@gajae-code/coding-agent 0.18.12 → 0.19.0

package/src/sdk/broker/session-index.ts

@@ -4,6 +4,7 @@ import path from "node:path";
 import { logger, resolveEquivalentPath } from "@gajae-code/utils";
 import { nativeProcessBindings } from "@gajae-code/utils/native-process";
 import {
+	describeRetainedLockTransition,
 	FILE_LOCK_ORPHAN_TRANSITION_ADOPTION_AGE_MS,
 	type FileLockOptions,
 	isFileLockAcquireTimeout,
@@ -11,6 +12,7 @@ import {
 } from "../../config/file-lock";
 import { repo } from "../../utils/git";
 import { endpointIncarnation } from "./endpoint-authority";
+import { brokerLockHostIdentity } from "./lock-host-identity";
 import { processIncarnation } from "./process-incarnation";
 import {
 	assertSupportedSessionIndexEventVersion,
@@ -848,7 +850,7 @@ const SESSION_INDEX_REPLAY_FRESHNESS_MS = 2_000;
  */
 const SESSION_HEARTBEAT_CHECKPOINT_ATTEMPTS = 4;
 /** One elapsed budget for queueing, probing and every reacquisition in a pass. */
-const SESSION_HEARTBEAT_CHECKPOINT_BUDGET_MS = 15_000;
+export const SESSION_HEARTBEAT_CHECKPOINT_BUDGET_MS = 15_000;
 const STALE_OBSERVATION: unique symbol = Symbol("stale-heartbeat-observation");
 /** Maximum distinct live identity probes admitted by one public status query. */
 const SESSION_GENERATION_PROBE_LIMIT = 32;
@@ -859,13 +861,26 @@ const SESSION_GENERATION_PROBE_LIMIT = 32;
  * is named in a slow-operation warning when it holds the lock beyond the
  * observation window and (b) the lock budget cannot silently drift per call
  * site.
+ *
+ * Every acquisition also carries this installation's host identity.
+ * `staleLockSnapshot` treats a host-qualified owner as foreign when the acquirer
+ * omits `ownerHostId`, so a same-host process that died holding the lock is never
+ * reclaimed and broker startup burns its whole retry budget. Identity is applied
+ * under the options a caller passed for budget and cancellation, so startup,
+ * heartbeat, and ordinary index transactions share one reclaim rule: a proven-dead
+ * same-host owner can be recovered, and a live or cross-host owner cannot.
  */
-function withSessionIndexLock<T>(
+async function sessionIndexLockOptions(agentDir: string, options: FileLockOptions): Promise<FileLockOptions> {
+	return { ...options, ...(await brokerLockHostIdentity(agentDir)) };
+}
+
+async function withSessionIndexLock<T>(
 	operation: string,
 	agentDir: string,
 	callback: () => Promise<T>,
 	options: FileLockOptions = SESSION_INDEX_LOCK_OPTIONS,
 ): Promise<T> {
+	const lockOptions = await sessionIndexLockOptions(agentDir, options);
 	return withFileLock(
 		logFor(agentDir),
 		async () => {
@@ -885,7 +900,7 @@ function withSessionIndexLock<T>(
 				clearTimeout(note);
 			}
 		},
-		options,
+		lockOptions,
 	);
 }
 function isValidSnapshot(snapshot: unknown): snapshot is { indexSeq: number; events: SessionIndexEvent[] } {
@@ -2445,7 +2460,17 @@ export class SessionIndex {
 			// so startup readiness never overtakes lock release.
 			return await checkpoint();
 		} catch (error) {
-			if (signal.aborted && error === signal.reason) return 0;
+			if (signal.aborted && error === signal.reason) {
+				// The budget abort can beat the lock layer's own exhaustion. Asking
+				// directly keeps a retained removal transition from being reported as
+				// an idle cycle: the blocked-renewal caller needs the reason and path.
+				const blocked = await describeRetainedLockTransition(
+					path.resolve(logFor(this.#agentDir)),
+					await sessionIndexLockOptions(this.#agentDir, SESSION_INDEX_LOCK_OPTIONS),
+				);
+				if (blocked) throw blocked;
+				return 0;
+			}
 			if (
 				performance.now() >= deadline &&
 				isFileLockAcquireTimeout(error) &&

package/src/sdk/broker/transport.ts

@@ -66,6 +66,7 @@ export class BrokerTransport {
 	readonly #requestedPort: number;
 	#server: Bun.Server<undefined> | null = null;
 	#port = 0;
+	readonly #clients = new Set<ServerWebSocket<unknown>>();
 	readonly #shutdownBackpressured = new WeakSet<ServerWebSocket<unknown>>();
 	#shutdownRequested = false;
 	constructor(broker: Broker, token: string, port = 0) {
@@ -77,6 +78,10 @@ export class BrokerTransport {
 		if (!this.#server) throw new Error("Broker transport is not running");
 		return this.#port;
 	}
+	/** Sockets a client holds open right now; the broker outlives every one of them. */
+	get clientCount(): number {
+		return this.#clients.size;
+	}
 	async start(): Promise<number> {
 		if (this.#server) return this.#port;
 		this.#server = Bun.serve({
@@ -95,8 +100,13 @@ export class BrokerTransport {
 			websocket: {
 				maxPayloadLength: MAX_BROKER_JSON_FRAME_BYTES * 2,
 				open: socket => {
+					this.#clients.add(socket);
 					send(socket, { type: "broker_hello", protocolVersion: PROTOCOL_VERSION });
 				},
+				close: socket => {
+					this.#clients.delete(socket);
+					this.#shutdownBackpressured.delete(socket);
+				},
 				message: (socket, message) => void this.#handleMessage(socket, message),
 				drain: socket => {
 					if (this.#shutdownBackpressured.delete(socket)) this.#scheduleStopAfterShutdownResponse();
@@ -147,7 +157,13 @@ export class BrokerTransport {
 			sendError(socket, frame.id, "invalid_input", "idempotencyKey must be a string");
 			return;
 		}
-		if (!this.#broker.ownsDiscovery) {
+		// After initial publication, the owner retains the authenticated endpoint
+		// so it can stop this exact broker even after discovery is lost. Before
+		// publication readiness, shutdown remains unavailable.
+		if (
+			!this.#broker.ownsDiscovery &&
+			!(frame.operation === "broker.shutdown" && this.#broker.hasPublishedDiscovery)
+		) {
 			sendError(socket, frame.id, "unavailable", "broker publication is unavailable");
 			return;
 		}

package/src/sdk/bus/chat-daemon-control.ts

@@ -141,8 +141,8 @@ export type ChatDaemonAction = "stop" | "reload";
  * crashes, recovery, and attachment retirement cannot replay ambiguous work.
  */
 export const CHAT_DAEMON_GENERATIONS: Readonly<Record<ChatDaemonKind, number>> = {
-	discord: 84,
-	slack: 91,
+	discord: 85,
+	slack: 93,
 };
 
 export function chatDaemonGeneration(kind: ChatDaemonKind): number {