feat(plugin): add chat.variant hook for programmatic thinking/variant control

[code-yeongyu]

Summary

Add chat.variant plugin hook that allows plugins to programmatically control the thinking/variant level (high, max) before LLM requests are made.

Changes

• Add chat.variant hook type definition in @opencode/plugin
• Trigger hook in llm.ts before variant options are applied
• Plugin receives current variant state and can override via output.variant or output.options

Use Case

Plugins can now programmatically control:

• Anthropic's thinking budget (high/max)
• OpenAI's reasoning effort
• Other provider-specific variant options

This enables use cases like:

• Always use high thinking for specific models
• Dynamic thinking level based on conversation context
• Conditional variant selection based on session/agent type

Example Usage

const myPlugin: Plugin = async () => ({
  "chat.variant": async (input, output) => {
    // Always use high thinking for Claude
    if (input.model.id.includes("claude")) {
      output.variant = "high"
    }
  }
})

---

Generated with assistance of OhMyOpenCode

[rekram1-node]

I don't think u need a brand new plugin for it.... See here:

    const variant = input.model.variants && input.user.variant ? input.model.variants[input.user.variant] : {}
    const options = pipe(
      ProviderTransform.options(input.model, input.sessionID, provider.options),
      mergeDeep(small),
      mergeDeep(input.model.options),
      mergeDeep(input.agent.options),
      mergeDeep(variant),
    )


    const params = await Plugin.trigger(
      "chat.params",
      {
        sessionID: input.sessionID,
        agent: input.agent,
        model: input.model,
        provider: Provider.getProvider(input.model.providerID),
        message: input.user,
      },
      {
        temperature: input.model.capabilities.temperature
          ? (input.agent.temperature ?? ProviderTransform.temperature(input.model))
          : undefined,
        topP: input.agent.topP ?? ProviderTransform.topP(input.model),
        topK: ProviderTransform.topK(input.model),
        options, << Variant stuff will already be here
      },
    )

But maybe thats too weird...

[rekram1-node]

accidentally clicked close instead of comment oops

[code-yeongyu]

Thanks for the feedback! I initially created this hook because I thought the variant changes wouldn't be reflected in the TUI. Let me do some more testing to verify this and get back to you.

[rekram1-node]

ah thats a fair point, I think u need to set the variant on the message then in that case

[code-yeongyu]

I tried implementing it with chat.params like this:

"chat.params": async (input, output) => {
  const agent = input.agent as unknown as { name: string }
  const model = input.model as { variants?: Record<string, Record<string, unknown>> }
  const maxVariant = model.variants?.["max"]
  if (agent.name === "Sisyphus" && maxVariant) {
    output.options = { ...output.options, ...maxVariant }
  }
},

But the TUI does not reflect the change (no "max" indicator shown).

It would be pretty cool if the TUI could display "max" when the plugin sets it. Is doing this via plugin a bad approach?