返回 Skill 列表
extension
分类: AI Agent 能力无需 API Key

model-switch

在运行时切换代理的模型并重新启动。当用户要求更换模型(例如,“切换到kimi”,“使用opus”,“换一个更便宜的”)或代理认为不同的模型更适合当前工作负载时使用。

person作者: jakexiaohubgithub

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.patchagents.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/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.