cli: opencode models prints empty list with exit 0 on a cold server #53724

Summary

opencode models silently prints nothing (empty list, exit code 0) when run against a freshly spawned server. Model/provider lists load asynchronously, and the command queries before loading finishes, so it always reports zero models on a cold server.

Environment

• opencode version: 2.0.24
• OS: Linux x86_64, kernel 6.8.0-110-generic (Ubuntu)
• Terminal: TERM=xterm-256color, TERM_PROGRAM=vscode, COLORTERM=truecolor
• Shell: /bin/bash
• Install/channel: npm global @opencode/cli via nvm-managed Node v25.3.0, channel=latest
• Active plugins: none

Reproduction

1. Disable the shared background service: opencode service set disabled true (or just use --standalone below)
2. Run opencode models --standalone
3. Observe: no output at all, exit code 0 (reproduced 3/3 times)
4. Contrast: start opencode serve --port <port>, wait ~2 minutes, then run opencode models --server http://localhost:<port> — the full model list prints instantly
5. On the fresh server, GET /api/model returns {"data":[]} right after boot and the complete list a couple of minutes later; /api/provider likewise shows only built-in integrations at first

Expected Behavior

opencode models waits for the initial provider/model load (bounded by a timeout), or prints a warning when the resulting list is empty, instead of silently succeeding with no output.

Actual Behavior

Prints nothing and exits 0. Looks like "no models configured" even though the config is valid.

Additional Context

Config defines one custom provider (redacted):

{
  "model": "<provider>/<model>",
  "provider": {
    "<provider>": {
      "npm": "@ai-sdk/openai-compatible",
      "options": { "apiKey": "<redacted>", "baseURL": "https://<private-proxy>/v1" },
      "models": { /* ~20 manually declared models with limits */ }
    }
  }
}

~/.config/opencode/service.json contains {"disabled": true}, so every CLI invocation spawns a private stdio server and hits the cold-start path. The TUI in the same directory works fine because it stays connected and models finish loading in the background. Slow model-list fetching (network latency) makes the race essentially deterministic here.