@integrity-labs/agt-cli 0.28.829 → 0.28.831

package/dist/{chunk-6HTCP6D5.js → chunk-N4QVPPYS.js}

@@ -18,7 +18,7 @@ import {
   rotateDailySession,
   sessionFileExists,
   todayLocalIso
-} from "./chunk-GJY3KZG6.js";
+} from "./chunk-4ZPOP7EW.js";
 import {
   reapOrphanChannelMcps
 } from "./chunk-XWVM4KPK.js";
@@ -5326,4 +5326,4 @@ export {
   stopAllSessionsAndWait,
   getProjectDir
 };
-//# sourceMappingURL=chunk-6HTCP6D5.js.map
+//# sourceMappingURL=chunk-N4QVPPYS.js.map

package/dist/{chunk-E75YCJTE.js → chunk-PODKITVQ.js}

@@ -16,7 +16,7 @@ import {
   parseEnvIntegrations,
   shellQuote,
   summarizeUnanswerablePane
-} from "./chunk-6HTCP6D5.js";
+} from "./chunk-N4QVPPYS.js";
 import {
   BIND_FAILURE_QUARANTINE_THRESHOLD,
   INTEGRATIONS_SECTION_END,
@@ -74,7 +74,7 @@ import {
   sessionTranscriptDir,
   worseConnectivityOutcome,
   wrapScheduledTaskPrompt
-} from "./chunk-GJY3KZG6.js";
+} from "./chunk-4ZPOP7EW.js";
 import {
   parsePsRows
 } from "./chunk-XWVM4KPK.js";
@@ -6535,7 +6535,7 @@ function exchangeFailureKind(err) {
 }
 
 // src/lib/api-client.ts
-var agtCliVersion = true ? "0.28.829" : "dev";
+var agtCliVersion = true ? "0.28.831" : "dev";
 var lastConfigHash = null;
 function setConfigHash(hash) {
   lastConfigHash = hash && hash.length > 0 ? hash : null;
@@ -10843,4 +10843,4 @@ export {
   managerInstallSystemUnitCommand,
   managerUninstallSystemUnitCommand
 };
-//# sourceMappingURL=chunk-E75YCJTE.js.map
+//# sourceMappingURL=chunk-PODKITVQ.js.map

package/dist/{claude-pair-runtime-UTDZXJKO.js → claude-pair-runtime-PUWAIK73.js}

@@ -100,7 +100,7 @@ async function spawnPairSession(session) {
     return { ok: true };
   } catch {
   }
-  const { resolveClaudeBinary } = await import("./persistent-session-6YTEDS3H.js");
+  const { resolveClaudeBinary } = await import("./persistent-session-W23U7GQU.js");
   const claudeBin = resolveClaudeBinary();
   const pairEnv = {
     ...process.env,
@@ -373,4 +373,4 @@ export {
   startClaudePair,
   submitClaudePairCode
 };
-//# sourceMappingURL=claude-pair-runtime-UTDZXJKO.js.map
+//# sourceMappingURL=claude-pair-runtime-PUWAIK73.js.map

