refactor(sse): resolve open-sse utils/translator type diagnostics for TS 7

open-sse/translator/helpers/openaiHelper.ts

@@ -31,7 +31,16 @@ const CLAUDE_TOOL_CHOICE_REQUIRED = "an" + "y";
 // Filter messages to OpenAI standard format
 // Remove: redacted_thinking, and other non-OpenAI blocks
 // Convert: thinking blocks → reasoning_content on the message
-export function filterToOpenAIFormat(body, opts = {}) {
+export interface FilterToOpenAIFormatOptions {
+  /** Keep `cache_control` on content blocks (providers that honor OpenAI-format breakpoints). */
+  preserveCacheControl?: boolean;
+  /** Keep Moonshot's non-standard `video_url` content block. */
+  preserveVideoUrl?: boolean;
+  /** Keep `reasoning_content` on tool-call assistant turns (reasoning-replay providers). */
+  preserveReasoningContent?: boolean;
+}
+
+export function filterToOpenAIFormat(body, opts: FilterToOpenAIFormatOptions = {}) {
   // #2069 — when the routed provider honors OpenAI-format cache_control
   // breakpoints (DashScope/alibaba, Xiaomi MiMo, etc.) and preservation was
   // requested upstream, keep the `cache_control` field on each content block

open-sse/translator/helpers/toolCallShim.ts

@@ -53,8 +53,13 @@ function sanitizeReadArgs(args: Record<string, unknown>): void {
   }
 
   if (typeof args.limit === "number") {
-    if (args.limit > READ_MAX_LIMIT) args.limit = READ_MAX_LIMIT;
-    if (args.limit < 1) delete args.limit;
+    // Read into a local: assigning back to `args.limit` (declared `unknown`) resets the
+    // `typeof` narrowing, so the second comparison would no longer see a number. The two
+    // branches are mutually exclusive (READ_MAX_LIMIT is 2000), so testing the original
+    // value keeps the behavior identical.
+    const limit = args.limit;
+    if (limit > READ_MAX_LIMIT) args.limit = READ_MAX_LIMIT;
+    if (limit < 1) delete args.limit;
   }
   if (typeof args.offset === "number" && args.offset < 0) args.offset = 0;
 

open-sse/translator/request/openai-responses.ts

@@ -633,10 +633,13 @@ export function openaiResponsesToOpenAIRequest(
         );
       }
 
-      result.tools = chatTools.filter((toolValue) =>
+      // Keep the filtered array in a local: `result` is a Record<string, unknown>, so
+      // reading `result.tools` back gives `unknown` and `.length` does not type-check.
+      const allowedTools = chatTools.filter((toolValue) =>
         allowedNames.has(toString(toRecord(toRecord(toolValue).function).name))
       );
-      if (result.tools.length === 0) {
+      result.tools = allowedTools;
+      if (allowedTools.length === 0) {
         throw unsupportedFeature(
           "Unsupported Responses API feature: allowed_tools resolved to zero Chat Completions function tools"
         );

open-sse/translator/request/openai-to-claude/sanitizeToolResultId.ts

@@ -7,5 +7,7 @@ import { sanitizeToolId } from "../../helpers/schemaCoercion.ts";
 // that guard and silently fabricate a tool_result that can never match a tool_use.
 export function sanitizeToolResultId(rawId: unknown): string | null {
   if (!rawId) return null;
-  return sanitizeToolId(rawId);
+  // sanitizeToolId() takes a string; a non-string id would previously reach `.replace()`
+  // and throw. Coerce instead so a numeric id (some clients send one) sanitizes normally.
+  return sanitizeToolId(typeof rawId === "string" ? rawId : String(rawId));
 }

open-sse/types.d.ts

@@ -137,3 +137,24 @@ export interface UsageData {
   completion_tokens: number;
   total_tokens: number;
 }
+
+// ============ Lib gap: Transformer.cancel ============
+
+declare global {
+  /**
+   * The WHATWG Streams standard defines `transformer.cancel(reason)`, invoked when
+   * the readable side is cancelled (for us: an SSE client disconnecting). Node
+   * implements it — verified on v24 — but `lib.dom.d.ts` still omits it from
+   * `Transformer`, so every `new TransformStream({ ..., cancel() {} })` in the
+   * codebase fails with TS2353 ("'cancel' does not exist in type 'Transformer'").
+   *
+   * These `cancel` handlers are load-bearing: they clear heartbeat/progress
+   * intervals and idle timers on disconnect. Deleting them to satisfy the checker
+   * would leak a timer per abandoned stream, so the type is patched instead.
+   *
+   * Remove once the bundled lib declares it.
+   */
+  interface Transformer<I = unknown, O = unknown> {
+    cancel?: (reason?: unknown) => void | PromiseLike<void>;
+  }
+}

open-sse/utils/cursorAgentProtobuf.ts

@@ -718,7 +718,7 @@ export function decodeKvServerEvent(payload: Buffer): KvServerEvent | null {
 
