

Export Workflow

When the user says things like "导出 claw", "export my setup", "backup my openclaw":

Step 1 — Gather context automatically

Before asking the user anything, silently gather:

CODEBLOCK5

Step 2 — Infer smart defaults, then confirm with user

Based on what you found:

• - 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.
• output: Default to ~/Desktop/<name>.claw (easy to find and share). On Linux, use ~/.
• description: Summarize the agent's personality/purpose from IDENTITY.md or SOUL.md in one sentence.
• include-memory: Default NO. Only suggest YES if the user explicitly mentions wanting to transfer memories.

Present your plan concisely:

I'll export your OpenClaw as user/x:v1 to ~/Desktop/x-v1.claw.
Description: "An assistant named X with ..."
Memory: not included
Skills bundled: agentar@0.3.0, custom-skill (2 skills from ~/.openclaw/skills/)
Proceed?

Step 3 — Execute and report

CODEBLOCK6

After success, tell the user:

1. 1. The file location and size
2. How many files were included
3. How many credentials were stripped (reassure them it's safe to share)
4. How to transfer it: "Copy this file to another machine, then tell that machine's OpenClaw '请安装这个 claw 包'"

Import Workflow

When the user says things like "导入 claw", "install this claw", "load this package", or provides a .claw file path:

Step 1 — Locate the .claw file

If the user didn't specify a path, search for it:

CODEBLOCK7

If multiple .claw files found, list them and ask which one. If none found, ask the user for the path.

Step 2 — Preview first (always)

ALWAYS dry-run before installing:

CODEBLOCK8

Present the preview to the user clearly:

• - Package name and description
• Number of workspace files (list them)
• Required credentials that need to be configured
• Required skills that need to be installed

Step 3 — Analyze conflicts

Before installing, check what already exists:

CODEBLOCK9

If existing files would be overwritten, WARN the user explicitly:

The following files already exist and will be overwritten:

• - IDENTITY.md
• SOUL.md

Your current agent personality will be replaced. A backup will be created automatically so you can rollback if needed. Continue?

Step 4 — Install (with auto-backup)

The load command automatically creates a backup before installing:

CODEBLOCK10

This will output something like:
CODEBLOCK11

Tell the user the backup id and reassure them they can rollback at any time.

Step 5 — Post-install intelligence

This is where AI adds the most value. The load command already handles:

• - Auto-backup before install (with rollback id)
• Workspace file installation
• Bundled skill installation (skills packaged in the .claw are auto-installed to ~/.openclaw/skills/)
• Existing skill version comparison (skips if same or newer version already installed)

After the load command completes, the agent should handle:

1. 1. Skill gap check — The load output shows "Additional skills referenced in config (not bundled, may be built-in)". For each:

CODEBLOCK12

1. 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:

> The imported config includes:
> - Telegram channel config (needs bot token)
> - Custom model provider "kimi-code"
> - Skill config for "nano-banana-pro"
>
> Your current config already has: [list]
> New items to merge: [list]
> Shall I merge the non-sensitive settings?

If the user agrees, read both JSON files, merge intelligently (keep current credentials, add new non-sensitive settings), and write back to openclaw.json.

1. 3. Credential checklist — For each $CLAW_PLACEHOLDER in the imported config, tell the user exactly what they need to set:

> To complete the setup, you need to configure:
> - Telegram bot token: openclaw config set channels.telegram.botToken <your-token>
> - Gateway auth token: INLINECODE30

1. 4. Verification — After everything is done, do a quick health check:

CODEBLOCK13

Rollback Workflow

When the user says "回滚", "rollback", "undo import", "restore my old config":

Step 1 — List available backups

CODEBLOCK14

Show the user what's available with timestamps and labels.

Step 2 — Confirm which backup to restore

If only one backup exists, confirm:

Found 1 backup: 20260308-143022 (before-load:my-bot.claw, 12 files)
Restore this? A safety backup of your current state will be created first.

If multiple, ask user to pick.

Step 3 — Execute rollback

CODEBLOCK15

The rollback command automatically creates a safety backup of the current state before restoring, so the user can undo the undo if needed.

Step 4 — Verify

CODEBLOCK16

Confirm the restored files match expectations.

Manual Backup Workflow

When the user says "备份一下当前状态", "save a checkpoint", "backup before I try something":

CODEBLOCK17

Report the backup id so they can reference it later.

Edge Cases

• - User says "导出" without any context: Gather info first, then propose defaults. Never ask more than 2 questions.
• User provides a URL instead of a file path: Download the file first with curl -o /tmp/downloaded.claw <url>, then proceed with import.
• 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).
• User wants to export + share in one step: After export, suggest easy transfer methods (AirDrop, scp, cloud drive).
• Multiple .claw files: If the user says "install all claw packages", iterate through them one by one with preview for each.
• Rollback chain: Each rollback creates a safety backup, so the user can always undo a rollback. Explain this when asked.
• Disk space: If the user has many backups, suggest cleaning old ones: rm -rf ~/.openclaw/.agentar-backups/<old-id>.

What Gets Exported

• - Workspace files (system prompt, agent personality, custom instructions)
• Sanitized configuration (credentials replaced with $CLAW_PLACEHOLDER)
• All user-installed skills from ~/.openclaw/skills/ (full files, auto-installed on import)
• Skill list and channel list (metadata for dependency tracking)

What Does NOT Get Exported

• - API keys, tokens, passwords (automatically stripped)
• Session logs and history
• Memory (unless --include-memory is specified)
• Credentials directory
• Binary files (images, keys, certificates)

What Gets Backed Up (on load/rollback)

• - All workspace files (~/.openclaw/workspace/)
• Configuration (~/.openclaw/openclaw.json)
• Stored in INLINECODE39