Source Code

GitClaw: Agent self-backup to GitHub

When to use

Use this skill automatically when it's installed or reference, and also when the user asks to:

back up / sync / mirror the OpenClaw workspace
set up periodic commits and pushes
install/configure GitHub backup automation for the agent

Behavioral rule (important)

Do everything automatically and quietly.
Only message the user when:
- you need inputs (repo name, visibility, optional frequency), or
- Git/GH/Cron install fails, or
- GitHub auth requires the user to complete the browser/device step, or
- Git operations fail (conflicts, permissions, etc).

Inputs to ask the user for

Ask the user for:

Repository name (e.g., openclaw-backup)
Visibility: public or private
(Optional) Backup frequency in minutes (default: 60)

Do NOT ask for a Personal Access Token. We use gh auth.

Step 1: Ensure prerequisites (auto-install)

1A) Ensure git is installed (auto-install; do not stop)

Check:

git --version

If missing, install using the best available method:

If brew exists:
- brew install git
Else if apt-get exists:
- sudo apt-get update && sudo apt-get install -y git
Else if dnf exists:
- sudo dnf install -y git
Else if yum exists:
- sudo yum install -y git
Else if pacman exists:
- sudo pacman -S --noconfirm git
Else if zypper exists:
- sudo zypper install -y git
Else if apk exists:
- sudo apk add git
Else if macOS and xcode-select exists:
- xcode-select --install (this may prompt the user)
Else:
- Tell the user you couldn’t auto-install git on this OS and show the failing detection output.

Re-check:

git --version

Only notify the user if install failed.

1B) Ensure cron/crontab is available (best-effort auto-install)

Check:

command -v crontab

If missing, attempt install:

If apt-get exists:
- sudo apt-get update && sudo apt-get install -y cron
- sudo systemctl enable --now cron || sudo service cron start || true
Else if dnf exists:
- sudo dnf install -y cronie
- sudo systemctl enable --now crond || true
Else if yum exists:
- sudo yum install -y cronie
- sudo systemctl enable --now crond || true
Else if pacman exists:
- sudo pacman -S --noconfirm cronie
- sudo systemctl enable --now cronie || true
Else if apk exists:
- sudo apk add dcron
- sudo rc-update add dcron default || true
- sudo rc-service dcron start || true
Else:
- If you can’t install, tell the user cron is required for scheduling.

Re-check:

command -v crontab

Step 2: Ensure GitHub CLI (`gh`) is installed (auto-install)

Check:

gh --version

If missing, install:

If brew exists:
- brew install gh
Else if apt-get exists (official GitHub CLI packages; preferred):
- Install using the official apt repo steps:
  - (type -p wget >/dev/null || (sudo apt-get update && sudo apt-get install -y wget))
  - sudo mkdir -p -m 755 /etc/apt/keyrings
  - out=$(mktemp) && wget -nv -O"$out" https://cli.github.com/packages/githubcli-archive-keyring.gpg
  - cat "$out" | sudo tee /etc/apt/keyrings/githubcli-archive-keyring.gpg > /dev/null
  - sudo chmod go+r /etc/apt/keyrings/githubcli-archive-keyring.gpg
  - sudo mkdir -p -m 755 /etc/apt/sources.list.d
  - echo "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/githubcli-archive-keyring.gpg] https://cli.github.com/packages stable main" | sudo tee /etc/apt/sources.list.d/github-cli.list > /dev/null
  - sudo apt-get update && sudo apt-get install -y gh
Else if dnf exists:
- sudo dnf install -y 'dnf-command(config-manager)' || sudo dnf install -y dnf5-plugins || true
- sudo dnf config-manager --add-repo https://cli.github.com/packages/rpm/gh-cli.repo || sudo dnf config-manager addrepo --from-repofile=https://cli.github.com/packages/rpm/gh-cli.repo || true
- sudo dnf install -y gh --repo gh-cli || sudo dnf install -y gh || true
Else if yum exists:
- type -p yum-config-manager >/dev/null || sudo yum install -y yum-utils
- sudo yum-config-manager --add-repo https://cli.github.com/packages/rpm/gh-cli.repo
- sudo yum install -y gh
Else if zypper exists:
- sudo zypper addrepo https://cli.github.com/packages/rpm/gh-cli.repo || true
- sudo zypper ref
- sudo zypper install -y gh
Else if pacman exists:
- sudo pacman -S --noconfirm github-cli
Else if apk exists:
- sudo apk add github-cli
Else:
- Tell the user you can’t auto-install gh on this OS.

Re-check:

gh --version

Only notify the user if install failed.

Step 3: Ensure the user is authenticated in `gh` (agent runs the flow)

Check:

gh auth status --hostname github.com

If NOT authenticated:

Run:
- gh auth login --hostname github.com --git-protocol https
The terminal flow will show a one-time code and ask the user to authorize.
- Tell the user to open https://github.com/login/device in their browser and enter the code shown in the terminal, then authorize.
After login:
- gh auth setup-git
Verify again:
- gh auth status --hostname github.com

If auth fails, stop and report the exact terminal output.

