Model Healthcheck Skill

When the user requests a model test, follow these steps:

Steps

1. Get all model list Use gateway config.get to read the current config and extract all provider/model combinations from models.providers.

2. Test all models concurrently For each model, use sessions_spawn to run a test:

   • task: "Reply with exactly: model ok. Only reply these two words."
   • model: the corresponding provider/model-id
   • runTimeoutSeconds: 30
   • cleanup: "delete"

   Fire all tests concurrently (multiple sessions_spawn calls at once), do not test sequentially.

3. Collect results and summarize Wait for all tests to complete, then summarize as a list:

   • provider/model-id — ✅ OK / ❌ Failed (Xs) — error message if any

   Fields:

   • Status: normal reply = ✅, timeout or error = ❌
   • Duration: time from spawn to result
   • Error: if failed, show the specific error (API error, timeout, auth failure, etc.)

4. Output results Send the summary to the user, with a final tally:

   • Total models tested: X
   • Passed: X
   • Failed: X (list failed model names and reasons)

Notes

• Skip the model currently in use (no need to test yourself)
• If the user specifies a particular model, only test that one
• Feishu/WhatsApp and similar platforms don't support markdown tables — use list format instead