fix(ai): strip partialJson from responses tool calls

Mutate persisted tool-call blocks in place on function_call completion,
remove partialJson, and emit the same reference on toolcall_end.
Add regression coverage for persisted block cleanup and event identity.

fixes #3078

packages/ai/src/providers/amazon-bedrock.ts

@@ -211,6 +211,7 @@ export const streamBedrock: StreamFunction<"bedrock-converse-stream", BedrockOpt
 		} catch (error) {
 			for (const block of output.content) {
 				delete (block as Block).index;
+				// partialJson is only a streaming scratch buffer; never persist it.
 				delete (block as Block).partialJson;
 			}
 			output.stopReason = options.signal?.aborted ? "aborted" : "error";
@@ -413,6 +414,8 @@ function handleContentBlockStop(
 			break;
 		case "toolCall":
 			block.arguments = parseStreamingJson(block.partialJson);
+			// Finalize in-place and strip the scratch buffer so replay only
+			// carries parsed arguments.
 			delete (block as Block).partialJson;
 			stream.push({ type: "toolcall_end", contentIndex: index, toolCall: block, partial: output });
 			break;

packages/ai/src/providers/anthropic.ts

@@ -384,7 +384,9 @@ export const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOpti
 							});
 						} else if (block.type === "toolCall") {
 							block.arguments = parseStreamingJson(block.partialJson);
-							delete (block as any).partialJson;
+							// Finalize in-place and strip the scratch buffer so replay only
+							// carries parsed arguments.
+							delete (block as { partialJson?: string }).partialJson;
 							stream.push({
 								type: "toolcall_end",
 								contentIndex: index,
@@ -429,7 +431,11 @@ export const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOpti
 			stream.push({ type: "done", reason: output.stopReason, message: output });
 			stream.end();
 		} catch (error) {
-			for (const block of output.content) delete (block as any).index;
+			for (const block of output.content) {
+				delete (block as { index?: number }).index;
+				// partialJson is only a streaming scratch buffer; never persist it.
+				delete (block as { partialJson?: string }).partialJson;
+			}
 			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
 			output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
 			stream.push({ type: "error", reason: output.stopReason, error: output });

packages/ai/src/providers/azure-openai-responses.ts

@@ -109,7 +109,11 @@ export const streamAzureOpenAIResponses: StreamFunction<"azure-openai-responses"
 			stream.push({ type: "done", reason: output.stopReason, message: output });
 			stream.end();
 		} catch (error) {
-			for (const block of output.content) delete (block as { index?: number }).index;
+			for (const block of output.content) {
+				delete (block as { index?: number }).index;
+				// partialJson is only a streaming scratch buffer; never persist it.
+				delete (block as { partialJson?: string }).partialJson;
+			}
 			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
 			output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
 			stream.push({ type: "error", reason: output.stopReason, error: output });

packages/ai/src/providers/mistral.ts

@@ -89,6 +89,10 @@ export const streamMistral: StreamFunction<"mistral-conversations", MistralOptio
 			stream.push({ type: "done", reason: output.stopReason, message: output });
 			stream.end();
 		} catch (error) {
+			for (const block of output.content) {
+				// partialArgs is only a streaming scratch buffer; never persist it.
+				delete (block as { partialArgs?: string }).partialArgs;
+			}
 			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
 			output.errorMessage = formatMistralError(error);
 			stream.push({ type: "error", reason: output.stopReason, error: output });
@@ -424,6 +428,8 @@ async function consumeChatStream(
 		if (block.type !== "toolCall") continue;
 		const toolBlock = block as ToolCall & { partialArgs?: string };
 		toolBlock.arguments = parseStreamingJson<Record<string, unknown>>(toolBlock.partialArgs);
+		// Finalize in-place and strip the scratch buffer so replay only
+		// carries parsed arguments.
 		delete toolBlock.partialArgs;
 		stream.push({
 			type: "toolcall_end",

packages/ai/src/providers/openai-codex-responses.ts

@@ -268,6 +268,10 @@ export const streamOpenAICodexResponses: StreamFunction<"openai-codex-responses"
 			stream.push({ type: "done", reason: output.stopReason as "stop" | "length" | "toolUse", message: output });
 			stream.end();
 		} catch (error) {
+			for (const block of output.content) {
+				// partialJson is only a streaming scratch buffer; never persist it.
+				delete (block as { partialJson?: string }).partialJson;
+			}
 			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
 			output.errorMessage = error instanceof Error ? error.message : String(error);
 			stream.push({ type: "error", reason: output.stopReason, error: output });

packages/ai/src/providers/openai-completions.ts

@@ -115,6 +115,8 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions", OpenA
 						});
 					} else if (block.type === "toolCall") {
 						block.arguments = parseStreamingJson(block.partialArgs);
+						// Finalize in-place and strip the scratch buffer so replay only
+						// carries parsed arguments.
 						delete block.partialArgs;
 						stream.push({
 							type: "toolcall_end",
@@ -289,7 +291,11 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions", OpenA
 			stream.push({ type: "done", reason: output.stopReason, message: output });
 			stream.end();
 		} catch (error) {
-			for (const block of output.content) delete (block as any).index;
+			for (const block of output.content) {
+				delete (block as { index?: number }).index;
+				// partialArgs is only a streaming scratch buffer; never persist it.
+				delete (block as { partialArgs?: string }).partialArgs;
+			}
 			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
 			output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
 			// Some providers via OpenRouter give additional information in this field.

