CLIProxy + OpenClaw

Use this skill when the user wants to:

• install or deploy CLIProxyAPI
• expose the CLIProxy dashboard or management UI
• connect OAuth-based CLI subscriptions like Claude Code, Gemini, Codex, Qwen, or iFlow
• generate a reusable API endpoint and API key
• use CLIProxy with OpenClaw or another OpenAI-compatible client

Outcome

The job is complete only when all of these are true:

1. CLIProxyAPI is installed and running
2. the intended dashboard or management URL is reachable
3. the user has added one or more OAuth-backed providers or accounts
4. a reusable API endpoint and API key are available
5. OpenClaw or the target client is configured to use CLIProxy
6. a smoke test succeeds against a real model

Default workflow

1. Determine the target mode:

   • local only
   • VPS or private LAN
   • public remote dashboard access

2. Inspect the environment before changing anything:

   • OS and package/runtime availability
   • whether Docker, systemd, nginx, Caddy, or another reverse proxy already exists
   • whether OpenClaw is already installed and how it is configured
   • firewall state and whether public exposure is actually desired

3. Install and start CLIProxyAPI.

   • Prefer a stable service deployment over an ad-hoc shell session.
   • Prefer systemd when available.
   • After install, verify the process is actually listening.

4. Expose access only as needed.

   • If the user wants remote access, prefer reverse proxy plus minimal port exposure.
   • Do not open management surfaces wider than necessary.
   • State clearly what URL and what ports will become reachable.

5. Guide provider onboarding.

   • Tell the user how to open the dashboard.
   • Have them add OAuth providers or accounts.
   • Confirm that models become visible and usable.

6. Capture integration details.

   • base URL
   • API key or token
   • model names
   • any special headers if the deployment requires them

7. Connect the result to OpenClaw.

   • Use the most direct compatible provider path available in OpenClaw.
   • If exact manual values are needed, provide them explicitly.

8. Run a smoke test.

   • list models if available
   • send a minimal request
   • verify the selected model returns a real response

Read references only when needed

• For install and service layout: references/install.md
• For dashboard exposure, reverse proxy, or ports: references/dashboard.md
• For adding OAuth providers and accounts: references/providers.md
• For connecting CLIProxy to OpenClaw: references/openclaw-integration.md
• For failures like 401, 403, 404, 429, 502, model-not-found, or streaming mismatches: references/troubleshooting.md

Operating rules

• Prefer fewer checkpointed steps over long blind command chains.
• Verify actual state after each major step before moving on.
• Treat API keys, OAuth tokens, session cookies, and dashboard credentials as sensitive.
• Do not assume public exposure is desired. If unclear, ask.
• The goal is not to "install the repo"; the goal is to produce one working API layer that OpenClaw or another client can really use.
• If the user wants this published on ClawHub, keep the operational guidance concise in SKILL.md and move detail into references.