fix(chat): an edit after a broken edit call no longer fails, found with Opus 5.5

- Claude Opus 5.5 sent an edit with invalid JSON, then the same edit
  again. The first call's streamed preview was never undone: its input
  has no operations, and the undo sat behind that check. The second edit
  then started from the preview, failed on a duplicate id, and the model
  had to try a third time. The undo now runs first, and an edit that
  starts in the same render uses the undone diagram.
- The SDK passes an invalid tool call's error as a string, which was
  wrapped as a provider error. streamErrorText keeps it as the text the
  model reads.
- Bedrock's "on-demand throughput isn't supported" gets the model id hint.
- The thinking header uses the page language ("Thought for 1 second" in
  English), from the dictionary entries that were already there.
This commit is contained in:
dayuan.jiang
2026-10-04 21:18:54 +09:00
parent 65e3dd1dde
commit 2ac4c54eb2
9 changed files with 271 additions and 18 deletions
+81 -2
View File
@@ -1,7 +1,20 @@
// @vitest-environment node
import { APICallError, InvalidToolInputError, RetryError } from "ai"
import {
APICallError,
InvalidToolInputError,
RetryError,
simulateReadableStream,
streamText,
tool,
} from "ai"
import { MockLanguageModelV3 } from "ai/test"
import { describe, expect, it } from "vitest"
import { classifyLLMError, isToolCallError } from "@/lib/llm-errors"
import { z } from "zod"
import {
classifyLLMError,
isToolCallError,
streamErrorText,
} from "@/lib/llm-errors"
const apiError = (statusCode: number, message: string, responseBody = "") =>
new APICallError({
@@ -89,6 +102,14 @@ describe("classifyLLMError", () => {
expect(classifyLLMError(timeout).code).toBe("timeout")
})
it("points to the model id when Bedrock wants an inference profile", () => {
const error = apiError(
400,
"Invocation of model ID anthropic.claude-sonnet-5-5 with on-demand throughput isn’t supported. Retry your request with the ID or ARN of an inference profile that contains this model.",
)
expect(classifyLLMError(error).code).toBe("model_not_found")
})
it("names a network error the SDK wrapped", () => {
const error = new APICallError({
message:
@@ -131,6 +152,64 @@ describe("classifyLLMError", () => {
})
})
describe("streamErrorText", () => {
it("keeps the text of a tool call the model got wrong", async () => {
// Seen with Claude Opus 5.5: a quote left unescaped in the input
const model = new MockLanguageModelV3({
doStream: (async () => ({
stream: simulateReadableStream({
chunks: [
{
type: "tool-call",
toolCallId: "c1",
toolName: "edit_diagram",
input: '{"operations": [{"new_xml": "as="x""}]}',
},
{
type: "finish",
finishReason: {
unified: "tool-calls",
raw: "tool_use",
},
usage: {
inputTokens: { total: 1 },
outputTokens: { total: 1 },
},
},
],
}),
})) as any,
})
const result = streamText({
model: model as any,
prompt: "edit",
tools: {
edit_diagram: tool({
inputSchema: z.object({ operations: z.array(z.any()) }),
}),
},
})
const errors: string[] = []
for await (const chunk of result.toUIMessageStream({
onError: streamErrorText,
})) {
if ("errorText" in chunk) errors.push(chunk.errorText)
}
expect(errors.length).toBeGreaterThan(0)
for (const text of errors) {
expect(text).toMatch(/^Invalid input for tool edit_diagram/)
}
})
it("classifies a provider error", () => {
expect(JSON.parse(streamErrorText(apiError(401, "bad key")))).toEqual({
type: "provider",
code: "invalid_api_key",
message: "bad key",
})
})
})
describe("isToolCallError", () => {
it("spots errors the model must see unchanged", () => {
const invalid = new InvalidToolInputError({