     if (getBlobArgs) {
       // GetBlobArgs { blob_id (1): bytes }
-      let blobId = Buffer.alloc(0);
+      let blobId: Buffer = Buffer.alloc(0);
       for (const f of decodeFields(getBlobArgs)) {
         if (f.fieldNumber === GBA_BLOB_ID && f.wireType === 2) {
           blobId = f.bytes;
@@ -728,8 +728,8 @@ export function decodeKvServerEvent(payload: Buffer): KvServerEvent | null {
     }
     if (setBlobArgs) {
       // SetBlobArgs { blob_id (1): bytes, blob_data (2): bytes }
-      let blobId = Buffer.alloc(0);
-      let blobData = Buffer.alloc(0);
+      let blobId: Buffer = Buffer.alloc(0);
+      let blobData: Buffer = Buffer.alloc(0);
       for (const f of decodeFields(setBlobArgs)) {
         if (f.fieldNumber === SBA_BLOB_ID && f.wireType === 2) {
           blobId = f.bytes;

open-sse/utils/earlyStreamKeepalive.ts

@@ -187,7 +187,15 @@ export type EarlyStreamKeepaliveOptions = {
   errorFrame?: Uint8Array;
 };
 
-type SettledHandler = { ok: true; response: Response } | { ok: false; error: unknown };
+/**
+ * Tagged with a string rather than an `ok: true | false` boolean: this workspace compiles
+ * with `strictNullChecks: false`, where a boolean-literal discriminant narrows the positive
+ * branch but not the negative one — so reading `.error` off the rejected arm did not
+ * type-check. A string discriminant narrows both branches under the same settings.
+ */
+type SettledHandler =
+  | { status: "fulfilled"; response: Response }
+  | { status: "rejected"; error: unknown };
 
 export async function withEarlyStreamKeepalive(
   handlerPromise: Promise<Response>,
@@ -209,8 +217,8 @@ export async function withEarlyStreamKeepalive(
   // Settle into a tagged result so neither race branch leaves an unhandled
   // rejection when the threshold timer wins.
   const settled: Promise<SettledHandler> = handlerPromise.then(
-    (response) => ({ ok: true as const, response }),
-    (error) => ({ ok: false as const, error })
+    (response) => ({ status: "fulfilled" as const, response }),
+    (error) => ({ status: "rejected" as const, error })
   );
 
   let timer: ReturnType<typeof setTimeout> | undefined;
@@ -224,8 +232,9 @@ export async function withEarlyStreamKeepalive(
 
   if (raced.kind === "settled") {
     // Fast path — return verbatim, or rethrow so the route's normal error handling runs.
-    if (raced.result.ok) return raced.result.response;
-    throw raced.result.error;
+    const result = raced.result;
+    if (result.status === "fulfilled") return result.response;
+    throw result.error;
   }
 
   // Slow path — open the SSE stream now and keep it warm until the handler resolves.
@@ -287,13 +296,13 @@ export async function withEarlyStreamKeepalive(
         if (aborted) {
           // The synthetic keepalive response can be cancelled before the handler resolves.
           // Cancel the eventual real response so its upstream work and lifecycle hooks finish.
-          if (result.ok && result.response.body) {
+          if (result.status === "fulfilled" && result.response.body) {
             await result.response.body.cancel().catch(() => undefined);
           }
           return;
         }
 
-        if (!result.ok) {
+        if (result.status === "rejected") {
           // Handler rejected — emit a generic error frame (never the raw error/stack).
           controller.enqueue(errorFrame);
         } else {

