From cdf373b77db58d590ceeb73706547a042d26e832 Mon Sep 17 00:00:00 2001
From: shoucandanghehe <wallfjjd@gmail.com>
Date: Sun, 26 Jul 2026 23:16:23 +0800
Subject: [PATCH 1/3] fix(coding-agent): handle asynchronous LSP pipe failures

---
 packages/coding-agent/CHANGELOG.md            |  6 +++
 packages/coding-agent/src/lsp/client.ts       | 40 ++++++++++--------
 .../test/tools/lsp-regressions.test.ts        | 42 +++++++++++++++++++
 3 files changed, 70 insertions(+), 18 deletions(-)

diff --git a/packages/coding-agent/CHANGELOG.md b/packages/coding-agent/CHANGELOG.md
index e4352d056..aa02c80e5 100644
--- a/packages/coding-agent/CHANGELOG.md
+++ b/packages/coding-agent/CHANGELOG.md
@@ -9,6 +9,12 @@
 - `omp usage` now surfaces auto-disabled credentials as red `✗` tombstone rows (identity, how long ago, the shortened upstream cause — e.g. `Refresh token expired` — and a re-login hint), including a provider section when no active credential remains. User-driven tombstones (`replaced by newer credential`, `deleted by user`) and API-key rows stay hidden. Requires a broker with `GET /v1/credentials/disabled`; older brokers degrade to no tombstone rows.
 - `omp usage` warns about Anthropic's ~30-day OAuth grant lifetime: accounts whose interactive login (`authorizedAt`) is within a week of the deadline get a yellow `⚠ re-login within <time>` line, and past-deadline accounts a red one. Grants die server-side exactly ~30 days after login regardless of refresh rotation, so this is the only warning before the broker auto-disables the row.
 
+### Fixed
+
+- Fixed Windows sessions crashing with an unhandled `EPIPE: broken pipe, write` when an LSP server closed its stdin between filesystem mutations; LSP writes now observe asynchronous `FileSink.write()` failures and route them through the existing request/notification failure path.
+- Fixed the Docker `natives-builder` stage failing to build releases ≥ 17.1.1: the native audio stack added bindgen (miniaudio needs libclang) and a bundled-opus CMake build (needs cmake + make), none of which were installed in the slim builder image.
+- Fixed `omp usage` duplicating org-less legacy accounts as "no usage data" rows whenever any sibling report carried an organization (mixed pools of pre-org-capture rows and fresh org-scoped logins): an org-less account is now covered by its own org-less report, while org-attributed sibling reports still never count as its coverage.
+- `omp usage` revalidates the broker credential snapshot before rendering: live usage reports were previously paired with a disk-cached account list up to an hour old, so a just-completed re-login (org-less row upserted to org-scoped) rendered as a phantom duplicate until the cache expired.
 ### Changed
 
 - Enabled Computer Use sessions now state the desktop-routing contract in the compact system prompt, retain their controller across model switches, expose effective native/function routing through `/computer status`, and emit structured lifecycle diagnostics without logging captured content.
diff --git a/packages/coding-agent/src/lsp/client.ts b/packages/coding-agent/src/lsp/client.ts
index c02e73cd6..cc7439b57 100644
--- a/packages/coding-agent/src/lsp/client.ts
+++ b/packages/coding-agent/src/lsp/client.ts
@@ -200,10 +200,10 @@ function abortReason(signal: AbortSignal): Error {
 	return signal.reason instanceof Error ? signal.reason : new ToolAbortError();
 }
 
-class LspFlushAbortError extends Error {
+class LspDrainAbortError extends Error {
 	constructor(readonly reason: Error) {
 		super(reason.message);
-		this.name = "LspFlushAbortError";
+		this.name = "LspDrainAbortError";
 	}
 }
 
