instar 1.3.107 → 1.3.108

package/dist/core/claudeActivityIndicators.d.ts

@@ -0,0 +1,37 @@
+/**
+ * claudeActivityIndicators — the canonical "Claude Code is mid-turn right now"
+ * footer hints.
+ *
+ * Claude Code renders these strings in the status footer ONLY while a turn is
+ * in flight (model generating, tool running, extended-think computing). They
+ * disappear the instant the turn lands and the session returns to its idle
+ * prompt. That structural property is exactly what makes them a reliable
+ * "actively working" tell: a session sitting idle at the prompt — or a session
+ * that has WEDGED and fast-fails every turn — never shows them.
+ *
+ * Why this is its own module: more than one recovery surface needs to answer
+ * "is this pane actively working?" before taking a disruptive action —
+ *   • StuckInputSentinel / SessionManager.verifyInjection refuse to fire a
+ *     recovery Enter into a working pane (the input is correctly queued and
+ *     will submit when the turn ends).
+ *   • CompactionSentinel refuses to RE-INJECT a recovery prompt into a working
+ *     pane — re-injecting buries the user's real message under stacked
+ *     recovery bootstraps (the false "session is restarting" loop).
+ * Centralizing the list keeps those surfaces from drifting apart.
+ *
+ * IMPORTANT — footer hints ONLY. We deliberately do NOT include the Braille
+ * spinner glyphs (⠋⠙⠹…) here: a frozen last frame of a dead pane can still
+ * contain a spinner glyph, which would make a genuinely-stuck session read as
+ * "working" and starve it of recovery. The footer hints are present only while
+ * a turn is actually in flight, so they cannot double-fire against a dead pane.
+ * (See frameworkActivitySignals.ts for the spinner-inclusive, framework-portable
+ * signal used by the triage nurse, where the trade-off is different.)
+ */
+export declare const CLAUDE_WORKING_INDICATORS: readonly string[];
+/**
+ * Pure check: does this captured tmux pane show Claude Code actively working?
+ * Footer-hint based — see CLAUDE_WORKING_INDICATORS for why that's the precise
+ * signal. Returns false for empty/missing panes.
+ */
+export declare function paneShowsClaudeWorking(pane: string | null | undefined): boolean;
+//# sourceMappingURL=claudeActivityIndicators.d.ts.map

package/dist/core/claudeActivityIndicators.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"claudeActivityIndicators.d.ts","sourceRoot":"","sources":["../../src/core/claudeActivityIndicators.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AACH,eAAO,MAAM,yBAAyB,EAAE,SAAS,MAAM,EAItD,CAAC;AAEF;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,GAAG,OAAO,CAG/E"}

package/dist/core/claudeActivityIndicators.js

@@ -0,0 +1,45 @@
+/**
+ * claudeActivityIndicators — the canonical "Claude Code is mid-turn right now"
+ * footer hints.
+ *
+ * Claude Code renders these strings in the status footer ONLY while a turn is
+ * in flight (model generating, tool running, extended-think computing). They
+ * disappear the instant the turn lands and the session returns to its idle
+ * prompt. That structural property is exactly what makes them a reliable
+ * "actively working" tell: a session sitting idle at the prompt — or a session
+ * that has WEDGED and fast-fails every turn — never shows them.
+ *
+ * Why this is its own module: more than one recovery surface needs to answer
+ * "is this pane actively working?" before taking a disruptive action —
+ *   • StuckInputSentinel / SessionManager.verifyInjection refuse to fire a
+ *     recovery Enter into a working pane (the input is correctly queued and
+ *     will submit when the turn ends).
+ *   • CompactionSentinel refuses to RE-INJECT a recovery prompt into a working
+ *     pane — re-injecting buries the user's real message under stacked
+ *     recovery bootstraps (the false "session is restarting" loop).
+ * Centralizing the list keeps those surfaces from drifting apart.
+ *
+ * IMPORTANT — footer hints ONLY. We deliberately do NOT include the Braille
+ * spinner glyphs (⠋⠙⠹…) here: a frozen last frame of a dead pane can still
+ * contain a spinner glyph, which would make a genuinely-stuck session read as
+ * "working" and starve it of recovery. The footer hints are present only while
+ * a turn is actually in flight, so they cannot double-fire against a dead pane.
+ * (See frameworkActivitySignals.ts for the spinner-inclusive, framework-portable
+ * signal used by the triage nurse, where the trade-off is different.)
+ */
+export const CLAUDE_WORKING_INDICATORS = [
+    'esc to interrupt', // Claude Code's "task in progress" hint
+    'ctrl+t to hide tasks', // Multi-task display
+    'tokens · esc', // Token-counting + interrupt hint
+];
+/**
+ * Pure check: does this captured tmux pane show Claude Code actively working?
+ * Footer-hint based — see CLAUDE_WORKING_INDICATORS for why that's the precise
+ * signal. Returns false for empty/missing panes.
+ */
+export function paneShowsClaudeWorking(pane) {
+    if (!pane)
+        return false;
+    return CLAUDE_WORKING_INDICATORS.some(ind => pane.includes(ind));
+}
+//# sourceMappingURL=claudeActivityIndicators.js.map

package/dist/core/claudeActivityIndicators.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"claudeActivityIndicators.js","sourceRoot":"","sources":["../../src/core/claudeActivityIndicators.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AACH,MAAM,CAAC,MAAM,yBAAyB,GAAsB;IAC1D,kBAAkB,EAAM,wCAAwC;IAChE,sBAAsB,EAAE,qBAAqB;IAC7C,cAAc,EAAU,kCAAkC;CAC3D,CAAC;AAEF;;;;GAIG;AACH,MAAM,UAAU,sBAAsB,CAAC,IAA+B;IACpE,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,yBAAyB,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC;AACnE,CAAC"}