packages/ai/src/providers/openai-responses-shared.ts

@@ -452,12 +452,22 @@ export async function processResponsesStream<TApi extends Api>(
 					currentBlock?.type === "toolCall" && currentBlock.partialJson
 						? parseStreamingJson(currentBlock.partialJson)
 						: parseStreamingJson(item.arguments || "{}");
-				const toolCall: ToolCall = {
-					type: "toolCall",
-					id: `${item.call_id}|${item.id}`,
-					name: item.name,
-					arguments: args,
-				};
+
+				let toolCall: ToolCall;
+				if (currentBlock?.type === "toolCall") {
+					// Finalize in-place and strip the scratch buffer so replay only
+					// carries parsed arguments.
+					currentBlock.arguments = args;
+					delete (currentBlock as { partialJson?: string }).partialJson;
+					toolCall = currentBlock;
+				} else {
+					toolCall = {
+						type: "toolCall",
+						id: `${item.call_id}|${item.id}`,
+						name: item.name,
+						arguments: args,
+					};
+				}
 
 				currentBlock = null;
 				stream.push({ type: "toolcall_end", contentIndex: blockIndex(), toolCall, partial: output });

packages/ai/src/providers/openai-responses.ts

@@ -116,7 +116,11 @@ export const streamOpenAIResponses: StreamFunction<"openai-responses", OpenAIRes
 			stream.push({ type: "done", reason: output.stopReason, message: output });
 			stream.end();
 		} catch (error) {
-			for (const block of output.content) delete (block as { index?: number }).index;
+			for (const block of output.content) {
+				delete (block as { index?: number }).index;
+				// partialJson is only a streaming scratch buffer; never persist it.
+				delete (block as { partialJson?: string }).partialJson;
+			}
 			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
 			output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
 			stream.push({ type: "error", reason: output.stopReason, error: output });

packages/ai/test/openai-responses-partial-json-cleanup.test.ts

@@ -0,0 +1,101 @@
+import type { ResponseStreamEvent } from "openai/resources/responses/responses.js";
+import { describe, expect, it, vi } from "vitest";
+import { processResponsesStream } from "../src/providers/openai-responses-shared.js";
+import type { AssistantMessage, AssistantMessageEvent, Model } from "../src/types.js";
+import { AssistantMessageEventStream } from "../src/utils/event-stream.js";
+
+function createOutput(model: Model<"openai-responses">): AssistantMessage {
+	return {
+		role: "assistant",
+		content: [],
+		api: model.api,
+		provider: model.provider,
+		model: model.id,
+		usage: {
+			input: 0,
+			output: 0,
+			cacheRead: 0,
+			cacheWrite: 0,
+			totalTokens: 0,
+			cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+		},
+		stopReason: "stop",
+		timestamp: Date.now(),
+	};
+}
+
+async function* createFunctionCallEvents(argumentsJson: string): AsyncIterable<ResponseStreamEvent> {
+	yield {
+		type: "response.output_item.added",
+		item: {
+			type: "function_call",
+			id: "fc_test",
+			call_id: "call_test",
+			name: "edit",
+			arguments: "",
+		},
+	} as ResponseStreamEvent;
+	yield {
+		type: "response.function_call_arguments.delta",
+		delta: '{"path":"README.md"',
+	} as ResponseStreamEvent;
+	yield {
+		type: "response.function_call_arguments.delta",
+		delta: ',"content":"updated"}',
+	} as ResponseStreamEvent;
+	yield {
+		type: "response.function_call_arguments.done",
+		arguments: argumentsJson,
+	} as ResponseStreamEvent;
+	yield {
+		type: "response.output_item.done",
+		item: {
+			type: "function_call",
+			id: "fc_test",
+			call_id: "call_test",
+			name: "edit",
+			arguments: argumentsJson,
+		},
+	} as ResponseStreamEvent;
+}
+
+describe("openai responses partialJson cleanup", () => {
+	it("removes partialJson from persisted tool-call blocks at output_item.done", async () => {
+		const model: Model<"openai-responses"> = {
+			id: "gpt-5-mini",
+			name: "GPT-5 Mini",
+			api: "openai-responses",
+			provider: "openai",
+			baseUrl: "https://api.openai.com/v1",
+			reasoning: true,
+			input: ["text"],
+			cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+			contextWindow: 400000,
+			maxTokens: 128000,
+		};
+		const output = createOutput(model);
+		const stream = new AssistantMessageEventStream();
+		const pushSpy = vi.spyOn(stream, "push");
+		const argumentsJson = '{"path":"README.md","content":"updated"}';
+
+		await processResponsesStream(createFunctionCallEvents(argumentsJson), output, stream, model);
+
+		expect(output.content).toHaveLength(1);
+		const persistedToolCall = output.content[0];
+		expect(persistedToolCall?.type).toBe("toolCall");
+		if (!persistedToolCall || persistedToolCall.type !== "toolCall") {
+			throw new Error("Expected toolCall block");
+		}
+		expect(persistedToolCall.arguments).toEqual({ path: "README.md", content: "updated" });
+		expect("partialJson" in persistedToolCall).toBe(false);
+
+		const emittedEvents = pushSpy.mock.calls.map(([event]) => event as AssistantMessageEvent);
+		const toolCallEnd = emittedEvents.find((event) => event.type === "toolcall_end");
+		expect(toolCallEnd).toBeDefined();
+		if (!toolCallEnd || toolCallEnd.type !== "toolcall_end") {
+			throw new Error("Expected toolcall_end event");
+		}
+		expect(toolCallEnd.toolCall).toBe(persistedToolCall);
+		expect("partialJson" in toolCallEnd.toolCall).toBe(false);
+	});
+});