Model Healthcheck Skill
When the user requests a model test, follow these steps:
Steps
- Get all model list
Use gateway config.get to read the current config and extract all provider/model combinations from models.providers.
- 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.
- 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.)
- 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