package/dist/core/cpuStarvation.d.ts

@@ -0,0 +1,30 @@
+/**
+ * cpuStarvation — a tiny shared helper for "is this machine so oversubscribed
+ * that a live process can't get scheduled?"
+ *
+ * The signal is `loadavg[0] / cpuCount`. Above ~1.5 the run-queue is longer
+ * than the core count, so a perfectly-alive event loop can stall for many
+ * seconds simply because it isn't getting CPU time. Two watchers need to tell
+ * this apart from a genuine fault, or they take a destructive action that makes
+ * things worse:
+ *   - SleepWakeDetector: a short heartbeat drift under CPU starvation is NOT a
+ *     real sleep/wake (it suppresses the wake — see SleepWakeDetector).
+ *   - ServerSupervisor: a server that can't answer /health under CPU starvation
+ *     is NOT dead — restarting it doesn't cure the starvation (the fresh server
+ *     is starved too), it just drops the in-flight message and loops.
+ *
+ * SleepWakeDetector keeps its own injectable providers for unit testing; this
+ * module is the canonical default ratio + computation that other callers
+ * (ServerSupervisor) share so the fleet uses one definition of "CPU-starved."
+ */
+/** Above this loadavg[0]/cpuCount ratio the machine is treated as CPU-starved. */
+export declare const DEFAULT_MAX_LOAD_RATIO = 1.5;
+/**
+ * 1-minute load average divided by CPU count. Returns 0 on any error (so a
+ * failure to read system load never trips a starvation branch). Providers are
+ * injectable for tests.
+ */
+export declare function cpuLoadRatio(loadAvg1?: number, cpuCount?: number): number;
+/** True when the machine is CPU-starved (load ratio strictly above maxRatio). */
+export declare function isCpuStarved(maxRatio?: number, loadAvg1?: number, cpuCount?: number): boolean;
+//# sourceMappingURL=cpuStarvation.d.ts.map

package/dist/core/cpuStarvation.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cpuStarvation.d.ts","sourceRoot":"","sources":["../../src/core/cpuStarvation.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAIH,kFAAkF;AAClF,eAAO,MAAM,sBAAsB,MAAM,CAAC;AAE1C;;;;GAIG;AACH,wBAAgB,YAAY,CAC1B,QAAQ,CAAC,EAAE,MAAM,EACjB,QAAQ,CAAC,EAAE,MAAM,GAChB,MAAM,CASR;AAED,iFAAiF;AACjF,wBAAgB,YAAY,CAC1B,QAAQ,GAAE,MAA+B,EACzC,QAAQ,CAAC,EAAE,MAAM,EACjB,QAAQ,CAAC,EAAE,MAAM,GAChB,OAAO,CAET"}

package/dist/core/cpuStarvation.js

@@ -0,0 +1,44 @@
+/**
+ * cpuStarvation — a tiny shared helper for "is this machine so oversubscribed
+ * that a live process can't get scheduled?"
+ *
+ * The signal is `loadavg[0] / cpuCount`. Above ~1.5 the run-queue is longer
+ * than the core count, so a perfectly-alive event loop can stall for many
+ * seconds simply because it isn't getting CPU time. Two watchers need to tell
+ * this apart from a genuine fault, or they take a destructive action that makes
+ * things worse:
+ *   - SleepWakeDetector: a short heartbeat drift under CPU starvation is NOT a
+ *     real sleep/wake (it suppresses the wake — see SleepWakeDetector).
+ *   - ServerSupervisor: a server that can't answer /health under CPU starvation
+ *     is NOT dead — restarting it doesn't cure the starvation (the fresh server
+ *     is starved too), it just drops the in-flight message and loops.
+ *
+ * SleepWakeDetector keeps its own injectable providers for unit testing; this
+ * module is the canonical default ratio + computation that other callers
+ * (ServerSupervisor) share so the fleet uses one definition of "CPU-starved."
+ */
+import os from 'node:os';
+/** Above this loadavg[0]/cpuCount ratio the machine is treated as CPU-starved. */
+export const DEFAULT_MAX_LOAD_RATIO = 1.5;
+/**
+ * 1-minute load average divided by CPU count. Returns 0 on any error (so a
+ * failure to read system load never trips a starvation branch). Providers are
+ * injectable for tests.
+ */
+export function cpuLoadRatio(loadAvg1, cpuCount) {
+    try {
+        const load = loadAvg1 ?? os.loadavg()[0];
+        const cpus = cpuCount ?? os.cpus().length;
+        if (!Number.isFinite(load) || !Number.isFinite(cpus) || cpus < 1)
+            return 0;
+        return load / cpus;
+    }
+    catch {
+        return 0;
+    }
+}
+/** True when the machine is CPU-starved (load ratio strictly above maxRatio). */
+export function isCpuStarved(maxRatio = DEFAULT_MAX_LOAD_RATIO, loadAvg1, cpuCount) {
+    return cpuLoadRatio(loadAvg1, cpuCount) > maxRatio;
+}
+//# sourceMappingURL=cpuStarvation.js.map

package/dist/core/cpuStarvation.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cpuStarvation.js","sourceRoot":"","sources":["../../src/core/cpuStarvation.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,EAAE,MAAM,SAAS,CAAC;AAEzB,kFAAkF;AAClF,MAAM,CAAC,MAAM,sBAAsB,GAAG,GAAG,CAAC;AAE1C;;;;GAIG;AACH,MAAM,UAAU,YAAY,CAC1B,QAAiB,EACjB,QAAiB;IAEjB,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,QAAQ,IAAI,EAAE,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,QAAQ,IAAI,EAAE,CAAC,IAAI,EAAE,CAAC,MAAM,CAAC;QAC1C,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,IAAI,GAAG,CAAC;YAAE,OAAO,CAAC,CAAC;QAC3E,OAAO,IAAI,GAAG,IAAI,CAAC;IACrB,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,CAAC,CAAC;IACX,CAAC;AACH,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,YAAY,CAC1B,WAAmB,sBAAsB,EACzC,QAAiB,EACjB,QAAiB;IAEjB,OAAO,YAAY,CAAC,QAAQ,EAAE,QAAQ,CAAC,GAAG,QAAQ,CAAC;AACrD,CAAC"}

