Model Healthcheck Skill
When the user requests a model test, follow these steps:
Steps
- Get all model list Use
gateway config.getto read the current config and extract allprovider/modelcombinations frommodels.providers. - Test all models concurrently For each model, use
sessions_spawnto run a test: Fire all tests concurrently (multiplesessions_spawncalls at once), do not test sequentially.
- task: "Reply with exactly: model ok. Only reply these two words." - model: the corresponding provider/model-id - runTimeoutSeconds: 30 - cleanup: "delete"
- Collect results and summarize Wait for all tests to complete, then summarize as a list: Fields:
- provider/model-id — ✅ OK / ❌ Failed (Xs) — error message if any - 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