@songsid/agend 2.3.1 → 2.3.2-beta.1

package/dist/daemon.js

@@ -46,7 +46,7 @@ import { buildFleetInstructions } from "./instructions.js";
 import { formatHumanInboundMessage, prepareDeliveryWire, steerInboundEnvelope, BTW_INBOUND_PREFIX, humanBodyHash, humanOwnerObjectId } from "./human-delivery-receipts.js";
 import { formatCrossInstanceInboundMessage, renderCrossInstanceHandoffMetadata, } from "./cross-instance-envelope.js";
 import { agendMessageInInput, bottomRowIsReady, inputAreaText, inputShowsPastedText, pastedTextSignature, pasteLeftInInput, strandedAgendMessageInInput } from "./pane-input-residue.js";
-import { TurnReplyGuard } from "./turn-reply-guard.js";
+import { TurnReplyGuard, formatReplyDropDetection, turnExpectsPersonReply } from "./turn-reply-guard.js";
 import { t } from "./locale.js";
 import { MuseUsageRelay, clearMuseUsageSnapshot } from "./muse-usage-relay.js";
 import { deliveryContentDigest } from "./delivery-outbox.js";
@@ -1245,6 +1245,8 @@ export class Daemon extends EventEmitter {
      * skipResume — the #1209 gate reads this, never the resettable flag.
      */
     bootSkippedResume = false;
+    recoveryCrashLoopFresh = false;
+    recoveryFreshChoice = false;
     startupAborted = false;
     /** First time the current on-screen blocking dialog was seen (0 = none); drives the parked report. */
     dialogParkedSince = 0;
@@ -2203,6 +2205,7 @@ export class Daemon extends EventEmitter {
     }
     async startAdmitted() {
         this.assertSessionOomAdmission();
+        this.recoveryBootWorkPending = true;
         ensureInstanceDir(this.instanceDir);
         writeFileSync(join(this.instanceDir, "daemon.pid"), String(process.pid));
         this.logger.info(`Starting ${this.name}`);
@@ -2211,6 +2214,8 @@ export class Daemon extends EventEmitter {
         // even when the JSON is corrupt so it cannot remain forever.
         const crashStatePath = join(this.instanceDir, "crash-state.json");
         this.bootSkippedResume = false;
+        this.recoveryCrashLoopFresh = false;
+        this.recoveryFreshChoice = false;
         try {
             if (existsSync(crashStatePath)) {
                 const content = readFileSync(crashStatePath, "utf-8");
@@ -2219,6 +2224,10 @@ export class Daemon extends EventEmitter {
                     if (state?.resumeDisabled === true) {
                         this.skipResume = true;
                         this.bootSkippedResume = true;
+                        this.recoveryCrashLoopFresh = (state.reason === undefined || state.reason === "crash_loop")
+                            && Number.isInteger(state.crashesInWindow) && state.crashesInWindow >= 3
+                            && Number.isFinite(state.lastCrashAt) && state.lastCrashAt > 0;
+                        this.recoveryFreshChoice = !this.recoveryCrashLoopFresh;
                         this.logger.warn("Previous crash loop detected — starting without resume");
                     }
                 }
@@ -2367,12 +2376,19 @@ export class Daemon extends EventEmitter {
         // #1209: conditional continuation for the turn interrupted by the last
         // restart, if the gates allow. Detached like the warmup notice; the gate
         // consumes its one-shot marker first and never throws into boot.
-        void this.maybeResumeInterruptedTurn().catch(() => { });
+        const legacyResume = this.maybeResumeInterruptedTurn().catch(() => "unknown");
         // Warmup: wait for CLI idle, then trigger steering reload — but only when
         // the instructions actually changed since the agent last saw them.
         // Skipping the no-op reload saves 10-30s of agent time on every restart
         // where instructions are unchanged.
