| name | model-switch |
| description | Switch the agent's model at runtime and restart. Use when the user asks to change models (e.g., "switch to kimi", "use opus", "go cheaper"), or when the agent decides a different model is better for the current workload. |
Model Switch
Overview
This skill switches the agent's primary model by patching the gateway config and letting the gateway auto-restart. Use it when a user explicitly requests a model change or when you determine a different model is better suited for the task (cost, speed, quality).
Workflow
1) Trigger
- User asks to change models ("switch to X", "use opus", "go cheaper")
- Or you decide a different model is better for the current workload
2) Map alias → full model ID
Use the reference table below to resolve common aliases. If the alias is unknown, ask for clarification or the exact model ID.
3) Patch gateway config
Update the primary model:
gateway config.patch → agents.defaults.model.primary = <full_model_id>
4) Gateway restarts automatically
No manual restart needed.
5) Confirm to the user
Confirm the model switch in plain language.
Reference: Known Model Aliases
| Alias | Full model ID |
|---|
| opus | anthropic/claude-opus-4-5 |
| sonnet | anthropic/claude-sonnet-4-5 |
| kimi | nvidia-build/moonshotai/kimi-k2.5 |
| codex | github-copilot/gpt-5.2-codex |
Notes
- Context resets after restart. Warn the user if the switch happens mid-conversation.
- Per-session override (no restart): You can use the
session_status tool with the model parameter to override the model for a single session without changing defaults.
