fix(session): prevent GPT-5.2 resume reasoning crash

packages/opencode/src/session/message-v2.ts

@@ -419,7 +419,13 @@ export namespace MessageV2 {
   })
   export type WithParts = z.infer<typeof WithParts>
 
-  export function toModelMessage(input: WithParts[]): ModelMessage[] {
+  export function toModelMessage(
+    input: WithParts[],
+    options?: {
+      dropReasoningOnlyAssistantMessages?: boolean
+    },
+  ): ModelMessage[] {
+    const drop = options?.dropReasoningOnlyAssistantMessages === true
     const result: UIMessage[] = []
 
     for (const msg of input) {
@@ -536,6 +542,10 @@ export namespace MessageV2 {
             })
           }
         }
+
+        if (drop && assistantMessage.parts.every((part) => part.type === "step-start" || part.type === "reasoning")) {
+          if (result[result.length - 1] === assistantMessage) result.pop()
+        }
       }
     }
 

packages/opencode/src/session/prompt.ts

@@ -527,14 +527,17 @@ export namespace SessionPrompt {
 
       await Plugin.trigger("experimental.chat.messages.transform", {}, { messages: sessionMessages })
 
+      const drop = model.api.id.startsWith("gpt-5.2")
       const result = await processor.process({
         user: lastUser,
         agent,
         abort,
         sessionID,
         system: [...(await SystemPrompt.environment()), ...(await SystemPrompt.custom())],
         messages: [
-          ...MessageV2.toModelMessage(sessionMessages),
+          ...MessageV2.toModelMessage(sessionMessages, {
+            dropReasoningOnlyAssistantMessages: drop,
+          }),
           ...(isLastStep
             ? [
                 {

packages/opencode/test/session/message-v2.test.ts

@@ -0,0 +1,93 @@
+import { describe, expect, test } from "bun:test"
+import { MessageV2 } from "../../src/session/message-v2"
+
+describe("MessageV2.toModelMessage", () => {
+  const sid = "session_1"
+  const mid = "message_1"
+  const pid = "message_0"
+
+  function assistant(parts: MessageV2.Part[]): MessageV2.WithParts {
+    return {
+      info: {
+        id: mid,
+        sessionID: sid,
+        role: "assistant",
+        time: {
+          created: 0,
+        },
+        parentID: pid,
+        modelID: "gpt-5.2",
+        providerID: "openai",
+        mode: "chat",
+        agent: "agent",
+        path: {
+          cwd: "/",
+          root: "/",
+        },
+        cost: 0,
+        tokens: {
+          input: 0,
+          output: 0,
+          reasoning: 0,
+          cache: {
+            read: 0,
+            write: 0,
+          },
+        },
+      },
+      parts,
+    }
+  }
+
+  function step(id: string): MessageV2.StepStartPart {
+    return {
+      id,
+      sessionID: sid,
+      messageID: mid,
+      type: "step-start",
+    }
+  }
+
+  function think(id: string): MessageV2.ReasoningPart {
+    return {
+      id,
+      sessionID: sid,
+      messageID: mid,
+      type: "reasoning",
+      text: "Let me think...",
+      time: {
+        start: 0,
+      },
+    }
+  }
+
+  function say(id: string): MessageV2.TextPart {
+    return {
+      id,
+      sessionID: sid,
+      messageID: mid,
+      type: "text",
+      text: "Answer",
+    }
+  }
+
+  test("drops assistant messages with only reasoning/step-start parts when enabled", () => {
+    const messages = [assistant([step("part_step"), think("part_reasoning")])]
+
+    expect(MessageV2.toModelMessage(messages, { dropReasoningOnlyAssistantMessages: true })).toEqual([])
+  })
+
+  test("keeps assistant messages with only reasoning by default", () => {
+    const messages = [assistant([think("part_reasoning")])]
+
+    expect(MessageV2.toModelMessage(messages)).toHaveLength(1)
+  })
+
+  test("keeps assistant messages when non-reasoning content exists", () => {
+    const messages = [assistant([think("part_reasoning"), say("part_text")])]
+
+    const result = MessageV2.toModelMessage(messages)
+    expect(result).toHaveLength(1)
+    expect(result[0].role).toBe("assistant")
+  })
+})