-        void this.runWarmupInstructionNotice();
+        const warmup = this.runWarmupInstructionNotice();
+        const recoveryOwner = JSON.stringify(this.interactionOwner());
+        void Promise.allSettled([legacyResume, warmup]).then(() => {
+            if (JSON.stringify(this.interactionOwner()) === recoveryOwner) {
+                this.recoveryBootWorkPending = false;
+                this.recoveryBootSettled = true;
+            }
+        });
         if (!this.config.lightweight) {
             // 3. Pipe-pane for prompt detection. Rotate first so a ballooned log from a
             // previous stuck splash (hundreds of MB of ANSI frames) is truncated before
@@ -3974,6 +3990,7 @@ export class Daemon extends EventEmitter {
     clearPendingDeliveries(fleetEpoch) {
         // FleetManager calls this synchronously before sendEscape. Cancel only the
         // current human turn; leave already-forwarded tool requests to settle.
+        this.recoveryCancelled = true;
         this.turnReplyGuard.cancelByUser();
         // #1241: a cancelled turn never confirms — its next edge completes it as
         // intentional. Drop a pending confirmation so it cannot promote afterwards.
@@ -3985,6 +4002,7 @@ export class Daemon extends EventEmitter {
         this.deliveryEpoch = fleetEpoch === undefined
             ? this.deliveryEpoch + 1
             : Math.max(this.deliveryEpoch, fleetEpoch);
+        this.cancelCapacityNudge("the user cancelled");
         this.publishWorkingState("cancelled");
         this.logger.info({ deliveryEpoch: this.deliveryEpoch }, "Pending delivery queue cleared by user cancel");
     }
@@ -4703,6 +4721,88 @@ export class Daemon extends EventEmitter {
     getInteractionSnapshot() {
         return this.interactionObservation.snapshot(this.interactionOwner());
     }
+    ownedWorkingSession = null;
+    recoveryBootSettled = false;
+    recoveryBootWorkPending = false;
+    recoveryLegacyCovered = false;
+    recoveryNativeWork = false;
+    recoveryContinuationId = null;
+    recoveryCancelled = false;
+    workingRecoveryInputOwner = null;
+    noteLegacyRestartReminder() { this.recoveryLegacyCovered = true; }
+    getWorkingRecoveryState() {
+        const observation = this.getWorkingStateObservation();
+        const owner = JSON.stringify(observation.owner), ownsInput = this.workingRecoveryInputOwner === owner;
+        return { owner, configuration: observation.configuration,
+            sessionId: this.ownedWorkingSession, idle: (ownsInput || this.instanceState === "idle") && this.pasteQueueDepth === 0,
+            paused: this.isPaused, stopped: this.isSessionOomStopped(), legacyPending: !this.recoveryBootSettled,
+            legacyCovered: this.recoveryLegacyCovered, nativeWork: !ownsInput && this.recoveryNativeWork,
+            freshChoice: this.recoveryFreshChoice || this.isNewSession && !this.bootSkippedResume, crashLoopFresh: this.recoveryCrashLoopFresh,
+            cancelled: this.recoveryCancelled, background: this.workingBackgroundObservation };
+    }
+    /** Local system work: no user turn, channel reply target, wake or replay of an original request. */
+    async deliverWorkingRecovery(content, delivery, admitted) {
+        const tmux = this.tmux, owner = JSON.stringify(this.interactionOwner()), epoch = this.deliveryEpoch;
+        const current = () => this.tmux === tmux && owner === JSON.stringify(this.interactionOwner())
+            && this.deliveryEpoch === epoch && !this.deliveryWritesStopping && !this.runtimeMonitorsFrozen
+            && !this.spawning && !this.startupAborted && !this.isPaused && !this.healthCheckPaused
+            && this.processStatus === "running" && this.pasteQueueDepth === 0
+            && (this.workingRecoveryInputOwner === owner || this.instanceState === "idle" && !this.recoveryNativeWork)
+            && this.recoveryBootSettled && !this.recoveryCancelled && !this.isSessionOomStopped() && admitted();
+        let begun = false;
+        try {
+            await this.paneWriteLock.run(async () => {
+                if (!tmux || !current())
+                    return;
+                const pane = await tmux.capturePane();
+                const input = this.backend?.readInputRow?.(pane);
+                if (!current() || !input || input.text !== "" || input.collapsedPastes !== 0 || input.queued
+                    || !this.paneAuthoritativelyIdle(pane) || this.inputTransientInPane(pane)
+                    || this.backend?.backgroundWorkRunning?.(pane) === true)
+                    return;
+                if (await tmux.getPaneInputMode?.() !== "raw" || !current())
+                    return;
+                const evidence = await this.durableAttemptEvidence(tmux.getWindowId(), false, false);
+                if (!current())
+                    return;
+                const signature = this.submissionSignature(content), baseline = this.paneEvidence(pane, signature);
+                const attempt = { ...delivery, contentSha256: deliveryContentDigest(content) };
+                if (!this.beginDurableDelivery(attempt, evidence))
+                    return;
+                begun = true;
+                // Own PTY input immediately produces a working observation. It must not
+                // cancel this same fenced write; external queue/cancel/owner guards stay.
