SKILL.md

metadata

name: openclaw-hf-deploy
description: >-
  Agent SOP for deploying this repository to Hugging Face Space with the project
  bootstrap scripts.

OpenClaw HF Deployment Skill

Use this skill when a user asks an agent to deploy openclaw-hf to Hugging Face.

Hard rules

• Always reuse project bootstrap scripts:
  • macOS/Linux: ./scripts/bootstrap-hf.sh
  • Windows PowerShell: powershell -ExecutionPolicy ByPass -File .\scripts\bootstrap-hf.ps1
• Do not hand-roll a separate deployment flow unless the user explicitly asks for manual fallback.
• Keep secrets masked in agent logs and summaries.

1. Preflight checks

Run and verify:

command -v git >/dev/null 2>&1 || MISSING_GIT=1
command -v hf >/dev/null 2>&1 || MISSING_HF=1
command -v python3 >/dev/null 2>&1 || MISSING_PYTHON3=1
git --version
hf version
python3 --version
python3 -c "import huggingface_hub" >/dev/null 2>&1 || python3 -m pip install --user 'huggingface_hub[cli]'

Install missing tools before continuing.

2. Clone repository (required before deployment)

Run:

git clone https://github.com/tenfyzhong/openclaw-hf.git
cd openclaw-hf

Do not continue until the shell is inside the cloned openclaw-hf directory.

3. HF auth flow must run first (before any other variables)

1. If user already provided HF_TOKEN in the conversation:
   • use this HF_TOKEN directly: hf auth login --token <HF_TOKEN>
   • if a different user is currently logged in, backup current token first and restore it after deployment
   • do not ask whether to keep current logged-in user
2. Otherwise, check login: hf auth whoami.
3. If not logged in:
   • ask user for HF_TOKEN (required)
   • run: hf auth login --token <HF_TOKEN>
4. If already logged in:
   • ask whether to use current logged-in user
   • if yes: continue
   • if no:
     • backup current token
     • ask for new HF_TOKEN
     • run: hf auth login --token <HF_TOKEN>
     • continue deployment with new account
     • after deployment, restore previous token

4. Collect remaining inputs in prompt order

After step 3 is completed, collect:

1. space_name (default openclaw-hf)
2. dataset_name (default <space_name>-backup)
3. OPENCLAW_VERSION (default latest detected from npm, fallback latest)
4. OPENCLAW_GATEWAY_TOKEN (optional, auto-generated if empty)
5. OPENCLAW_GATEWAY_PASSWORD (optional, auto-generated if empty)
6. Configure custom LLM now? (y/N)
   • if yes: OPENCLAW_LLM_BASE_URL, OPENCLAW_LLM_MODEL, OPENCLAW_LLM_API_KEY
   • if no: OPENCLAW_SSHX_AUTO_START (Y/n)
7. Proceed with these settings? defaults to Y; agent should press Enter directly and not collect this input from user.

If user explicitly asks to cancel before execution, answer n at final confirmation and stop without creating/updating remote resources.

5. Execute deployment

From the cloned openclaw-hf repo root, run the platform-specific bootstrap script and follow prompts in order.

The script will:

• create private Space and Dataset
• upload repository to Space
• configure Space variables and secrets, including OPENCLAW_BACKUP_DATASET_REPO, OPENCLAW_VERSION, HF_TOKEN, gateway secrets, and optional LLM/SSHX settings

6. Required final hand-off

Agent final message must include:

• Space repo ID: <owner>/<space_name>
• Space page URL: https://huggingface.co/spaces/<owner>/<space_name>
• Dataset repo ID: <owner>/<dataset_name>
• OPENCLAW_VERSION used
• App URL: https://<owner>-<space_name>.hf.space
• Health URL: https://<owner>-<space_name>.hf.space/healthz
• auto-generated gateway token/password if generated
• keep-alive note: free tier may sleep; paid hardware + Sleep time=Never for strict 24/7 uptime

7. Cleanup local clone after deployment

After deployment and final hand-off are done, remove the local cloned repository:

cd ..
rm -rf openclaw-hf