package/dist/lifeline/ServerSupervisor.d.ts

@@ -127,6 +127,9 @@ export declare class ServerSupervisor extends EventEmitter {
     private consecutiveBindFailures;
     private readonly bindFailureEscalationThreshold;
     private readonly processAliveThreshold;
+    private readonly maxLoadRatio;
+    private readonly starvationRestartThreshold;
+    private readonly loadRatioProvider;
     private stateDir;
     private maintenanceWaitStartedAt;
     private maintenanceWaitMs;
@@ -165,6 +168,8 @@ export declare class ServerSupervisor extends EventEmitter {
         maintenanceWaitMinutes?: number;
         /** How long to wait after spawning before starting health checks. Default: 180 seconds (3 minutes). */
         startupGraceSeconds?: number;
+        /** Injectable CPU-load-ratio source (loadavg[0]/cpuCount) for tests. Default: cpuLoadRatio. */
+        loadRatioProvider?: () => number;
     });
     /**
      * Start the server and begin monitoring.
@@ -343,6 +348,26 @@ export declare class ServerSupervisor extends EventEmitter {
     getPendingRemediatorRequestCount(): number;
     /** Test helper — simulate a healthy tick driving completion notifications. */
     triggerHealthyTickForTests(): void;
+    /**
+     * Decide what to do about a server that has failed `>= unhealthyThreshold`
+     * consecutive health checks. Shared by the health-check loop's two failure
+     * paths (unhealthy /health response, and a thrown check).
+     *
+     *   - Process dead → restart immediately.
+     *   - Process alive but unresponsive, below the (load-lenient) threshold →
+     *     keep waiting.
+     *   - Process alive, threshold reached, but the box is CPU-starved → DEFER the
+     *     restart (restarting a starved server only drops the in-flight message
+     *     and loops) until starvation eases or the hard cap forces it.
+     *   - Process alive, threshold reached, not starved → restart.
+     */
+    private evaluateUnhealthyServer;
+    /**
+     * True when we should hold off restarting an alive-but-unresponsive server
+     * because the machine is CPU-starved — but only up to the hard cap. Past the
+     * cap we restart regardless (the server may be genuinely hung, not starved).
+     */
+    private deferRestartForCpuStarvation;
     private handleUnhealthy;
     /**
      * Capture crash output from multiple sources:

package/dist/lifeline/ServerSupervisor.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"ServerSupervisor.d.ts","sourceRoot":"","sources":["../../src/lifeline/ServerSupervisor.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAIH,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAiB3C,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,EAAE,EAAE,CAAC;IACb,UAAU,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC7B,gBAAgB,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;IACpC,aAAa,EAAE,CAAC,aAAa,EAAE,MAAM,EAAE,eAAe,EAAE,MAAM,CAAC,CAAC;IAChE,qBAAqB,EAAE,CAAC,OAAO,EAAE;QAAE,cAAc,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAClF,mBAAmB,EAAE,CAAC,IAAI,EAAE;QAAE,cAAc,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACxE;wFACoF;IACpF,aAAa,EAAE,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC;CACxC;AAYD;;;;;;GAMG;AACH,MAAM,WAAW,uBAAuB;IACtC,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,SAAS,GAAG,SAAS,GAAG,OAAO,CAAC;IAC7C,gEAAgE;IAChE,WAAW,EAAE,MAAM,CAAC;IACpB,oEAAoE;IACpE,WAAW,EAAE,MAAM,CAAC;IACpB,6DAA6D;IAC7D,gBAAgB,EAAE,MAAM,CAAC;IACzB,iFAAiF;IACjF,IAAI,EAAE,MAAM,CAAC;CACd;AAED,kEAAkE;AAClE,MAAM,WAAW,qBAAqB;IACpC,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,OAAO,CAAC;IAClB;oFACgF;IAChF,MAAM,EAAE,MAAM,CAAC;IACf,+DAA+D;IAC/D,0BAA0B,EAAE,MAAM,CAAC;IACnC,4DAA4D;IAC5D,iBAAiB,EAAE,MAAM,CAAC;CAC3B;AAUD;;;;;;GAMG;AACH,MAAM,WAAW,qCAAqC;IACpD,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,WAAW,CAAC;IACzB,4EAA4E;IAC5E,iBAAiB,EAAE,MAAM,CAAC;IAC1B,qEAAqE;IACrE,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,iEAAiE;IACjE,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;;;GAIG;AACH,MAAM,WAAW,mCAAmC;IAClD,aAAa,CAAC,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,MAAM,GAAG,MAAM,CAAC;CAC/D;AAED,MAAM,WAAW,mCAAmC;IAClD,OAAO,EAAE,SAAS,GAAG,SAAS,CAAC;IAC/B,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAClC;AAyDD,wEAAwE;AACxE,MAAM,WAAW,oBAAoB;IACnC,8DAA8D;IAC9D,iBAAiB,EAAE,CAAC,GAAG,EAAE;QAAE,SAAS,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC;IACxD,0EAA0E;IAC1E,oBAAoB,EAAE,MAAM,MAAM,CAAC;CACpC;AAED,qBAAa,gBAAiB,SAAQ,YAAY;IAChD,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,WAAW,CAAS;IAC5B,OAAO,CAAC,IAAI,CAAS;IACrB,OAAO,CAAC,QAAQ,CAAgB;IAChC,OAAO,CAAC,iBAAiB,CAAS;IAClC,OAAO,CAAC,mBAAmB,CAA+C;IAC1E,OAAO,CAAC,iBAAiB,CAAK;IAC9B,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAc;IAC7C,OAAO,CAAC,eAAe,CAAK;IAC5B,OAAO,CAAC,kBAAkB,CAAK;IAC/B,OAAO,CAAC,gBAAgB,CAAQ;IAChC,OAAO,CAAC,SAAS,CAAS;IAC1B,OAAO,CAAC,WAAW,CAAK;IACxB,OAAO,CAAC,cAAc,CAAW;IACjC,OAAO,CAAC,SAAS,CAAK;IACtB,OAAO,CAAC,eAAe,CAAc;IACrC,OAAO,CAAC,qBAAqB,CAAK;IAClC,OAAO,CAAC,mBAAmB,CAAK;IAChC,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAK;IAMxC,OAAO,CAAC,uBAAuB,CAAK;IACpC,OAAO,CAAC,QAAQ,CAAC,8BAA8B,CAAK;IACpD,OAAO,CAAC,QAAQ,CAAC,qBAAqB,CAAK;IAC3C,OAAO,CAAC,QAAQ,CAAgB;IAGhC,OAAO,CAAC,wBAAwB,CAAK;IACrC,OAAO,CAAC,iBAAiB,CAAc;IACvC,OAAO,CAAC,oBAAoB,CAAuB;IAGnD,OAAO,CAAC,aAAa,CAAK;IAC1B,OAAO,CAAC,uBAAuB,CAAK;IACpC,OAAO,CAAC,QAAQ,CAAC,uBAAuB,CAAM;IAC9C,OAAO,CAAC,QAAQ,CAAC,sBAAsB,CAAe;IACtD,OAAO,CAAC,aAAa,CAAS;IAC9B,OAAO,CAAC,uBAAuB,CAAK;IACpC,OAAO,CAAC,wBAAwB,CAAK;IACrC,OAAO,CAAC,QAAQ,CAAC,6BAA6B,CAAe;IAC7D,OAAO,CAAC,QAAQ,CAAC,wBAAwB,CAAK;IAC9C,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAmB;IACvD,OAAO,CAAC,kBAAkB,CAAK;IAC/B,OAAO,CAAC,eAAe,CAAM;IAC7B,OAAO,CAAC,mBAAmB,CAAuB;IAClD,OAAO,CAAC,iBAAiB,CAAkC;IAC3D,OAAO,CAAC,mBAAmB,CAAK;IAChC,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAU;IAG3C,0EAA0E;IAC1E,MAAM,CAAC,QAAQ,CAAC,0BAA0B,KAAK;IAC/C,yEAAyE;IACzE,MAAM,CAAC,QAAQ,CAAC,0BAA0B,SAAc;IACxD,OAAO,CAAC,oBAAoB,CAAqC;IACjE,qFAAqF;IACrF,OAAO,CAAC,yBAAyB,CAAmF;IACpH,oFAAoF;IACpF,OAAO,CAAC,iBAAiB,CAA+D;gBAE5E,OAAO,EAAE;QACnB,UAAU,EAAE,MAAM,CAAC;QACnB,WAAW,EAAE,MAAM,CAAC;QACpB,IAAI,EAAE,MAAM,CAAC;QACb,QAAQ,CAAC,EAAE,MAAM,CAAC;QAClB,yGAAyG;QACzG,sBAAsB,CAAC,EAAE,MAAM,CAAC;QAChC,uGAAuG;QACvG,mBAAmB,CAAC,EAAE,MAAM,CAAC;KAC9B;IAiBD;;OAEG;IACG,KAAK,IAAI,OAAO,CAAC,OAAO,CAAC;IAyD/B;;OAEG;IACG,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC;IAyB3B;;OAEG;IACH,IAAI,OAAO,IAAI,OAAO,CAErB;IAED;;OAEG;IACH,SAAS,IAAI;QACX,OAAO,EAAE,OAAO,CAAC;QACjB,OAAO,EAAE,OAAO,CAAC;QACjB,eAAe,EAAE,MAAM,CAAC;QACxB,WAAW,EAAE,MAAM,CAAC;QACpB,aAAa,EAAE,MAAM,CAAC;QACtB,WAAW,EAAE,OAAO,CAAC;QACrB,mBAAmB,EAAE,MAAM,CAAC;QAC5B,aAAa,EAAE,OAAO,CAAC;QACvB,aAAa,EAAE,MAAM,CAAC;QACtB,eAAe,EAAE,MAAM,CAAC;QACxB,wBAAwB,EAAE,MAAM,CAAC;QACjC,wBAAwB,EAAE,MAAM,CAAC;QACjC,iBAAiB,EAAE,OAAO,CAAC;QAC3B,wBAAwB,EAAE,MAAM,CAAC;QACjC,gBAAgB,EAAE,OAAO,CAAC;QAC1B,yBAAyB,EAAE,MAAM,CAAC;KACnC;IA2BD;;;OAGG;IACH,mBAAmB,IAAI,IAAI;IAa3B;;;OAGG;IACH,sBAAsB,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI;IAI5C;;;;;OAKG;IACG,sBAAsB,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;IAsC9D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;OA8BG;IACG,oBAAoB,CACxB,GAAG,EAAE,qCAAqC,EAC1C,QAAQ,CAAC,EAAE,mCAAmC,GAC7C,OAAO,CAAC,mCAAmC,CAAC;IAmG/C;;;OAGG;IACH,OAAO,CAAC,iBAAiB;IA8QzB;;OAEG;IACH,OAAO,CAAC,WAAW;IAoBnB;;OAEG;IACH,OAAO,CAAC,YAAY;IAgBpB,OAAO,CAAC,WAAW;IAmFnB,OAAO,CAAC,oBAAoB;IAY5B,OAAO,CAAC,iBAAiB;IAkKzB,OAAO,CAAC,gBAAgB;YAWV,WAAW;IAoBzB;;;;;;;;;;OAUG;YACW,sBAAsB;IAYpC;;;;OAIG;IACH,OAAO,CAAC,mBAAmB;IA2E3B;;;OAGG;IACH,OAAO,CAAC,wBAAwB;IAiDhC;;OAEG;IACH,OAAO,CAAC,mBAAmB;IAoC3B;;;;;OAKG;IACH,kBAAkB,CAAC,UAAU,EAAE,oBAAoB,GAAG,IAAI;IAK1D,8DAA8D;IAC9D,2BAA2B,IAAI,MAAM;IAIrC,2DAA2D;IAC3D,oBAAoB,IAAI,MAAM;IAI9B,oEAAoE;IACpE,oBAAoB,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI;IAK3C;;;;;;;;;;;;;;OAcG;IACG,sBAAsB,CAAC,OAAO,EAAE,uBAAuB,GAAG,OAAO,CAAC,qBAAqB,CAAC;IAwG9F;;;;;;OAMG;IACH,OAAO,CAAC,wCAAwC;IAkBhD;;;;;;;OAOG;IACH,OAAO,CAAC,4BAA4B;IAiBpC,kEAAkE;IAClE,gCAAgC,IAAI,MAAM;IAI1C,8EAA8E;IAC9E,0BAA0B,IAAI,IAAI;IAMlC,OAAO,CAAC,eAAe;IAwKvB;;;;OAIG;IAOH,OAAO,CAAC,kBAAkB;IAgC1B;;;OAGG;IACH,OAAO,CAAC,qBAAqB;IAgC7B;;;OAGG;IACH,OAAO,CAAC,sBAAsB;IAgB9B,OAAO,CAAC,sBAAsB;IAa9B;;;;;;;;;;OAUG;IACH,OAAO,CAAC,uBAAuB;IAwC/B;;OAEG;IACH,OAAO,CAAC,sBAAsB;CAS/B;AAED;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,sEAAsE;IACtE,UAAU,EAAE,MAAM,CAAC;IACnB,kDAAkD;IAClD,UAAU,EAAE,MAAM,CAAC;IACnB,oEAAoE;IACpE,SAAS,EAAE,MAAM,CAAC;CACnB;AAED;;;;;;;;;;;;;GAaG;AACH,wBAAgB,uBAAuB,CAAC,eAAe,EAAE,MAAM,GAAG,iBAAiB,EAAE,CAqDpF;AAID;;;;;;;;;;;;;;;;;;GAkBG;AACH,wBAAgB,6BAA6B,CAAC,OAAO,EAAE,uBAAuB,GAAG,MAAM,CAuCtF"}