+                this.workingRecoveryInputOwner = owner;
+                // Conservatively commit possible input BEFORE paste, not merely before Enter.
+                // A process death at any later boundary must never replay this continuation.
+                if (!this.deliveryOutbox?.markEnterStarted(delivery.deliveryId, this.bootId, delivery.attemptNo))
+                    return;
+                if (!current() || !await tmux.pasteBuffer(content, { guard: current }) || !current())
+                    return;
+                await new Promise(resolve => setTimeout(resolve, NORMAL_ENTER_SETTLE_MS));
+                if (!current() || !await this.sendDeliveryEnterAfterAvailability("working-recovery", current, attempt))
+                    return;
+                const proof = await this.confirmSubmitted(signature, baseline);
+                const committed = this.deliveryOutbox.complete(delivery.deliveryId, this.bootId, delivery.attemptNo, proof === "submitted" ? "delivered" : "uncertain", "local continuation Enter attempted once; no replay");
+                if (committed && proof === "submitted" && current()) {
+                    this.recoveryContinuationId = delivery.deliveryId;
+                    this.publishWorkingState("execution");
+                }
+            });
+        }
+        catch { /* The durable fence below owns ambiguity; never retry a begun write. */ }
+        finally {
+            if (this.workingRecoveryInputOwner === owner) {
+                this.workingRecoveryInputOwner = null;
+                if (this.tmux === tmux && owner === JSON.stringify(this.interactionOwner()))
+                    this.recoveryNativeWork = true;
+            }
+            if (begun)
+                this.deliveryOutbox?.complete(delivery.deliveryId, this.bootId, delivery.attemptNo, "uncertain", "local continuation may have reached input; not replayed");
+            else
+                this.deliveryOutbox?.retryBeforeBegin(delivery.deliveryId, this.bootId, delivery.attemptNo, "working recovery prerequisite changed before input", 5_000);
+        }
+    }
     workingTurnDelivery = null;
     workingBackgroundObservation = null;
     /** Pure metadata read: never discovers sessions, captures a pane or reads backend files. */
@@ -4714,7 +4814,8 @@ export class Daemon extends EventEmitter {
             owner: this.interactionOwner(), backend,
             configuration: createHash("sha256").update(JSON.stringify([backend, this.config.working_directory, profile])).digest("hex"),
             executionState: this.processStatus === "running" ? this.isPaused ? "paused" : this.instanceState : null,
-            processStatus: this.getProcessStatus(), sessionId: null, backendTurnId: null,
+            processStatus: this.getProcessStatus(), sessionId: this.ownedWorkingSession, backendTurnId: null,
+            ...(this.recoveryContinuationId ? { recoveryDeliveryId: this.recoveryContinuationId } : {}),
             backgroundWork: this.workingBackgroundObservation,
             turn: turn ? { generation: turn.generation,
                 deliveryId: this.workingTurnDelivery?.generation === turn.generation ? this.workingTurnDelivery.deliveryId : null,
@@ -5174,6 +5275,9 @@ export class Daemon extends EventEmitter {
             if (unblockedTurn && unblockedTurn.phase === "awaiting")
                 this.maybeProxyReplyOnTurnEnd(pane);
         }
+        if (snapshot.state === "working" && this.recoveryBootSettled
+            && this.workingRecoveryInputOwner !== JSON.stringify(this.interactionOwner()))
+            this.recoveryNativeWork = true;
         this.instanceState = snapshot.state;
         this.acceptedInstanceSnapshot = { ...snapshot };
         // Reuse an already-captured pane and the backend's pure parser, never capture for the journal.