@@ -216,26 +216,30 @@ async function writeMessage(
 		throw abortReason(signal);
 	}
 	const content = JSON.stringify(message);
-	sink.write(`Content-Length: ${Buffer.byteLength(content, "utf-8")}\r\n\r\n${content}`);
-	const flush = Promise.resolve(sink.flush());
+	const write = Promise.resolve(
+		sink.write(`Content-Length: ${Buffer.byteLength(content, "utf-8")}\r\n\r\n${content}`),
+	);
+	// Attach before flush(): it may throw synchronously after write() returned a
+	// rejected Promise, and leaving that rejection unobserved kills the host.
+	void write.catch(() => {});
+	const drain = Promise.all([write, Promise.resolve(sink.flush())]).then(() => {});
 	if (!signal) {
-		await flush;
+		await drain;
 		return;
 	}
-	// The sink's flush blocks on the OS-level pipe drain: if the server is
-	// alive but stopped reading stdin, `await sink.flush()` never resolves.
-	// Race the flush against the caller's signal so a wedged server surfaces
-	// as the tool's normal timeout/cancel instead of a permanent hang.
+	// Either sink operation can block on the OS-level pipe drain when a live
+	// server stops reading stdin. Race the combined drain against the caller's
+	// signal so a wedged server surfaces as the tool's normal timeout/cancel.
 	const { promise, resolve, reject } = Promise.withResolvers<void>();
 	const onAbort = () => {
 		signal.removeEventListener("abort", onAbort);
-		// The underlying flush stays pending in the background; suppress its
+		// The underlying drain stays pending in the background; suppress its
 		// eventual settlement so we do not surface an unhandled rejection.
-		flush.catch(() => {});
-		reject(new LspFlushAbortError(abortReason(signal)));
+		drain.catch(() => {});
+		reject(new LspDrainAbortError(abortReason(signal)));
 	};
 	signal.addEventListener("abort", onAbort, { once: true });
-	flush.then(
+	drain.then(
 		() => {
 			signal.removeEventListener("abort", onAbort);
 			resolve();
@@ -249,8 +253,8 @@ async function writeMessage(
 }
 
 /**
- * Kill a client whose write queue is stuck (aborted flush left the sink's
- * flush promise pending, so subsequent writes queue behind a wedge forever).
+ * Kill a client whose write queue is stuck (an aborted drain left a sink
+ * operation pending, so subsequent writes queue behind the wedge forever).
  * Remove it from `clients` immediately so concurrent `getOrCreateClient`
  * callers do not grab the corpse before `proc.exited` cleans up.
  */
@@ -270,8 +274,8 @@ function queueWriteMessage(
 ): Promise<void> {
 	const write = client.writeQueue.catch(() => {}).then(() => writeMessage(client.proc.stdin, message, signal));
 	const result = write.catch((err: unknown) => {
-		if (err instanceof LspFlushAbortError) {
-			// Only an abort that raced this write's in-flight flush leaves
+		if (err instanceof LspDrainAbortError) {
+			// Only an abort that raced this write's in-flight drain leaves
 			// the sink pending. Pre-write aborts and queued caller timeouts
 			// must not kill a healthy shared client.
 			teardownWedgedClient(client);
@@ -1067,7 +1071,7 @@ const WATCHED_FILES_NOTIFY_TIMEOUT_MS = 2_000;
  * This covers sibling files that are not open text documents, such as generated
  * CSS modules or type files that another edited document imports immediately.
  *
- * The underlying stdin flush is self-bounded by
+ * The underlying stdin write drain is self-bounded by
  * {@link WATCHED_FILES_NOTIFY_TIMEOUT_MS}; only an abort of the caller's
  * `signal` rejects.
  */
diff --git a/packages/coding-agent/test/tools/lsp-regressions.test.ts b/packages/coding-agent/test/tools/lsp-regressions.test.ts
index 0b562f9ed..a9b2470c9 100644
--- a/packages/coding-agent/test/tools/lsp-regressions.test.ts
+++ b/packages/coding-agent/test/tools/lsp-regressions.test.ts
@@ -3079,6 +3079,48 @@ describe("lsp regressions", () => {
 			expect(writes).toHaveLength(1);
 		});
 
+		it("surfaces an asynchronous stdin write rejection instead of resolving the notification", async () => {
+			const epipe = Object.assign(new Error("EPIPE: broken pipe, write"), {
+				code: "EPIPE",
+				syscall: "write",
+			});
+			const client: LspClient = {
+				name: "fake-lsp-write-epipe:/tmp",
+				cwd: "/tmp",
+				config: { command: "fake-lsp-write-epipe", fileTypes: ["ts"], rootMarkers: [] },
+				proc: {
+					exited: new Promise<number>(() => {}),
+					exitCode: null,
+					stdin: {
+						write: () => Promise.reject(epipe),
+						flush: () => 0,
+					},
+					stdout: new ReadableStream<Uint8Array>(),
+					peekStderr: () => "",
+					kill() {},
+				} as unknown as LspClient["proc"],
+				requestId: 0,
+				diagnostics: new Map(),
+				diagnosticsVersion: 0,
+				openFiles: new Map(),
+				pendingRequests: new Map(),
+				messageBuffer: new Uint8Array(0),
+				isReading: false,
+				status: "ready",
+				lastActivity: Date.now(),
+				writeQueue: Promise.resolve(),
+				activeProgressTokens: new Set(),
+				projectLoaded: Promise.resolve(),
+				resolveProjectLoaded: () => {},
+			};
+
+			await expect(
+				lspClient.sendNotification(client, "workspace/didChangeWatchedFiles", {
+					changes: [{ uri: "file:///tmp/example.ts", type: 2 }],
+				}),
+			).rejects.toBe(epipe);
+		});
+
 		it("bounds a wedged notification flush on the caller signal and tears down the client", async () => {
 			// Custom fake: stdin.flush is gated by a controllable promise so we
 			// can simulate a server that stopped draining stdin AFTER init has

