@songsid/agend 2.1.1-beta.15 → 2.1.1-beta.17

package/dist/daemon.js

@@ -10,6 +10,7 @@ import { TmuxManager, resolveTmuxLogicalSize } from "./tmux-manager.js";
 import { TranscriptMonitor } from "./transcript-monitor.js";
 import { ContextGuardian } from "./context-guardian.js";
 import { IpcServer } from "./channel/ipc-bridge.js";
+import { daemonBudgetMs } from "./channel/ipc-timeouts.js";
 import { MessageBus } from "./channel/message-bus.js";
 import { shellQuote } from "./backend/types.js";
 import { getTmuxSession } from "./config.js";
@@ -848,6 +849,7 @@ export class Daemon extends EventEmitter {
                         this.logger.warn({ instanceDir: this.instanceDir }, "Instance directory missing — stopping health check");
                         this.healthCheckPaused = true;
                         this.healthCheckTimer = null;
+                        this.emitSupervisionEnded("instance directory was removed from disk", "Recreate the instance, or delete it from fleet.yaml.");
                         return;
                     }
                     if (!this.tmux || this.spawning || this.healthCheckPaused || Daemon.tmuxServerPaused) {
@@ -898,6 +900,7 @@ export class Daemon extends EventEmitter {
                         this.logger.info("CLI exited normally (code 0) — pausing health check");
                         await this.tmux.killWindow();
                         this.healthCheckPaused = true;
+                        this.emitSupervisionEnded("the CLI exited normally (code 0)", "Nothing crashed — start it again when you need it.");
                         return;
                     }
                     this.setProcessStatus("crashed");
@@ -1007,6 +1010,7 @@ export class Daemon extends EventEmitter {
                     if (max_retries <= 0) {
                         this.healthCheckPaused = true;
                         this.logger.warn(`${cliLabel} window died — automatic restart is disabled`);
+                        this.emitSupervisionEnded("the CLI died and automatic restart is disabled (restart_policy.max_retries is 0)", "Start it manually, or raise max_retries in fleet.yaml.");
                         return;
                     }
                     // Detect rapid crash: sliding window — 3+ crashes in 5 minutes
@@ -1036,6 +1040,8 @@ export class Daemon extends EventEmitter {
                     this.lastCrashAt = Date.now();
                     if (this.crashCount > max_retries) {
                         this.logger.error({ crashCount: this.crashCount, maxRetries: max_retries }, "Max crash retries exceeded — not respawning");
+                        this.healthCheckPaused = true;
+                        this.emitSupervisionEnded(`it crashed ${this.crashCount} times, exceeding restart_policy.max_retries (${max_retries})`, "Check the logs for the cause, then restart it.");
                         return; // don't schedule next — given up
                     }
                     // Calculate backoff delay
@@ -1119,6 +1125,21 @@ export class Daemon extends EventEmitter {
      * A dead remain-on-exit pane still contains the old ready prompt, so allowing
      * the pane monitor to capture it would re-create a false idle state.
      */
+    /**
+     * Announce that this instance is no longer being supervised.
+     *
+     * Four health-check exits set `healthCheckPaused = true` and returned without
+     * telling anyone: a clean CLI exit, `max_retries <= 0`, crash retries exhausted,
+     * and the instance directory being deleted. Only the crash-LOOP case emitted an
+     * event, so a fleet could quietly contain a dead instance that still looked fine
+     * on the dashboard — while messages routed to it queued or failed with a bare ❌.
+     *
+     * `crash_loop` already had this treatment; this is the same bridge for the other
+     * four, carrying a human-readable cause and the operator's next step.
+     */
+    emitSupervisionEnded(reason, remedy) {
+        this.emit("supervision_ended", { name: this.name, reason, remedy });
+    }
     setProcessStatus(status) {
         if (this.processStatus === status)
             return;
@@ -2236,8 +2257,8 @@ export class Daemon extends EventEmitter {
             });
             const timeout = setTimeout(() => {
                 this.pendingIpcRequests.delete(fleetReqId);
-                respond(null, "Task operation timed out after 30s");
-            }, 30_000);
+                respond(null, `Task operation timed out after ${daemonBudgetMs(tool) / 1000}s`);
+            }, daemonBudgetMs(tool));
             this.pendingIpcRequests.set(fleetReqId, (respMsg) => {
                 clearTimeout(timeout);
                 respond(respMsg.result, respMsg.error);
@@ -2259,8 +2280,8 @@ export class Daemon extends EventEmitter {
             });
             const timeout = setTimeout(() => {
                 this.pendingIpcRequests.delete(fleetReqId);
-                respond(null, "Decision operation timed out after 30s");
-            }, 30_000);
+                respond(null, `Decision operation timed out after ${daemonBudgetMs(tool) / 1000}s`);
+            }, daemonBudgetMs(tool));
             this.pendingIpcRequests.set(fleetReqId, (respMsg) => {
                 clearTimeout(timeout);
                 respond(respMsg.result, respMsg.error);
@@ -2286,8 +2307,8 @@ export class Daemon extends EventEmitter {
             // Wait for fleet_schedule_response via pending request map
             const timeout = setTimeout(() => {
                 this.pendingIpcRequests.delete(fleetReqId);
-                respond(null, "Schedule operation timed out after 30s");
-            }, 30_000);
+                respond(null, `Schedule operation timed out after ${daemonBudgetMs(tool) / 1000}s`);
+            }, daemonBudgetMs(tool));
             this.pendingIpcRequests.set(fleetReqId, (respMsg) => {
                 clearTimeout(timeout);
                 respond(respMsg.result, respMsg.error);
@@ -2308,7 +2329,9 @@ export class Daemon extends EventEmitter {
                     fleetRequestId: fleetReqId,
                     senderSessionName,
                 });
-                const crossTimeoutMs = (tool === "start_instance" || tool === "create_instance" || tool === "replace_instance") ? 60_000 : 30_000;
+                // Shared table, so delete_instance and the deployment tools get the same
+                // budget here as the MCP client expects (the ad-hoc list omitted them).
+                const crossTimeoutMs = daemonBudgetMs(tool);
                 const timeout = setTimeout(() => {
                     this.pendingIpcRequests.delete(fleetReqId);
                     respond(null, `Cross-instance operation timed out after ${crossTimeoutMs / 1000}s`);
@@ -2351,8 +2374,8 @@ export class Daemon extends EventEmitter {
             this.ipcServer?.broadcast({ type: "fleet_outbound", tool, args, fleetRequestId: fleetReqId });
             const timeout = setTimeout(() => {
                 this.pendingIpcRequests.delete(outboundKey);
-                respond(null, "Fleet outbound timed out after 30s");
-            }, 30_000);
+                respond(null, `Fleet outbound timed out after ${daemonBudgetMs(tool) / 1000}s`);
+            }, daemonBudgetMs(tool));
             this.pendingIpcRequests.set(outboundKey, (respMsg) => {
                 clearTimeout(timeout);
                 respond(respMsg.result, respMsg.error);