@ouro.bot/cli 0.1.0-alpha.601 → 0.1.0-alpha.602

package/changelog.json

@@ -1,6 +1,12 @@
 {
   "_note": "This changelog is maintained as part of the PR/version-bump workflow. Agent-curated, not auto-generated. Agents read this file directly via read_file to understand what changed between versions.",
   "versions": [
+    {
+      "version": "0.1.0-alpha.602",
+      "changes": [
+        "Two-part fix for the 2026-05-11 BlueBubbles wedge that drove the user (Ari) up the wall: Slugger's BB session showed the same user message replayed 76 times, because each death-spiral cycle re-injected the inbound. Root cause was the daemon's HTTP health probe (`createHttpHealthProbe(\"bluebubbles:<agent>\", port)`) GETting the sense's /health endpoint every ~60 s with a 5 s timeout — busy BB sense (e.g. VLM image-describe at 20+ s) timed out, daemon declared 'critical', SIGTERM'd the sense mid-work, respawned, hit the same image, killed again, forever. Part 1: removed the HTTP probe entirely from `listHealthProbes()`. Process supervision (`processManager` child-process exit handler) already catches dead processes; for 'alive but hung' we now rely on the agent's own awareness via `pendingRecoveryCount` / `lastRecoveredAt` in the BB runtime state surfaced into the prompt, plus the agent's new `restart_runtime` tool (from alpha.598 / #723). Part 2: defense-in-depth respawn-loop guard in `processManager.restartAgent` — if anything triggers more than `RESPAWN_GUARD_MAX_RESTARTS = 5` orchestrated restarts in `RESPAWN_GUARD_WINDOW_MS = 10 min`, refuse further restarts (`daemon.agent_respawn_loop_tripped` nerves event, errorReason + fixHint set on the snapshot). Trip self-clears once timestamps age out of the window, and `startAgent` (= `ouro up`) bypasses the guard so the operator can always recover. Even if some other future cause re-introduces a tight respawn loop, the guard bounds it. The 2026-05-11 spiral was ~60 restarts/hr — well above 5/10min, so this would have caught it."
+      ]
+    },
     {
       "version": "0.1.0-alpha.601",
       "changes": [

package/dist/heart/daemon/process-manager.js

@@ -33,7 +33,7 @@ var __importStar = (this && this.__importStar) || (function () {
     };
 })();
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.DaemonProcessManager = void 0;
+exports.DaemonProcessManager = exports.RESPAWN_GUARD_WINDOW_MS = exports.RESPAWN_GUARD_MAX_RESTARTS = void 0;
 const child_process_1 = require("child_process");
 const path = __importStar(require("path"));
 const identity_1 = require("../identity");
@@ -41,6 +41,18 @@ const runtime_1 = require("../../nerves/runtime");
 function startOfHour(ms) {
     return ms - 60 * 60 * 1000;
 }
+/**
+ * Respawn-loop guard: refuse `restartAgent` if we've already orchestrated
+ * RESPAWN_GUARD_MAX_RESTARTS in the past RESPAWN_GUARD_WINDOW_MS.
+ *
+ * Calibrated for the 2026-05-11 BB sense incident: a misconfigured probe
+ * was triggering `restartAgent` every ~60s for hours. Five restarts in
+ * 10 minutes is well above the rate of legitimate operational restarts
+ * (a single human-initiated `ouro down && ouro up` produces one) and well
+ * below the rate of a death spiral (60/hr ⇒ 10/10min).
+ */
+exports.RESPAWN_GUARD_MAX_RESTARTS = 5;
+exports.RESPAWN_GUARD_WINDOW_MS = 10 * 60_000;
 class DaemonProcessManager {
     agents = new Map();
     maxRestartsPerHour;
@@ -149,6 +161,8 @@ class DaemonProcessManager {
                 startAttemptId: 0,
                 restartTimer: null,
                 crashTimestamps: [],
+                orchestratedRestartTimestamps: [],
+                respawnLoopTripped: false,
                 stopRequested: false,
                 cooldownTimer: null,
                 cooldownRetryCount: 0,
@@ -370,6 +384,11 @@ class DaemonProcessManager {
         this.clearRestartTimer(state);
         this.clearCooldownTimer(state);
         state.stopRequested = true;
+        // NOTE: do not touch state.respawnLoopTripped / orchestratedRestartTimestamps
+        // here. restartAgent calls stopAgent internally; clearing the guard here
+        // would reset the window every cycle and defeat the loop-detection. The
+        // guard self-clears when timestamps age out of the window (handled inside
+        // restartAgent at the prune step).
         if (!state.process) {
             state.snapshot.status = "stopped";
             state.snapshot.pid = null;
@@ -396,6 +415,61 @@ class DaemonProcessManager {
     }
     async restartAgent(agent) {
         const state = this.requireAgent(agent);
+        // Respawn-loop guard: prune timestamps outside the window, then check
+        // whether we've already restarted this agent too many times in it.
+        const now = this.now();
+        const windowStart = now - exports.RESPAWN_GUARD_WINDOW_MS;
+        state.orchestratedRestartTimestamps = state.orchestratedRestartTimestamps.filter((ts) => ts >= windowStart);
+        // If the window is now empty, the trip naturally self-clears. That means
+        // after RESPAWN_GUARD_WINDOW_MS of no restart attempts, the daemon is
+        // willing to try again (e.g. for a fresh health probe failure that has
+        // nothing to do with the original loop).
+        if (state.respawnLoopTripped && state.orchestratedRestartTimestamps.length === 0) {
+            state.respawnLoopTripped = false;
+            state.snapshot.errorReason = null;
+            state.snapshot.fixHint = null;
+            (0, runtime_1.emitNervesEvent)({
+                component: "daemon",
+                event: "daemon.agent_respawn_loop_cleared",
+                message: "respawn-loop guard cleared by window-aging",
+                meta: { agent, windowMs: exports.RESPAWN_GUARD_WINDOW_MS },
+            });
+            this.notifySnapshotChange(state.snapshot);
+        }
+        if (state.respawnLoopTripped) {
+            (0, runtime_1.emitNervesEvent)({
+                level: "error",
+                component: "daemon",
+                event: "daemon.agent_respawn_loop_blocked",
+                message: "refused agent restart — respawn-loop guard tripped; manual intervention required",
+                meta: {
+                    agent,
+                    recentRestartCount: state.orchestratedRestartTimestamps.length,
+                    windowMs: exports.RESPAWN_GUARD_WINDOW_MS,
+                },
+            });
+            return;
+        }
+        if (state.orchestratedRestartTimestamps.length >= exports.RESPAWN_GUARD_MAX_RESTARTS) {
+            state.respawnLoopTripped = true;
+            state.snapshot.errorReason = `respawn loop detected: ${exports.RESPAWN_GUARD_MAX_RESTARTS}+ restarts in ${Math.round(exports.RESPAWN_GUARD_WINDOW_MS / 60_000)}min — refusing further restarts`;
+            state.snapshot.fixHint = "investigate the root cause then run `ouro up` to resume";
+            (0, runtime_1.emitNervesEvent)({
+                level: "error",
+                component: "daemon",
+                event: "daemon.agent_respawn_loop_tripped",
+                message: "respawn-loop guard tripped; further restarts blocked",
+                meta: {
+                    agent,
+                    restartCount: state.orchestratedRestartTimestamps.length,
+                    windowMs: exports.RESPAWN_GUARD_WINDOW_MS,
+                    maxRestarts: exports.RESPAWN_GUARD_MAX_RESTARTS,
+                },
+            });
+            this.notifySnapshotChange(state.snapshot);
+            return;
+        }
+        state.orchestratedRestartTimestamps.push(now);
         if (state.startInFlight && !state.process) {
             const startedAt = state.startAttemptedAtMs;
             /* v8 ignore next -- defensive: startInFlight always records a start timestamp @preserve */