@@ -5290,14 +5394,21 @@ export class Daemon extends EventEmitter {
         // Its late success must not re-arm the turn that the user just cancelled.
         if (!this.isDeliveryEpochCurrent(deliveryEpoch))
             return;
-        // Channel turns only. A cross-instance inbound (from_instance, empty
-        // chat_id) does not update lastChatId, so its proxy reply would land in
-        // whatever USER topic spoke to this instance last — the wrong audience for
-        // a task result, and a stale one (sol's review of #515).
-        // A web message on a fleet with no chat platform has no chat_id (its reply goes to the web chat through the fleet's
-        // web-only sink), but it is a person waiting for an answer all the same: the guard arms for it (alpha.2).
-        if (meta.from_instance || (!meta.chat_id && meta.source !== "web"))
+        // Channel turns started by a person. Cross-instance, schedules (silent or
+        // not), and AgEnD system notices must not expect a reply to a person (#2544):
+        // a proxy or recovery notice would land in the last USER topic, or flag a
+        // heartbeat that was told not to report. A web message on a fleet with no
+        // chat platform has no chat_id (its reply goes to the web chat through the
+        // fleet's web-only sink), but it is a person waiting for an answer all the
+        // same: the guard arms for it (alpha.2).
+        if (!turnExpectsPersonReply(meta)) {
+            this.logger.debug({
+                source: meta.source || null,
+                user: meta.user || null,
+                from_instance: meta.from_instance || null,
+            }, "Reply guard skipped — turn was not started by a person");
             return;
+        }
         // The last non-empty line of what we pasted: everything on screen after it
         // is the agent's own output.
         const inboundMarker = deliveredText.split(/\r?\n/).map(l => l.trim()).filter(Boolean).pop();
@@ -5314,7 +5425,8 @@ export class Daemon extends EventEmitter {
             deliveredText,
             deliveredAt,
         }, this.pendingWork.lastInboundTimestamp());
-        this.workingTurnDelivery = { generation: this.turnReplyGuard.snapshot().generation, deliveryId: meta.delivery_id || null };
+        const armed = this.turnReplyGuard.snapshot();
+        this.workingTurnDelivery = { generation: armed.generation, deliveryId: meta.delivery_id || null };
         this.publishWorkingState("turn_started");
         // #1209: persist the in-flight turn with a seam checkpoint. A marker
         // that survives a restart is an interrupted turn by construction —
@@ -5323,7 +5435,9 @@ export class Daemon extends EventEmitter {
         // the interruption. Reads are windowed/bounded (see turn-resume.ts).
         // A resumed continuation never re-arms (P3a): episode-once means a crash
         // during the continuation boots clean instead of chaining resumes.
-        if (!meta.resumedContinuationOf) {
+        // #2544: a reply that settled before this arm already answered the person —
+        // writing a marker would re-inject a continuation for a finished turn.