Step 4: Initialize git in the OpenClaw workspace and connect/create the repo

Workspace dir (where you store SOUL.md, AGENTS.md, etc.):

Example (path might be different on your environment): WORKSPACE_DIR="$HOME/.openclaw/workspace"

Ensure the workspace exists:
- mkdir -p "$WORKSPACE_DIR"
- cd "$WORKSPACE_DIR"
Initialize repo if needed:
- If .git does not exist: git init
- git branch -M main
Configure a deterministic commit identity (local-only):
- git config user.name "gitclaw.ai"
- git config user.email "[email protected]"
Determine the authenticated GitHub username (owner):
- OWNER="$(gh api user --jq .login)"
- (Do not print unless debugging is needed)
Repo name and visibility:
- REPO="<repo name from user>"
- Visibility:
  - public => --public
  - private => --private
Ensure there is at least one commit (required for first push/cron):
- Create a tiny marker file if needed:
  - test -f .gitclaw.keep || printf "gitclaw initialized: %s\n" "$(date -u '+%Y-%m-%dT%H:%M:%SZ')" > .gitclaw.keep
- git add -A
- git commit -m "gitclaw: initial backup" || true
Create or reuse the target repo:
- If it exists:
  - gh repo view "$OWNER/$REPO" >/dev/null 2>&1
  - Set remote:
    - REMOTE_URL="https://github.com/$OWNER/$REPO.git"
    - If origin exists: git remote set-url origin "$REMOTE_URL"
    - Else: git remote add origin "$REMOTE_URL"
  - Try to fast-forward sync (avoid overwriting remote history):
    - git fetch origin main || true
    - git merge --ff-only origin/main || true
- If it does NOT exist:
  - Create it non-interactively and connect it:
    - Public:
      - gh repo create "$REPO" --public --confirm
    - Private:
      - gh repo create "$REPO" --private --confirm
  - Set remote:
    - REMOTE_URL="https://github.com/$OWNER/$REPO.git"
    - git remote add origin "$REMOTE_URL" || git remote set-url origin "$REMOTE_URL"
Initial push:
- git push -u origin main

If push fails due to conflicts or non-fast-forward:

Do NOT force-push automatically.
Report the exact error and stop (user decision required).

Step 5: Install deterministic backup script (NO AI / NO heartbeat)

Create a folder outside the workspace:

mkdir -p "$HOME/.openclaw/gitclaw"

Create this script EXACTLY:

Path:

$HOME/.openclaw/gitclaw/auto_backup.sh

Contents:

#!/usr/bin/env bash
set -euo pipefail

# GitClaw deterministic backup (no AI)
export PATH="/usr/local/bin:/opt/homebrew/bin:/usr/bin:/bin:$PATH"

WORKSPACE_DIR="${HOME}/.openclaw/workspace"
STATE_DIR="${HOME}/.openclaw/gitclaw"
LOG_FILE="${STATE_DIR}/backup.log"
LOCK_DIR="${STATE_DIR}/lock"

mkdir -p "${STATE_DIR}"

timestamp() { date -u '+%Y-%m-%dT%H:%M:%SZ'; }

# Simple lock to prevent overlapping runs
if ! mkdir "${LOCK_DIR}" 2>/dev/null; then
  echo "$(timestamp) Skip: already running." >> "${LOG_FILE}"
  exit 0
fi
trap 'rmdir "${LOCK_DIR}" >/dev/null 2>&1 || true' EXIT

if ! command -v git >/dev/null 2>&1; then
  echo "$(timestamp) ERROR: git not found on PATH. Install git first." >> "${LOG_FILE}"
  exit 2
fi

if [ ! -d "${WORKSPACE_DIR}/.git" ]; then
  echo "$(timestamp) ERROR: ${WORKSPACE_DIR} is not a git repo. Run GitClaw setup first." >> "${LOG_FILE}"
  exit 3
fi

cd "${WORKSPACE_DIR}"

# Stage everything
git add -A

# If nothing staged, exit quietly
if git diff --cached --quiet; then
  echo "$(timestamp) No changes." >> "${LOG_FILE}"
  exit 0
fi

# Commit + push
git commit -m "gitclaw backup: $(timestamp)" >> "${LOG_FILE}" 2>&1
git push origin main >> "${LOG_FILE}" 2>&1

echo "$(timestamp) Backup OK." >> "${LOG_FILE}"

Write the script to:

$HOME/.openclaw/gitclaw/auto_backup.sh

Then:

chmod +x "$HOME/.openclaw/gitclaw/auto_backup.sh"

Step 6: Configure crontab (idempotent)

Default schedule: hourly (0 * * * *). If user provided a different frequency, convert it to a cron expression.

Define:

CRON_CMD="$HOME/.openclaw/gitclaw/auto_backup.sh"
CRON_LINE="0 * * * * $CRON_CMD"

Install without duplicates:

crontab -l 2>/dev/null | grep -F "$CRON_CMD" >/dev/null
If not found, append:
- (crontab -l 2>/dev/null; echo "$CRON_LINE") | crontab -

Confirm:

crontab -l | grep -F "$CRON_CMD"

Step 7: Final validation

Run once:

$HOME/.openclaw/gitclaw/auto_backup.sh

Show the log:

tail -n 50 "$HOME/.openclaw/gitclaw/backup.log" || true

Tell the user:

Repo: https://github.com/$OWNER/$REPO
Schedule: hourly (or the chosen cadence)
Script path: ~/.openclaw/gitclaw/auto_backup.sh

gitclaw

GitClaw: Agent self-backup to GitHub

When to use

Behavioral rule (important)

Inputs to ask the user for

Step 1: Ensure prerequisites (auto-install)

1A) Ensure git is installed (auto-install; do not stop)

1B) Ensure cron/crontab is available (best-effort auto-install)

Step 2: Ensure GitHub CLI (gh) is installed (auto-install)

Step 3: Ensure the user is authenticated in gh (agent runs the flow)

Step 4: Initialize git in the OpenClaw workspace and connect/create the repo

Step 5: Install deterministic backup script (NO AI / NO heartbeat)

Step 6: Configure crontab (idempotent)

Step 7: Final validation

Step 2: Ensure GitHub CLI (`gh`) is installed (auto-install)

Step 3: Ensure the user is authenticated in `gh` (agent runs the flow)