From b61c864e04122f606a8a55cedfcb33fd9842cc7d Mon Sep 17 00:00:00 2001
From: shoucandanghehe <wallfjjd@gmail.com>
Date: Sun, 26 Jul 2026 23:38:55 +0800
Subject: [PATCH 2/3] chore(coding-agent): keep EPIPE fix unreleased

---
 packages/coding-agent/CHANGELOG.md | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/packages/coding-agent/CHANGELOG.md b/packages/coding-agent/CHANGELOG.md
index aa02c80e5..7f16ba2f5 100644
--- a/packages/coding-agent/CHANGELOG.md
+++ b/packages/coding-agent/CHANGELOG.md
@@ -2,6 +2,10 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Fixed Windows sessions crashing with an unhandled `EPIPE: broken pipe, write` when an LSP server closed its stdin between filesystem mutations; LSP writes now observe asynchronous `FileSink.write()` failures and route them through the existing request/notification failure path.
+
 ## [17.1.4] - 2026-07-26
 
 ### Added
@@ -9,12 +13,6 @@
 - `omp usage` now surfaces auto-disabled credentials as red `✗` tombstone rows (identity, how long ago, the shortened upstream cause — e.g. `Refresh token expired` — and a re-login hint), including a provider section when no active credential remains. User-driven tombstones (`replaced by newer credential`, `deleted by user`) and API-key rows stay hidden. Requires a broker with `GET /v1/credentials/disabled`; older brokers degrade to no tombstone rows.
 - `omp usage` warns about Anthropic's ~30-day OAuth grant lifetime: accounts whose interactive login (`authorizedAt`) is within a week of the deadline get a yellow `⚠ re-login within <time>` line, and past-deadline accounts a red one. Grants die server-side exactly ~30 days after login regardless of refresh rotation, so this is the only warning before the broker auto-disables the row.
 
-### Fixed
-
-- Fixed Windows sessions crashing with an unhandled `EPIPE: broken pipe, write` when an LSP server closed its stdin between filesystem mutations; LSP writes now observe asynchronous `FileSink.write()` failures and route them through the existing request/notification failure path.
-- Fixed the Docker `natives-builder` stage failing to build releases ≥ 17.1.1: the native audio stack added bindgen (miniaudio needs libclang) and a bundled-opus CMake build (needs cmake + make), none of which were installed in the slim builder image.
-- Fixed `omp usage` duplicating org-less legacy accounts as "no usage data" rows whenever any sibling report carried an organization (mixed pools of pre-org-capture rows and fresh org-scoped logins): an org-less account is now covered by its own org-less report, while org-attributed sibling reports still never count as its coverage.
-- `omp usage` revalidates the broker credential snapshot before rendering: live usage reports were previously paired with a disk-cached account list up to an hour old, so a just-completed re-login (org-less row upserted to org-scoped) rendered as a phantom duplicate until the cache expired.
 ### Changed
 
 - Enabled Computer Use sessions now state the desktop-routing contract in the compact system prompt, retain their controller across model switches, expose effective native/function routing through `/computer status`, and emit structured lifecycle diagnostics without logging captured content.

From 62e24ffc160d6f27fd56c0e767c505f2db4d84f1 Mon Sep 17 00:00:00 2001
From: shoucandanghehe <wallfjjd@gmail.com>
Date: Sun, 26 Jul 2026 23:56:27 +0800
Subject: [PATCH 3/3] test(coding-agent): follow promise fixture convention

---
 packages/coding-agent/test/tools/lsp-regressions.test.ts | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/packages/coding-agent/test/tools/lsp-regressions.test.ts b/packages/coding-agent/test/tools/lsp-regressions.test.ts
index a9b2470c9..4272a8e86 100644
--- a/packages/coding-agent/test/tools/lsp-regressions.test.ts
+++ b/packages/coding-agent/test/tools/lsp-regressions.test.ts
@@ -3089,7 +3089,7 @@ describe("lsp regressions", () => {
 				cwd: "/tmp",
 				config: { command: "fake-lsp-write-epipe", fileTypes: ["ts"], rootMarkers: [] },
 				proc: {
-					exited: new Promise<number>(() => {}),
+					exited: Promise.withResolvers<number>().promise,
 					exitCode: null,
 					stdin: {
 						write: () => Promise.reject(epipe),