package/dist/heart/daemon/sense-manager.js

@@ -44,7 +44,6 @@ const provider_credentials_1 = require("../provider-credentials");
 const sense_truth_1 = require("../sense-truth");
 const machine_identity_1 = require("../machine-identity");
 const process_manager_1 = require("./process-manager");
-const http_health_probe_1 = require("./http-health-probe");
 const DEFAULT_TEAMS_PORT = 3978;
 const DEFAULT_BLUEBUBBLES_PORT = 18790;
 const DEFAULT_BLUEBUBBLES_WEBHOOK_PATH = "/bluebubbles-webhook";
@@ -631,13 +630,33 @@ class DaemonSenseManager {
             if (!context.senses.bluebubbles.enabled || !context.facts.bluebubbles.configured || !machineRuntimeConfig.ok) {
                 continue;
             }
-            const machinePayload = machineRuntimeConfig.config;
-            const bluebubblesChannel = machinePayload.bluebubblesChannel;
-            const port = numberField(bluebubblesChannel, "port", DEFAULT_BLUEBUBBLES_PORT);
-            probes.push({
-                ...(0, http_health_probe_1.createHttpHealthProbe)(`bluebubbles:${agent}`, port),
-                managedName: `${agent}:bluebubbles`,
-            });
+            // DELIBERATELY no HTTP health probe for BlueBubbles.
+            //
+            // We used to register `createHttpHealthProbe(...)` here, which GETs the
+            // sense's /health endpoint every ~60s with a 5s timeout. On 2026-05-11
+            // that caused a death spiral:
+            //   1. Sense gets busy with real work (e.g. VLM image describe → 20+s)
+            //   2. /health probe times out at 5s
+            //   3. Daemon declares the sense "critical" → SIGTERMs it mid-work
+            //   4. Sense respawns, recovery loop replays the same inbound message
+            //      into the agent's BB session (visible side-effect — slugger saw
+            //      the same user text injected 76 times)
+            //   5. New sense hits the same VLM call, gets killed at 5s, repeat
+            //
+            // The probe was redundant supervision: dead processes are already
+            // recaptured by `processManager`'s child-process exit handler. The
+            // probe specifically caught "alive but hung" cases — but the cost
+            // (killing genuinely-busy processes and replaying messages) far
+            // outweighed the benefit. For "alive but hung" detection we now
+            // rely on the agent's own awareness: BB sense's runtime.json carries
+            // pendingRecoveryCount + lastRecoveredAt, surfaced in the agent
+            // prompt. If recovery has been wedged for too long, the agent can
+            // call `restart_runtime` itself (see alpha.598 / PR #723).
+            //
+            // The respawn-loop guard in processManager is the backstop: even if
+            // something else triggers a tight respawn cycle for any reason, the
+            // guard fires and refuses further restarts after N attempts in M
+            // minutes, so we can never re-enter the 2026-05-11 spiral.
         }
         return probes;
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ouro.bot/cli",
-  "version": "0.1.0-alpha.601",
+  "version": "0.1.0-alpha.602",
   "main": "dist/heart/daemon/ouro-entry.js",
   "bin": {
     "cli": "dist/heart/daemon/ouro-bot-entry.js",