+{"version":3,"file":"ServerSupervisor.d.ts","sourceRoot":"","sources":["../../src/lifeline/ServerSupervisor.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAIH,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAkB3C,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,EAAE,EAAE,CAAC;IACb,UAAU,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC7B,gBAAgB,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;IACpC,aAAa,EAAE,CAAC,aAAa,EAAE,MAAM,EAAE,eAAe,EAAE,MAAM,CAAC,CAAC;IAChE,qBAAqB,EAAE,CAAC,OAAO,EAAE;QAAE,cAAc,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAClF,mBAAmB,EAAE,CAAC,IAAI,EAAE;QAAE,cAAc,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACxE;wFACoF;IACpF,aAAa,EAAE,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC;CACxC;AAYD;;;;;;GAMG;AACH,MAAM,WAAW,uBAAuB;IACtC,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,SAAS,GAAG,SAAS,GAAG,OAAO,CAAC;IAC7C,gEAAgE;IAChE,WAAW,EAAE,MAAM,CAAC;IACpB,oEAAoE;IACpE,WAAW,EAAE,MAAM,CAAC;IACpB,6DAA6D;IAC7D,gBAAgB,EAAE,MAAM,CAAC;IACzB,iFAAiF;IACjF,IAAI,EAAE,MAAM,CAAC;CACd;AAED,kEAAkE;AAClE,MAAM,WAAW,qBAAqB;IACpC,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,OAAO,CAAC;IAClB;oFACgF;IAChF,MAAM,EAAE,MAAM,CAAC;IACf,+DAA+D;IAC/D,0BAA0B,EAAE,MAAM,CAAC;IACnC,4DAA4D;IAC5D,iBAAiB,EAAE,MAAM,CAAC;CAC3B;AAUD;;;;;;GAMG;AACH,MAAM,WAAW,qCAAqC;IACpD,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,WAAW,CAAC;IACzB,4EAA4E;IAC5E,iBAAiB,EAAE,MAAM,CAAC;IAC1B,qEAAqE;IACrE,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,iEAAiE;IACjE,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;;;GAIG;AACH,MAAM,WAAW,mCAAmC;IAClD,aAAa,CAAC,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,MAAM,GAAG,MAAM,CAAC;CAC/D;AAED,MAAM,WAAW,mCAAmC;IAClD,OAAO,EAAE,SAAS,GAAG,SAAS,CAAC;IAC/B,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CAClC;AAyDD,wEAAwE;AACxE,MAAM,WAAW,oBAAoB;IACnC,8DAA8D;IAC9D,iBAAiB,EAAE,CAAC,GAAG,EAAE;QAAE,SAAS,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,CAAC;IACxD,0EAA0E;IAC1E,oBAAoB,EAAE,MAAM,MAAM,CAAC;CACpC;AAED,qBAAa,gBAAiB,SAAQ,YAAY;IAChD,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,WAAW,CAAS;IAC5B,OAAO,CAAC,IAAI,CAAS;IACrB,OAAO,CAAC,QAAQ,CAAgB;IAChC,OAAO,CAAC,iBAAiB,CAAS;IAClC,OAAO,CAAC,mBAAmB,CAA+C;IAC1E,OAAO,CAAC,iBAAiB,CAAK;IAC9B,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAc;IAC7C,OAAO,CAAC,eAAe,CAAK;IAC5B,OAAO,CAAC,kBAAkB,CAAK;IAC/B,OAAO,CAAC,gBAAgB,CAAQ;IAChC,OAAO,CAAC,SAAS,CAAS;IAC1B,OAAO,CAAC,WAAW,CAAK;IACxB,OAAO,CAAC,cAAc,CAAW;IACjC,OAAO,CAAC,SAAS,CAAK;IACtB,OAAO,CAAC,eAAe,CAAc;IACrC,OAAO,CAAC,qBAAqB,CAAK;IAClC,OAAO,CAAC,mBAAmB,CAAK;IAChC,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAK;IAMxC,OAAO,CAAC,uBAAuB,CAAK;IACpC,OAAO,CAAC,QAAQ,CAAC,8BAA8B,CAAK;IACpD,OAAO,CAAC,QAAQ,CAAC,qBAAqB,CAAK;IAQ3C,OAAO,CAAC,QAAQ,CAAC,YAAY,CAA0B;IACvD,OAAO,CAAC,QAAQ,CAAC,0BAA0B,CAAM;IACjD,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAAe;IACjD,OAAO,CAAC,QAAQ,CAAgB;IAGhC,OAAO,CAAC,wBAAwB,CAAK;IACrC,OAAO,CAAC,iBAAiB,CAAc;IACvC,OAAO,CAAC,oBAAoB,CAAuB;IAGnD,OAAO,CAAC,aAAa,CAAK;IAC1B,OAAO,CAAC,uBAAuB,CAAK;IACpC,OAAO,CAAC,QAAQ,CAAC,uBAAuB,CAAM;IAC9C,OAAO,CAAC,QAAQ,CAAC,sBAAsB,CAAe;IACtD,OAAO,CAAC,aAAa,CAAS;IAC9B,OAAO,CAAC,uBAAuB,CAAK;IACpC,OAAO,CAAC,wBAAwB,CAAK;IACrC,OAAO,CAAC,QAAQ,CAAC,6BAA6B,CAAe;IAC7D,OAAO,CAAC,QAAQ,CAAC,wBAAwB,CAAK;