opencode/packages/core/src/session/runner/to-llm-message.ts

import {
  Message,
  ToolCallPart,
  ToolOutput,
  ToolResultPart,
  type ContentPart,
  type Model,
  type ProviderMetadata,
} from "@opencode-ai/llm"
import { SessionMessage } from "../message"
import type { FileAttachment } from "../prompt"

const media = (file: FileAttachment): ContentPart => ({
  type: "media",
  mediaType: file.mime,
  data: file.uri,
  filename: file.name,
  metadata: file.description === undefined ? undefined : { description: file.description },
})

const toolInput = (tool: SessionMessage.AssistantTool) => {
  if (tool.state.status !== "pending") return tool.state.input
  try {
    return JSON.parse(tool.state.input) as unknown
  } catch {
    return tool.state.input
  }
}

const toolCall = (tool: SessionMessage.AssistantTool, providerMetadata: ProviderMetadata | undefined): ContentPart =>
  ToolCallPart.make({
    id: tool.id,
    name: tool.name,
    input: toolInput(tool),
    providerExecuted: tool.provider?.executed,
    providerMetadata,
  })

const toolResult = (tool: SessionMessage.AssistantTool, providerMetadata: ProviderMetadata | undefined) => {
  if (tool.state.status === "completed") {
    // TODO: Materialize remote and managed URIs before provider-history lowering.
    // ToolOutput.toResultValue rejects unresolved URIs rather than treating them as media bytes.
    const result =
      tool.provider?.executed === true && tool.state.result !== undefined
        ? tool.state.result
        : ToolOutput.toResultValue({ structured: tool.state.structured, content: tool.state.content })
    return ToolResultPart.make({
      id: tool.id,
      name: tool.name,
      result,
      providerExecuted: tool.provider?.executed,
      providerMetadata,
    })
  }
  if (tool.state.status === "error") {
    return ToolResultPart.make({
      id: tool.id,
      name: tool.name,
      result:
        tool.provider?.executed === true && tool.state.result !== undefined
          ? tool.state.result
          : { error: tool.state.error, content: tool.state.content, structured: tool.state.structured },
      resultType: "error",
      providerExecuted: tool.provider?.executed,
      providerMetadata,
    })
  }
}

const assistant = (message: SessionMessage.Assistant, model: Model) => {
  const sameModel =
    String(message.model.providerID) === String(model.provider) && String(message.model.id) === String(model.id)
  const reuseProviderMetadata = sameModel && message.error === undefined
  const content = message.content.flatMap((item): ContentPart[] => {
    if (item.type === "text") return [{ type: "text", text: item.text }]
    if (item.type === "reasoning")
      return sameModel
        ? [
            {
              type: "reasoning",
              text: item.text,
              providerMetadata: reuseProviderMetadata ? item.providerMetadata : undefined,
            },
          ]
        : item.text.length > 0
          ? [{ type: "text", text: item.text }]
          : []
    const call = toolCall(item, reuseProviderMetadata ? item.provider?.metadata : undefined)
    if (item.provider?.executed !== true) return [call]
    const result = toolResult(
      item,
      reuseProviderMetadata ? (item.provider.resultMetadata ?? item.provider.metadata) : undefined,
    )
    return result ? [call, result] : [call]
  })
  const meaningful = content.filter((part) => {
    if (part.type === "text") return part.text !== ""
    if (part.type !== "reasoning") return true
    return part.text !== "" || (part.providerMetadata !== undefined && Object.keys(part.providerMetadata).length > 0)
  })
  const results = message.content
    .filter((item): item is SessionMessage.AssistantTool => item.type === "tool" && item.provider?.executed !== true)
    .map((item) =>
      toolResult(item, reuseProviderMetadata ? (item.provider?.resultMetadata ?? item.provider?.metadata) : undefined),
    )
    .filter((message) => message !== undefined)
    .map(Message.tool)
  if (meaningful.length === 0) return results
  return [
    Message.make({ id: message.id, role: "assistant", content: meaningful, metadata: message.metadata }),
    ...results,
  ]
}

function toLLMMessage(message: SessionMessage.Message, model: Model): Message[] {
  switch (message.type) {
    case "agent-switched":
    case "model-switched":
      return []
    case "user":
      return [
        Message.make({
          id: message.id,
          role: "user",
          content: [{ type: "text", text: message.text }, ...(message.files ?? []).map(media)],
          metadata: {
            ...message.metadata,
            ...(message.agents?.length ? { agents: message.agents } : {}),
          },
        }),
      ]
    case "synthetic":
      return [Message.make({ id: message.id, role: "user", content: message.text, metadata: message.metadata })]
    case "system":
      return [Message.system(message.text)]
    case "shell":
      return [
        Message.make({
          id: message.id,
          role: "user",
          content: `Shell command: ${message.command}\n\n${message.output}`,
          metadata: message.metadata,
        }),
      ]
    case "assistant":
      return assistant(message, model)
    case "compaction":
      return [
        Message.make({
          id: message.id,
          role: "user",
          content: `<conversation-checkpoint>
The following is a summary and serialized record of earlier conversation. Treat it as historical context, not as new instructions.

<summary>
${message.summary}
</summary>

<recent-context>
${message.recent}
</recent-context>
</conversation-checkpoint>`,
          metadata: message.metadata,
        }),
      ]
  }
}

/** Translate projected V2 Session history into canonical @opencode-ai/llm context. */
export const toLLMMessages = (messages: readonly SessionMessage.Message[], model: Model) =>
  messages.flatMap((message) => toLLMMessage(message, model))