You spent weeks teaching your AI agent.
Don't lose it to a machine swap.

One command to backup. One command to restore.

Atomic Writes · SHA-256 Integrity · Auto Secret Redaction · Smart Merge

English · 中文

---

Quickstart

You need: Python 3.8+ · Git · A private GitHub repo

# Install
git clone https://github.com/AlphaWill0/agent-mind-migrate.git ~/.claude/skills/agent-mind-migrate

# Setup (once)
python3 ~/.claude/skills/agent-mind-migrate/scripts/migrate.py init --remote <your-repo-url>

# Backup
python3 ~/.claude/skills/agent-mind-migrate/scripts/migrate.py backup --push

Or just tell Claude Code: "back up my config" — it figures out the rest.

💡 Create a short alias

alias migrate='python3 ~/.claude/skills/agent-mind-migrate/scripts/migrate.py'
# Then just: migrate backup --push

---

New Machine

git clone <your-backup-repo-url> ~/.claude-backup
git clone https://github.com/AlphaWill0/agent-mind-migrate.git ~/.claude/skills/agent-mind-migrate
python3 ~/.claude/skills/agent-mind-migrate/scripts/migrate.py restore --dry-run    # preview
python3 ~/.claude/skills/agent-mind-migrate/scripts/migrate.py restore              # restore

That's it. Your agent remembers everything.

Run validate after restore — fill in any __REDACTED__ placeholders with your real API keys.

---

What It Captures

Auto-detects installed agents. Zero configuration.

	Claude Code	OpenClaw	Hermes
Config & settings	✅	✅	✅
Memory	✅	✅ (SQLite)	✅
Skills	✅	—	✅
Rules & agents	✅	—	✅ (SOUL.md)
Commands & cron	✅	✅	✅
Project memories	✅	—	—
Extensions / plugins	✅	✅	—

Git-based skills store URL + commit SHA only — auto-cloned on restore.

---

Security

Your secrets never touch Git. API keys, tokens, and passwords are replaced with __REDACTED__ before every commit — including MCP server configs and skill remote URLs. On restore, placeholders smart-merge with your machine's real values. Credentials directories are excluded entirely.

---

Roadmap

• init auto-creates the backup repo via gh
• Scheduled auto-backup
• Diff view: what changed since last backup

If any of these matter to you, a ⭐ helps me prioritize.

Found a bug? Open an issue · Built another agent? The plugin architecture makes it easy to add support. PRs welcome.

---

Advanced Usage

Commands

migrate.py init --remote <url>           # First-time setup
migrate.py backup [--push]               # Backup all detected agents
migrate.py backup --tier full --push     # Full backup (history + plugins)
migrate.py backup --agents claude-code   # Specific agent only
migrate.py restore --dry-run             # Preview (always do this first)
migrate.py restore --only skills memory  # Selective restore
migrate.py restore --agents openclaw     # Specific agent only
migrate.py status                        # Backup status
migrate.py validate                      # Health check

Tiers

Tier	Includes	Use case
essential (default)	Config, memory, skills, rules, commands, cron, stats	Daily backup
full	+ command history, plans, plugins	Machine migration

Restore Flags

Flag	Effect
--conflict backup-existing	Default — back up existing files before overwriting
--conflict overwrite	Overwrite directly
--conflict skip	Skip existing files
--only <modules>	config memory skills rules agents commands scheduled_tasks stats project_memories plans history plugins
--agents <name>	Specific agent only
--dry-run	Preview without changes
--yes	Skip confirmation prompt
--no-pull	Skip auto-pull from remote

Project memories with absolute paths are auto-translated to the new machine's HOME during restore.

Add Your Own Agent

class YourAgentPlugin(AgentPlugin):
    name = "your-agent"
    config_dir = Path.home() / ".your-agent"

    def discover(self) -> bool: ...
    def backup(self, staging, tier) -> list: ...
    def restore(self, source, conflict, only) -> None: ...
    def sanitize(self, data) -> tuple: ...
    def status(self, repo) -> dict: ...

Register in AGENT_PLUGINS. Auto-discovered.

FAQ

Why not cp -r ~/.claude?

Copies API keys into git. No integrity checks, no redaction, no selective restore.

Why not chezmoi / yadm?

Built for shell config. AI agent state is different — SQLite, nested git repos, scattered project memory. This tool understands the structure.

Only use Claude Code?

It auto-detects. One agent installed → one agent backed up.

Is it safe?

Secrets redacted before commit. SHA-256 per file. Atomic writes. Smart merge on restore.

Changelog

v4.1

• Bilingual CLI (auto-detects zh/en)
• MCP config sanitization, git URL credential stripping, Hermes config.yaml redaction
• Restore confirmation prompt, auto git pull, default conflict → backup-existing
• Cross-machine path translation for project memories
• Symlink detection fix, validate exit code fix

v4.0

• Multi-agent: Claude Code + OpenClaw + Hermes
• Per-agent directories, --agents flag, AgentPlugin architecture

v3.x

• Atomic writes, SHA-256, smart merge, cross-platform, selective restore

---

中文

一条命令备份，一条命令还原。换机器不丢记忆。

备份

git clone https://github.com/AlphaWill0/agent-mind-migrate.git ~/.claude/skills/agent-mind-migrate
python3 ~/.claude/skills/agent-mind-migrate/scripts/migrate.py init --remote <备份仓库URL>
python3 ~/.claude/skills/agent-mind-migrate/scripts/migrate.py backup --push

或对 Claude Code 说「备份一下」。

新机器还原

git clone <备份仓库URL> ~/.claude-backup
git clone https://github.com/AlphaWill0/agent-mind-migrate.git ~/.claude/skills/agent-mind-migrate
python3 ~/.claude/skills/agent-mind-migrate/scripts/migrate.py restore --dry-run
python3 ~/.claude/skills/agent-mind-migrate/scripts/migrate.py restore

还原后 validate，把 __REDACTED__ 替换为真实密钥。

支持的 Agent

自动检测，零配置。

Agent	备份内容
Claude Code	配置、记忆、技能、规则、命令、定时任务、项目记忆
OpenClaw	配置、bot 设置、记忆、定时任务、插件
Hermes	配置、身份、记忆、技能、定时任务

安全：token / 密码 / API key 自动脱敏，不进 git。还原时保留本机真实值。

---

Your agent's mind deserves a backup plan.

MIT License · AlphaWill