+        if (!meta.resumedContinuationOf && !armed.completionDelivered) {
             const resumeBackend = this.config.backend ?? this.backend?.binaryName ?? this.runtimeIdentity?.backend;
             const resumeCwd = this.config.working_directory ?? "";
             writeInFlightTurnMarker(this.instanceDir, {
@@ -5371,6 +5485,7 @@ export class Daemon extends EventEmitter {
         const marker = consumeInFlightTurnMarker(this.instanceDir);
         if (!marker)
             return "none";
+        this.recoveryLegacyCovered = true;
         // P1 fence: capture before any await. A cancel (deliveryEpoch),
         // respawn (spawnGeneration) or stop/pause (launchFenceEpoch) during the
         // grace hold must stale this gate — the push below rechecks, and the
@@ -5737,7 +5852,7 @@ export class Daemon extends EventEmitter {
             reason,
             recoveryStarted: true,
         });
-        void this.deliverDaemonReply(t("inst.reply_drop_retrying"), "replydrop", "Reply-drop status", turn.target, true);
+        this.deliverReplyDropNotice("inst.reply_drop_retrying", "replydrop", "Reply-drop status", turn, reason, false);
         this.queueReplyRecoveryPrompt(turn);
     }
     reportReplyDropWithoutRetry(turn, reason) {
@@ -5748,7 +5863,7 @@ export class Daemon extends EventEmitter {
             reason,
             recoveryStarted: false,
         });
-        void this.deliverDaemonReply(t("inst.reply_drop_unknown"), "replydrop", "Unconfirmed reply status", turn.target, true);
+        this.deliverReplyDropNotice("inst.reply_drop_unknown", "replydrop", "Unconfirmed reply status", turn, reason, true);
     }
     queueReplyRecoveryPrompt(turn) {
         const deliveryEpoch = this.deliveryEpoch;
@@ -5841,12 +5956,20 @@ export class Daemon extends EventEmitter {
             generation: turn.generation,
             reason,
         });
-        const now = Date.now();
-        if (this.lastReplyDropWarningAt !== 0
-            && now - this.lastReplyDropWarningAt < REPLY_DROP_WARNING_COOLDOWN_MS)
-            return;
-        this.lastReplyDropWarningAt = now;
-        void this.deliverDaemonReply(t("inst.reply_drop_unrecovered"), "replydropwarn", "Reply-drop recovery warning", turn.target, true);
+        this.deliverReplyDropNotice("inst.reply_drop_unrecovered", "replydropwarn", "Reply-drop recovery warning", turn, reason, true);
+    }
+    /** Channel notice for a reply-drop. `rateLimit` shares the per-instance cooldown (#2544). */
+    deliverReplyDropNotice(key, kind, title, turn, reason, rateLimit) {
+        if (rateLimit) {
+            const now = Date.now();
+            if (this.lastReplyDropWarningAt !== 0
+                && now - this.lastReplyDropWarningAt < REPLY_DROP_WARNING_COOLDOWN_MS) {
+                this.logger.debug({ reason, kind }, "Reply-drop channel notice rate-limited");
+                return;
+            }
+            this.lastReplyDropWarningAt = now;
+        }
+        void this.deliverDaemonReply(t(key, formatReplyDropDetection(turn, reason)), kind, title, turn.target, true);
     }
     /** Relay the pane's final text to the channel, marked as a daemon proxy reply. */
     async sendProxyReply(pane, target) {
@@ -6444,6 +6567,7 @@ export class Daemon extends EventEmitter {
     }
     /** Stop every runtime poller/watcher while preserving IPC and daemon state. */
     freezeRuntimeMonitors() {
+        this.cancelCapacityNudge("the instance was stopped, paused or respawned");
         this.runtimeMonitorsFrozen = true;
         this.launchFenceEpoch++;
         this.interactionObservation.reset(this.interactionOwner());
@@ -6556,6 +6680,10 @@ export class Daemon extends EventEmitter {
             this.config.warm_cap = update.warm_cap;
         else if (update.warm_cap === null)
             delete this.config.warm_cap;
+        if (typeof update.resume_interrupted_work === "boolean")
+            this.config.resume_interrupted_work = update.resume_interrupted_work;
+        else if (update.resume_interrupted_work === null)
+            delete this.config.resume_interrupted_work;
         const nextAutoPause = typeof update.auto_pause_after === "number"
             && Number.isFinite(update.auto_pause_after)
             && update.auto_pause_after >= 0
@@ -7828,7 +7956,8 @@ export class Daemon extends EventEmitter {
     /**
      * Arm the "keep going" nudge for the error occurrence just reported (`pattern` must be the one the lifecycle was
      * handed). The nudge is sent once, `delayMs` from now on the MONOTONIC clock (a wall-clock step changes nothing),
-     * and only if nothing has changed by then — see {@link tickCapacityNudge}. Returns false when there is nothing to arm.
