live-preview（实时预览）

Create and report a live preview endpoint for the current repository.

Use arguments:

• Raw arguments: $ARGUMENTS
• First argument: $0
• Second argument: $1

Follow this workflow.

1) Resolve runtime plan

• Prefer explicit user arguments for run command, port, and tunnel provider.
• Otherwise infer a dev/start command from project files (e.g., package.json, pyproject.toml, requirements.txt, Cargo.toml, go.mod) and repository docs.
• Prefer loopback bind target (127.0.0.1).

2) Handle local port and app process

• Before starting a new service, check whether the target port is already in use.
• If port is free, start the app in background and capture PID.
• If port is occupied, inspect the owning process command/cwd.
• If the process appears to be the same project and same service (matching repository path and compatible run command), reuse that existing service and keep the same port.
• If the process is different (other project or different service), choose one alternate port and start a new service once.
• Include in message whether the preview reused an existing service or started a new one.

3) Verify local readiness

• Probe http://127.0.0.1:<port> for up to 60 seconds.
• Treat any HTTP response as reachable preview surface.
• If unreachable, return an error result in the structured Markdown format.

4) Start tunnel

• Try Cloudflare first: cloudflared tunnel --url http://127.0.0.1:<port>.
• Parse https://*.trycloudflare.com from output.
• If unavailable or failed, try ngrok (ngrok http <port>) and read URL from http://127.0.0.1:4040/api/tunnels.
• If both fail due to cloudflare or ngrok not installed, install them and try again.
• If tunnel setup fails, return an error result that includes a concrete install/action hint.

5) Emit structured Markdown result

• Print final output as Markdown bullets only, with * at the beginning of each line.
• Extract the tunnel URL from provider output and include it as public_url.
• Include exactly these fields, one per line:

• status: ok or error
• provider: cloudflare, ngrok, or none
• local_url: local URL such as http://127.0.0.1:3000
• port: numeric local port
• pid: process id or null
• message: short actionable summary
• public_url: tunnel URL or null

• Do not print secrets.
• Keep message concise and actionable.

6) Keep lifecycle predictable

• Keep preview processes running unless the user asked for one-shot verification only.
• If user asked to stop/cleanup, terminate local app and tunnel processes and return success result in the same structured Markdown format.