IAC9C,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAmB;IACvD,OAAO,CAAC,kBAAkB,CAAK;IAC/B,OAAO,CAAC,eAAe,CAAM;IAC7B,OAAO,CAAC,mBAAmB,CAAuB;IAClD,OAAO,CAAC,iBAAiB,CAAkC;IAC3D,OAAO,CAAC,mBAAmB,CAAK;IAChC,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAU;IAG3C,0EAA0E;IAC1E,MAAM,CAAC,QAAQ,CAAC,0BAA0B,KAAK;IAC/C,yEAAyE;IACzE,MAAM,CAAC,QAAQ,CAAC,0BAA0B,SAAc;IACxD,OAAO,CAAC,oBAAoB,CAAqC;IACjE,qFAAqF;IACrF,OAAO,CAAC,yBAAyB,CAAmF;IACpH,oFAAoF;IACpF,OAAO,CAAC,iBAAiB,CAA+D;gBAE5E,OAAO,EAAE;QACnB,UAAU,EAAE,MAAM,CAAC;QACnB,WAAW,EAAE,MAAM,CAAC;QACpB,IAAI,EAAE,MAAM,CAAC;QACb,QAAQ,CAAC,EAAE,MAAM,CAAC;QAClB,yGAAyG;QACzG,sBAAsB,CAAC,EAAE,MAAM,CAAC;QAChC,uGAAuG;QACvG,mBAAmB,CAAC,EAAE,MAAM,CAAC;QAC7B,+FAA+F;QAC/F,iBAAiB,CAAC,EAAE,MAAM,MAAM,CAAC;KAClC;IAkBD;;OAEG;IACG,KAAK,IAAI,OAAO,CAAC,OAAO,CAAC;IAyD/B;;OAEG;IACG,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC;IAyB3B;;OAEG;IACH,IAAI,OAAO,IAAI,OAAO,CAErB;IAED;;OAEG;IACH,SAAS,IAAI;QACX,OAAO,EAAE,OAAO,CAAC;QACjB,OAAO,EAAE,OAAO,CAAC;QACjB,eAAe,EAAE,MAAM,CAAC;QACxB,WAAW,EAAE,MAAM,CAAC;QACpB,aAAa,EAAE,MAAM,CAAC;QACtB,WAAW,EAAE,OAAO,CAAC;QACrB,mBAAmB,EAAE,MAAM,CAAC;QAC5B,aAAa,EAAE,OAAO,CAAC;QACvB,aAAa,EAAE,MAAM,CAAC;QACtB,eAAe,EAAE,MAAM,CAAC;QACxB,wBAAwB,EAAE,MAAM,CAAC;QACjC,wBAAwB,EAAE,MAAM,CAAC;QACjC,iBAAiB,EAAE,OAAO,CAAC;QAC3B,wBAAwB,EAAE,MAAM,CAAC;QACjC,gBAAgB,EAAE,OAAO,CAAC;QAC1B,yBAAyB,EAAE,MAAM,CAAC;KACnC;IA2BD;;;OAGG;IACH,mBAAmB,IAAI,IAAI;IAa3B;;;OAGG;IACH,sBAAsB,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI;IAI5C;;;;;OAKG;IACG,sBAAsB,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;IAsC9D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;OA8BG;IACG,oBAAoB,CACxB,GAAG,EAAE,qCAAqC,EAC1C,QAAQ,CAAC,EAAE,mCAAmC,GAC7C,OAAO,CAAC,mCAAmC,CAAC;IAmG/C;;;OAGG;IACH,OAAO,CAAC,iBAAiB;IA8QzB;;OAEG;IACH,OAAO,CAAC,WAAW;IAoBnB;;OAEG;IACH,OAAO,CAAC,YAAY;IAgBpB,OAAO,CAAC,WAAW;IAmFnB,OAAO,CAAC,oBAAoB;IAY5B,OAAO,CAAC,iBAAiB;IA0HzB,OAAO,CAAC,gBAAgB;YAWV,WAAW;IAoBzB;;;;;;;;;;OAUG;YACW,sBAAsB;IAYpC;;;;OAIG;IACH,OAAO,CAAC,mBAAmB;IA2E3B;;;OAGG;IACH,OAAO,CAAC,wBAAwB;IAiDhC;;OAEG;IACH,OAAO,CAAC,mBAAmB;IAoC3B;;;;;OAKG;IACH,kBAAkB,CAAC,UAAU,EAAE,oBAAoB,GAAG,IAAI;IAK1D,8DAA8D;IAC9D,2BAA2B,IAAI,MAAM;IAIrC,2DAA2D;IAC3D,oBAAoB,IAAI,MAAM;IAI9B,oEAAoE;IACpE,oBAAoB,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI;IAK3C;;;;;;;;;;;;;;OAcG;IACG,sBAAsB,CAAC,OAAO,EAAE,uBAAuB,GAAG,OAAO,CAAC,qBAAqB,CAAC;IAwG9F;;;;;;OAMG;IACH,OAAO,CAAC,wCAAwC;IAkBhD;;;;;;;OAOG;IACH,OAAO,CAAC,4BAA4B;IAiBpC,kEAAkE;IAClE,gCAAgC,IAAI,MAAM;IAI1C,8EAA8E;IAC9E,0BAA0B,IAAI,IAAI;IAMlC;;;;;;;;;;;;OAYG;IACH,OAAO,CAAC,uBAAuB;IAoC/B;;;;OAIG;IACH,OAAO,CAAC,4BAA4B;IAKpC,OAAO,CAAC,eAAe;IAwKvB;;;;OAIG;IAOH,OAAO,CAAC,kBAAkB;IAgC1B;;;OAGG;IACH,OAAO,CAAC,qBAAqB;IAgC7B;;;OAGG;IACH,OAAO,CAAC,sBAAsB;IAgB9B,OAAO,CAAC,sBAAsB;IAa9B;;;;;;;;;;OAUG;IACH,OAAO,CAAC,uBAAuB;IAwC/B;;OAEG;IACH,OAAO,CAAC,sBAAsB;CAS/B;AAED;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,sEAAsE;IACtE,UAAU,EAAE,MAAM,CAAC;IACnB,kDAAkD;IAClD,UAAU,EAAE,MAAM,CAAC;IACnB,oEAAoE;IACpE,SAAS,EAAE,MAAM,CAAC;CACnB;AAED;;;;;;;;;;;;;GAaG;AACH,wBAAgB,uBAAuB,CAAC,eAAe,EAAE,MAAM,GAAG,iBAAiB,EAAE,CAqDpF;AAID;;;;;;;;;;;;;;;;;;GAkBG;AACH,wBAAgB,6BAA6B,CAAC,OAAO,EAAE,uBAAuB,GAAG,MAAM,CAuCtF"}