+     * and only if the failed turn still owns an empty idle composer — see {@link tickCapacityNudge}.
+     * Cosmetic footer redraws are not new work. Returns false when there is nothing to arm.
      */
     armCapacityNudge(pattern, delayMs) {
         const episode = this.lastErrorEpisode;
@@ -7847,9 +7976,12 @@ export class Daemon extends EventEmitter {
         // lookahead on the counting pattern: whole-pane dedup would max at 1 and miss later hits. See #1215.)
         if (!codexLiveRowMatches(episode.pane, pattern))
             return false;
+        this.cancelCapacityNudge("the instance has work of its own");
         this.capacityNudge = {
             pane: episode.pane, pattern, spawn: episode.spawn, fence: episode.fence,
             epoch: this.deliveryEpoch, dueAt: performance.now() + delayMs,
+            episode,
+            promiseTarget: null,
         };
         this.logger.info({ delayMs }, "Model-capacity nudge armed");
         return true;
@@ -7864,7 +7996,7 @@ export class Daemon extends EventEmitter {
         const nudge = this.capacityNudge, episode = this.lastErrorEpisode;
         const tmux = this.tmux, attempt = this.launchAttempt, boot = this.bootId;
         this.cancelCapacityNudge("retries exhausted");
-        if (!nudge || !episode || !tmux)
+        if (!nudge || !episode || nudge.episode !== episode || !tmux)
             return null;
         const window = tmux.getWindowId?.();
         const current = () => this.tmux === tmux && tmux.getWindowId?.() === window
@@ -7872,9 +8004,7 @@ export class Daemon extends EventEmitter {
             && episode === this.lastErrorEpisode && this.isDeliveryEpochCurrent(nudge.epoch)
             && !this.startupAborted && !this.spawning && !this.isPaused
             && this.getProcessStatus() === "running" && this.instanceState === "idle" && this.pasteQueueDepth === 0;
-        const accepts = (pane) => pane === nudge.pane && this.paneAuthoritativelyIdle(pane)
-            && this.backend?.isPeriodicRedrawIdlePane?.(pane) === true
-            && codexLiveRowMatches(pane, nudge.pattern) && !this.backend?.backgroundWorkRunning?.(pane);
+        const accepts = (pane) => this.capacityNudgePaneCurrent(nudge, pane);
         const revalidate = () => this.paneWriteLock.run(async () => {
             if (!current())
                 return false;
@@ -7896,10 +8026,35 @@ export class Daemon extends EventEmitter {
     }
     /** Drop an armed nudge (the user stopped / paused the instance, or the lifecycle gave up on it). */
     cancelCapacityNudge(reason) {
-        if (!this.capacityNudge)
+        const nudge = this.capacityNudge;
+        if (!nudge)
             return;
         this.capacityNudge = null;
         this.logger.info({ reason }, "Model-capacity nudge cancelled");
+        this.reportCapacityNudgeCancelled(nudge, reason);
+    }
+    /** Only a dispatched channel promise needs a later cancellation notice. */
+    markCapacityNudgePromised(target) {
+        if (this.capacityNudge)
+            this.capacityNudge.promiseTarget = target;
+    }
+    reportCapacityNudgeCancelled(nudge, reason) {
+        const target = nudge.promiseTarget;
+        if (!target)
+            return;
+        nudge.promiseTarget = null;
+        this.emit("capacity_nudge_cancelled", { name: this.name, reason, target });
+    }
+    /**
+     * A failed turn must still be the last live transcript item, with no new
+     * error occurrence, draft, queue, busy status or background work. Codex's
+     * own periodic-redraw idle classifier recognises its empty composer/footer;
+     * never infer inactivity from byte equality or a cached idle state alone.