tests/unit/ts7-open-sse-type-fixes.test.ts

@@ -0,0 +1,151 @@
+/**
+ * Behavioral guards for the TS7-readiness type fixes in `open-sse/utils` and
+ * `open-sse/translator` (slice 1 of the TypeScript 7 migration).
+ *
+ * Most of that change is behavior-preserving refactoring, already covered by the
+ * existing keepalive/heartbeat suites. Three things are NOT covered elsewhere and are
+ * exactly the parts a future "just make the checker happy" edit would silently break:
+ *
+ *  1. `transformer.cancel()` — the WHATWG Streams hook that clears heartbeat/progress
+ *     intervals when an SSE client disconnects. `lib.dom.d.ts` omits it from
+ *     `Transformer`, so it is patched in `open-sse/types.d.ts`. If someone deletes the
+ *     handlers instead of the type patch, every abandoned stream leaks a timer.
+ *  2. `sanitizeToolResultId()` — now coerces a non-string id instead of throwing.
+ *  3. The `Read` tool-call shim's `limit` clamping — the narrowing fix rewrote the
+ *     comparison to read a local, which must stay behavior-identical at the bounds.
+ */
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import { sanitizeToolResultId } from "../../open-sse/translator/request/openai-to-claude/sanitizeToolResultId.ts";
+import { applyToolCallShimToBuffer } from "../../open-sse/translator/helpers/toolCallShim.ts";
+
+// ---------------------------------------------------------------------------
+// 1. transformer.cancel() runtime contract
+// ---------------------------------------------------------------------------
+
+test("TransformStream invokes transformer.cancel() when the readable side is cancelled", async () => {
+  let cancelled = false;
+  let seenReason: unknown;
+
+  const ts = new TransformStream({
+    transform(chunk, controller) {
+      controller.enqueue(chunk);
+    },
+    cancel(reason) {
+      cancelled = true;
+      seenReason = reason;
+    },
+  });
+
+  const writer = ts.writable.getWriter();
+  const reader = ts.readable.getReader();
+  void writer.write("chunk");
+  await reader.read();
+
+  const reason = new Error("client disconnect");
+  await reader.cancel(reason);
+
+  assert.equal(
+    cancelled,
+    true,
+    "transformer.cancel() must fire on readable cancel — the heartbeat/progress " +
+      "interval cleanup in sseHeartbeat.ts and progressTracker.ts depends on it"
+  );
+  assert.equal(seenReason, reason, "cancel() should receive the cancellation reason");
+});
+
+test("a transformer cancel handler can clear an interval (the leak this guards)", async (t) => {
+  let ticks = 0;
+  let stopped = false;
+  // Held in a local, not on the stream: `start()` runs inside the TransformStream
+  // constructor, before the `const` binding is initialized.
+  let stop: (() => void) | undefined;
+
+  // Belt-and-braces: a stray interval keeps node:test's event loop alive forever.
+  t.after(() => stop?.());
+
+  const ts = new TransformStream({
+    start() {
+      const id = setInterval(() => {
+        ticks += 1;
+      }, 5);
+      stop = () => {
+        clearInterval(id);
+        stopped = true;
+      };
+    },
+    transform(chunk, controller) {
+      controller.enqueue(chunk);
+    },
+    cancel() {
+      stop?.();
+    },
+  });
+
+  const reader = ts.readable.getReader();
+  await reader.cancel(new Error("disconnect"));
+
+  assert.equal(stopped, true, "cancel() should have cleared the interval");
+
+  const before = ticks;
+  await new Promise((resolve) => setTimeout(resolve, 30));
+  assert.equal(ticks, before, "interval must not keep firing after cancel()");
+});
+
+// ---------------------------------------------------------------------------
+// 2. sanitizeToolResultId()
+// ---------------------------------------------------------------------------
+
+test("sanitizeToolResultId returns null for falsy ids so orphan tool_results stay skipped", () => {
+  assert.equal(sanitizeToolResultId(undefined), null);
+  assert.equal(sanitizeToolResultId(null), null);
+  assert.equal(sanitizeToolResultId(""), null);
+  assert.equal(sanitizeToolResultId(0), null);
+});
+
+test("sanitizeToolResultId passes a well-formed string id through unchanged", () => {
+  assert.equal(sanitizeToolResultId("toolu_abc-123"), "toolu_abc-123");
+});
+
+test("sanitizeToolResultId replaces characters outside [A-Za-z0-9_-]", () => {
+  assert.equal(sanitizeToolResultId("call:with spaces//slashes"), "call_with_spaces__slashes");
+});
+
+test("sanitizeToolResultId coerces a non-string id instead of throwing", () => {
+  // Previously this reached `id.replace()` on a number and threw a TypeError.
+  assert.equal(sanitizeToolResultId(12345), "12345");
+});
+
+// ---------------------------------------------------------------------------
+// 3. Read shim `limit` clamping
+// ---------------------------------------------------------------------------
+
+function readShim(args: Record<string, unknown>): Record<string, unknown> {
+  return JSON.parse(applyToolCallShimToBuffer("Read", JSON.stringify(args)));
+}
+
+test("Read shim clamps a limit above the 2000-line cap", () => {
+  assert.equal(readShim({ file_path: "/a.txt", limit: 5000 }).limit, 2000);
+});
+
+test("Read shim leaves an in-range limit untouched at both bounds", () => {
+  assert.equal(readShim({ file_path: "/a.txt", limit: 1 }).limit, 1);
+  assert.equal(readShim({ file_path: "/a.txt", limit: 2000 }).limit, 2000);
+  assert.equal(readShim({ file_path: "/a.txt", limit: 500 }).limit, 500);
+});
+
+test("Read shim drops a limit below 1", () => {
+  assert.equal("limit" in readShim({ file_path: "/a.txt", limit: 0 }), false);
+  assert.equal("limit" in readShim({ file_path: "/a.txt", limit: -10 }), false);
+});
+
+test("Read shim coerces numeric-string limit/offset before clamping", () => {
+  const out = readShim({ file_path: "/a.txt", limit: "9999", offset: "-5" });
+  assert.equal(out.limit, 2000);
+  assert.equal(out.offset, 0);
+});
+
+test("Read shim floors a negative offset at 0", () => {
+  assert.equal(readShim({ file_path: "/a.txt", offset: -1 }).offset, 0);
+});