package/dist/lifeline/ServerSupervisor.js

@@ -23,6 +23,7 @@ import { fileURLToPath } from 'node:url';
 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 import { detectTmuxPath } from '../core/Config.js';
 import { SleepWakeDetector } from '../core/SleepWakeDetector.js';
+import { cpuLoadRatio, DEFAULT_MAX_LOAD_RATIO } from '../core/cpuStarvation.js';
 import { SafeFsExecutor } from '../core/SafeFsExecutor.js';
 import { SafeGitExecutor } from '../core/SafeGitExecutor.js';
 /** Execute a shell command safely, returning stdout. */
@@ -110,6 +111,16 @@ export class ServerSupervisor extends EventEmitter {
     consecutiveBindFailures = 0;
     bindFailureEscalationThreshold = 2;
     processAliveThreshold = 6; // When process is alive but unresponsive (e.g., high CPU load), require 6 failures (~60s) before restarting
+    // CPU-starvation restart guard: when the box is so oversubscribed (load >>
+    // cores) that the live server can't answer /health, restarting it does NOT
+    // cure the starvation — the fresh server is starved too, it just drops the
+    // in-flight message and loops (the 2026-05-29 restart-loop incident). So while
+    // CPU-starved we DEFER the restart up to a hard cap, then force it (in case
+    // the server is genuinely hung rather than merely starved). Same load-ratio
+    // signal SleepWakeDetector uses (see core/cpuStarvation).
+    maxLoadRatio = DEFAULT_MAX_LOAD_RATIO; // loadavg[0]/cpuCount above this = starved
+    starvationRestartThreshold = 30; // ~5min at 10s checks — force a restart even if starved if unresponsiveness persists this long
+    loadRatioProvider; // injectable for tests; default cpuLoadRatio
     stateDir;
     // Planned restart / maintenance wait — suppress alerts during expected downtime
     maintenanceWaitStartedAt = 0;
@@ -150,6 +161,7 @@ export class ServerSupervisor extends EventEmitter {
         this.stateDir = options.stateDir ?? null;
         this.tmuxPath = detectTmuxPath();
         this.serverSessionName = `${this.projectName}-server`;
+        this.loadRatioProvider = options.loadRatioProvider ?? (() => cpuLoadRatio());
         if (options.maintenanceWaitMinutes !== undefined) {
             this.maintenanceWaitMs = options.maintenanceWaitMinutes * 60_000;
         }
@@ -993,58 +1005,14 @@ export class ServerSupervisor extends EventEmitter {
                 else {
                     this.consecutiveFailures++;
                     if (this.consecutiveFailures >= this.unhealthyThreshold) {
-                        const processAlive = this.isServerSessionAlive();
-                        if (processAlive) {
-                            // Server process exists but isn't responding to health checks.
-                            // Under high CPU load (or during wake transitions), this is normal —
-                            // the event loop is stalled, not the process dead. Use a much higher
-                            // threshold to avoid killing a server that would recover on its own.
-                            const effectiveThreshold = Date.now() < this.wakeTransitionUntil
-                                ? this.unhealthyThreshold // During wake transition: already lenient via counter reset
-                                : this.processAliveThreshold;
-                            if (this.consecutiveFailures < effectiveThreshold) {
-                                if (this.consecutiveFailures === this.unhealthyThreshold) {
-                                    console.log(`[Supervisor] Health check failed but server process is alive — waiting for ${effectiveThreshold} consecutive failures before restart (${this.consecutiveFailures}/${effectiveThreshold})`);
-                                }
-                            }
-                            else {
-                                console.log(`[Supervisor] Server process alive but unresponsive for ${this.consecutiveFailures} checks (~${this.consecutiveFailures * 10}s) — restarting`);
-                                this.handleUnhealthy();
-                            }
-                            if (Date.now() < this.wakeTransitionUntil) {
-                                this.consecutiveFailures = 0; // Reset during wake transition as before
-                            }
-                        }
-                        else {
-                            this.handleUnhealthy();
-                        }
+                        this.evaluateUnhealthyServer();
                     }
                 }
             }
             catch {
                 this.consecutiveFailures++;
                 if (this.consecutiveFailures >= this.unhealthyThreshold) {
-                    const processAlive = this.isServerSessionAlive();
-                    if (processAlive) {
-                        const effectiveThreshold = Date.now() < this.wakeTransitionUntil
-                            ? this.unhealthyThreshold
-                            : this.processAliveThreshold;
-                        if (this.consecutiveFailures < effectiveThreshold) {
-                            if (this.consecutiveFailures === this.unhealthyThreshold) {
-                                console.log(`[Supervisor] Health check failed but server process is alive — waiting for ${effectiveThreshold} consecutive failures before restart (${this.consecutiveFailures}/${effectiveThreshold})`);
-                            }
-                        }
-                        else {
-                            console.log(`[Supervisor] Server process alive but unresponsive for ${this.consecutiveFailures} checks (~${this.consecutiveFailures * 10}s) — restarting`);
-                            this.handleUnhealthy();
-                        }
-                        if (Date.now() < this.wakeTransitionUntil) {
-                            this.consecutiveFailures = 0;
-                        }
-                    }
-                    else {
-                        this.handleUnhealthy();
-                    }
+                    this.evaluateUnhealthyServer();
                 }
             }
             // Check for restart requests from the server (e.g., auto-updater)
@@ -1462,6 +1430,63 @@ export class ServerSupervisor extends EventEmitter {
         this.notifyPendingRemediatorRequestsOnHealthy();
     }
     // ── Unhealthy handling ──────────────────────────────────────────
+    /**
+     * Decide what to do about a server that has failed `>= unhealthyThreshold`
+     * consecutive health checks. Shared by the health-check loop's two failure
+     * paths (unhealthy /health response, and a thrown check).
+     *
+     *   - Process dead → restart immediately.
+     *   - Process alive but unresponsive, below the (load-lenient) threshold →
+     *     keep waiting.
+     *   - Process alive, threshold reached, but the box is CPU-starved → DEFER the
+     *     restart (restarting a starved server only drops the in-flight message
+     *     and loops) until starvation eases or the hard cap forces it.
+     *   - Process alive, threshold reached, not starved → restart.
+     */
+    evaluateUnhealthyServer() {
+        if (!this.isServerSessionAlive()) {
+            // Process is genuinely gone — restart immediately.
+            this.handleUnhealthy();
+            return;
+        }
+        // Server process exists but isn't responding to health checks. Under high
+        // CPU load (or during wake transitions), this is normal — the event loop is
+        // stalled, not the process dead. Use a much higher threshold to avoid
+        // killing a server that would recover on its own.
+        const inWakeTransition = Date.now() < this.wakeTransitionUntil;
+        const effectiveThreshold = inWakeTransition
+            ? this.unhealthyThreshold // During wake transition: already lenient via counter reset
+            : this.processAliveThreshold;
+        if (this.consecutiveFailures < effectiveThreshold) {
+            if (this.consecutiveFailures === this.unhealthyThreshold) {
+                console.log(`[Supervisor] Health check failed but server process is alive — waiting for ${effectiveThreshold} consecutive failures before restart (${this.consecutiveFailures}/${effectiveThreshold})`);
+            }
+        }
+        else if (this.deferRestartForCpuStarvation()) {
+            // Box is CPU-starved — bouncing the server won't help, it only drops the
+            // in-flight message. Defer; the next healthy tick resets the counter.
+            if (this.consecutiveFailures === effectiveThreshold) {
+                console.log(`[Supervisor] Server alive but unresponsive (${this.consecutiveFailures} checks) AND the box is CPU-starved (load ratio ${this.loadRatioProvider().toFixed(2)} > ${this.maxLoadRatio}) — DEFERRING restart; restarting a starved server only drops in-flight messages. Will force-restart at ${this.starvationRestartThreshold} checks (~${this.starvationRestartThreshold * 10}s) if it persists.`);
+            }
+        }
+        else {
+            console.log(`[Supervisor] Server process alive but unresponsive for ${this.consecutiveFailures} checks (~${this.consecutiveFailures * 10}s) — restarting`);
+            this.handleUnhealthy();
+        }
+        if (inWakeTransition) {
+            this.consecutiveFailures = 0; // Reset during wake transition as before
+        }
+    }
+    /**
+     * True when we should hold off restarting an alive-but-unresponsive server
+     * because the machine is CPU-starved — but only up to the hard cap. Past the
+     * cap we restart regardless (the server may be genuinely hung, not starved).
+     */
+    deferRestartForCpuStarvation() {
+        if (this.consecutiveFailures >= this.starvationRestartThreshold)
+            return false; // hard cap — restart even if starved
+        return this.loadRatioProvider() > this.maxLoadRatio;
+    }
     handleUnhealthy() {
         // Circuit breaker — periodic retry instead of permanent death
         if (this.circuitBroken) {