+     */
+    capacityNudgePaneCurrent(nudge, pane) {
+        return this.paneAuthoritativelyIdle(pane) && this.backend?.isPeriodicRedrawIdlePane?.(pane) === true
+            && !this.backend?.backgroundWorkRunning?.(pane) && codexLiveRowMatches(pane, nudge.pattern)
+            && Daemon.countOccurrences(nudge.pattern, pane) === Daemon.countOccurrences(nudge.pattern, nudge.pane);
     }
     /** The first live composer of a spawn fixes what is history (see {@link nudgeBaselines}). */
     takeNudgeBaselines(pane, patterns) {
@@ -7922,8 +8077,8 @@ export class Daemon extends EventEmitter {
     }
     /**
      * One monitor poll's look at an armed nudge. It is CANCELLED — never postponed — when anything about the instance
-     * moved: a stop / pause / respawn (spawn generation or monitor epoch), a user cancel (delivery epoch), the pane no
-     * longer being exactly the one the error was seen on (it moved on by itself, a person typed, another dialog), the
+     * moved: a stop / pause / respawn (spawn generation or monitor epoch), a user cancel (delivery epoch), the pane
+     * no longer belonging to the failed turn (new output, a draft, another dialog), the
      * agent working, or a message waiting for delivery. Only when it is due and still untouched is it fired — once — and
      * the injection re-verifies everything under the pane-write lock.
      */
@@ -7933,10 +8088,11 @@ export class Daemon extends EventEmitter {
             return;
         const gone = nudge.spawn !== this.spawnGeneration || nudge.fence !== this.launchFenceEpoch ? "the instance was stopped, paused or respawned"
             : !this.isDeliveryEpochCurrent(nudge.epoch) ? "the user cancelled"
-                : this.isPaused || this.healthCheckPaused || this.getProcessStatus() === "stopped" ? "the instance is not running"
-                    : pane !== nudge.pane ? "the screen changed"
+                : nudge.episode !== this.lastErrorEpisode ? "the instance has work of its own"
+                    : this.isPaused || this.healthCheckPaused || this.getProcessStatus() === "stopped" ? "the instance is not running"
                         : this.instanceState === "working" || this.pasteQueueDepth > 0 ? "the instance has work of its own"
-                            : null;
+                            : !this.capacityNudgePaneCurrent(nudge, pane) ? "the failed turn is no longer idle with an empty input"
+                                : null;
         if (gone) {
             this.cancelCapacityNudge(gone);
             return;
@@ -7944,29 +8100,38 @@ export class Daemon extends EventEmitter {
         if (performance.now() < nudge.dueAt)
             return;
         this.capacityNudge = null; // fired: once per episode
-        void this.sendCapacityNudge(nudge).catch(err => this.logger.warn({ err: err.message }, "Model-capacity nudge failed"));
+        void this.sendCapacityNudge(nudge).catch(err => {
+            this.logger.warn({ err: err.message }, "Model-capacity nudge failed");
+            this.reportCapacityNudgeCancelled(nudge, "the retry could not be submitted");
+        });
     }
     async sendCapacityNudge(nudge) {
         // Everything the nudge was armed under, asked again after EVERY await that precedes a write — inside the paste
         // primitive too (its capture, its paste, its pause, its Enter, its retry): once any of it is false nothing more is
         // written. A message queued behind the lock, or the agent starting a turn, is "work of its own" like at the poll.
         const current = () => nudge.spawn === this.spawnGeneration && nudge.fence === this.launchFenceEpoch
+            && nudge.episode === this.lastErrorEpisode
             && this.isDeliveryEpochCurrent(nudge.epoch) && !this.isPaused && !this.healthCheckPaused
             && !this.startupAborted && this.getProcessStatus() !== "stopped"
             && this.pasteQueueDepth === 0 && this.instanceState !== "working";
         // Exclusive with every delivery: nothing can paste a turn between the verdict below and the nudge.
-        await this.paneWriteLock.run(async () => {
-            if (!this.tmux)
-                return;
-            // ONE capture decides: it must be exactly the screen the error was seen on AND an idle composer. (A second picture
-            // vouching for the first — "same screen" here, "idle" there — would let a half-typed draft through.)
-            const sent = await this.submitSystemPaste(t("inst.codex_capacity_nudge_text"), "capacity-continue", {
-                current,
-                accept: pane => pane === nudge.pane && this.paneAuthoritativelyIdle(pane)
-                    && codexLiveRowMatches(pane, nudge.pattern),
+        let sent = false;
+        try {
+            await this.paneWriteLock.run(async () => {
+                if (!this.tmux)
+                    return;
+                // ONE capture decides failed-turn ownership and empty/idle input.
+                sent = await this.submitSystemPaste(t("inst.codex_capacity_nudge_text"), "capacity-continue", {
+                    current,
+                    accept: pane => this.capacityNudgePaneCurrent(nudge, pane),
+                });
+                this.logger.info({ sent }, "Model-capacity nudge attempted");
             });
-            this.logger.info({ sent }, "Model-capacity nudge attempted");
-        });
+        }
+        finally {
+            if (!sent)
+                this.reportCapacityNudgeCancelled(nudge, "the retry could not be submitted");
+        }
     }
     /** What identifies the request on screen: the backend's own identity when it has one, else the whole screen verbatim. */
     static screenOf(dialog, pane) {
@@ -10233,7 +10398,9 @@ export class Daemon extends EventEmitter {
                 // false and must not clear the window either.
                 if (replyAttempt?.completionAction) {
                     const snap = this.turnReplyGuard.snapshot();
-                    if (snap?.generation === replyAttempt.generation && snap.phase === "awaiting"
+                    // generation 0 is a pre-arm token applied to the live generation (#2544).
+                    if (snap && (snap.generation === replyAttempt.generation || replyAttempt.generation === 0)
+                        && snap.phase === "awaiting"
                         && snap.completionDelivered && !snap.cancelledByUser) {
                         this.clearReplyGuardConfirm();
                     }
@@ -10883,9 +11050,14 @@ export class Daemon extends EventEmitter {
     beginSpawn() {
         this.unclearOomStopped = false;
         this.publishWorkingState("before_spawn");
+        this.recoveryContinuationId = null;
         this.workingBackgroundObservation = null;
+        this.ownedWorkingSession = null;
+        this.recoveryBootSettled = false;
+        this.recoveryNativeWork = false;
+        this.recoveryCancelled = false;
         this.dialogAnswers = null;
-        this.capacityNudge = null;
+        this.cancelCapacityNudge("the instance was stopped, paused or respawned");
         this.lastErrorEpisode = null;
         this.nudgeBaselines = null;
         // A relaunch settles any outstanding quit watch: the CLI came back.
@@ -10916,6 +11088,8 @@ export class Daemon extends EventEmitter {
         this.spawnDepth = Math.max(0, this.spawnDepth - 1);
         if (this.spawnDepth > 0)
             return; // an enclosing spawn is still running
+        if (!this.recoveryBootWorkPending)
+            this.recoveryBootSettled = true;
         this.spawning = false;
         const resolve = this.resolveSpawnSettled;
         this.spawnSettled = null;
@@ -12369,6 +12543,13 @@ export class Daemon extends EventEmitter {
             const sid = this.backend?.getSessionId();
             if (!sid)
                 return;
+            if ((this.config.backend ?? "claude-code") === "claude-code" && Buffer.byteLength(sid) <= 256
+                && !/[\u0000-\u001f\u007f]/.test(sid) && this.ownedWorkingSession !== sid) {
+                this.ownedWorkingSession = sid;
+                // Reuse this existing owned checkpoint. An idle instance may have no
+                // later state edge before an OOM; its journal must not stay unknown.
+                this.publishWorkingState("process");
+            }
             const path = join(this.instanceDir, "session-id");
             // Idle observations happen after turns. Avoid needless writes (and mtime
             // churn) while still updating the marker when /new changes the session.