package/dist/lib/manager-worker.js

@@ -60,7 +60,7 @@ import {
   safeWriteJsonAtomic,
   setConfigHash,
   tripClass
-} from "../chunk-E75YCJTE.js";
+} from "../chunk-PODKITVQ.js";
 import {
   getProjectDir as getProjectDir2,
   getReadyTasks,
@@ -126,7 +126,7 @@ import {
   takeZombieDetection,
   toOpencodeModel,
   writeEgressAllowlist
-} from "../chunk-6HTCP6D5.js";
+} from "../chunk-N4QVPPYS.js";
 import {
   ACCOUNT_ENFORCEMENT_MARKER_FILENAME,
   AnchorSessionClient,
@@ -221,7 +221,7 @@ import {
   subagentActivityAgeSeconds,
   sumTranscriptUsageInWindow,
   transcriptActivityAgeSeconds
-} from "../chunk-GJY3KZG6.js";
+} from "../chunk-4ZPOP7EW.js";
 import {
   reapOrphanChannelMcps
 } from "../chunk-XWVM4KPK.js";
@@ -12928,7 +12928,7 @@ var pendingDayRolloverReset = /* @__PURE__ */ new Set();
 var dayRolloverInboundHold = /* @__PURE__ */ new Map();
 var INBOUND_HOLD_EPISODE_GAP_MS = 12e4;
 async function channelInboundActivityAgeSecondsFor(codeName) {
-  const { newestPendingInboundActivityMtimeMs } = await import("../responsiveness-probe-5B2WA5G3.js");
+  const { newestPendingInboundActivityMtimeMs } = await import("../responsiveness-probe-IRQKIEBW.js");
   const newest = newestPendingInboundActivityMtimeMs(dirname10(paneLogPath(codeName)));
   if (newest === null) return null;
   return Math.max(0, Math.floor((Date.now() - newest) / 1e3));
@@ -13596,7 +13596,7 @@ var agentRestartTimezoneInputs = /* @__PURE__ */ new Map();
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
 var lastResponsivenessProbeAt = 0;
-var agtCliVersion = true ? "0.28.829" : "dev";
+var agtCliVersion = true ? "0.28.831" : "dev";
 function resolveBrewPath(execFileSync2) {
   try {
     const out = execFileSync2("which", ["brew"], { timeout: 5e3 }).toString().trim();
@@ -15186,7 +15186,7 @@ function flushRestartedAgentDiagnostics(hostId, codeNames) {
   if (codeNames.length === 0) return;
   void (async () => {
     try {
-      const { collectDiagnostics } = await import("../persistent-session-6YTEDS3H.js");
+      const { collectDiagnostics } = await import("../persistent-session-W23U7GQU.js");
       await api.post("/host/heartbeat", {
         host_id: hostId,
         agent_diagnostics: collectDiagnostics(codeNames, quarantineEntriesFor, claudeMdSizeFor, spawnOutcomeForDiagnostics, pidPressureFor, forwardedToolsFor)
@@ -15328,7 +15328,7 @@ async function pollCycleInner() {
     }
     try {
       const { detectHostSecurity } = await import("../host-security-6PDFG7F5.js");
-      const { collectDiagnostics } = await import("../persistent-session-6YTEDS3H.js");
+      const { collectDiagnostics } = await import("../persistent-session-W23U7GQU.js");
       const diagCodeNames = [...agentState.persistentSessionAgents];
       const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames, quarantineEntriesFor, claudeMdSizeFor, spawnOutcomeForDiagnostics, pidPressureFor, forwardedToolsFor) : void 0;
       let tailscaleHostname;
@@ -15497,7 +15497,7 @@ async function pollCycleInner() {
         collectPanelessActivityProbes,
         getResponsivenessIntervalMs,
         occupancyQualificationClassifications
-      } = await import("../responsiveness-probe-5B2WA5G3.js");
+      } = await import("../responsiveness-probe-IRQKIEBW.js");
       const probeIntervalMs = getResponsivenessIntervalMs();
       if (now - lastResponsivenessProbeAt > probeIntervalMs) {
         const probeCodeNames = [...agentState.persistentSessionAgents];
@@ -15614,7 +15614,7 @@ async function pollCycleInner() {
           collectResponsivenessProbes,
           livePendingInboundOldestAgeSeconds,
           parkPendingInbound
-        } = await import("../responsiveness-probe-5B2WA5G3.js");
+        } = await import("../responsiveness-probe-IRQKIEBW.js");
         const { getProjectDir: wedgeProjectDir } = await import("../scheduler-engine-NDP36U7O.js");
         const wedgeNow = /* @__PURE__ */ new Date();
         const liveAgents = agentState.persistentSessionAgents;
@@ -15770,7 +15770,7 @@ async function pollCycleInner() {
     }
     try {
       const { scrapeMcpFailedBannerCount } = await import("../pane-mcp-banner-scraper-JA437JIB.js");
-      const { probeSessionAuth } = await import("../session-auth-dead-LFNU6AUU.js");
+      const { probeSessionAuth } = await import("../session-auth-dead-VDDXVJTX.js");
       const observations = [];
       const pendingCacheCommits = [];
       const modelApiErrorReportingOn = hostFlagStore().getBoolean("model-api-error-reporting");
@@ -20003,7 +20003,7 @@ async function handleRestartDoorbell(agentId, requestedAt, restartReason) {
     void api.post("/host/restart-ack", { host_id: hostId, agent_id: agentId, restart_requested_at: requestedAt }).catch((err) => log(`[restart-lane] ack failed for '${codeName}': ${err.message}`));
     void (async () => {
       try {
-        const { collectDiagnostics } = await import("../persistent-session-6YTEDS3H.js");
+        const { collectDiagnostics } = await import("../persistent-session-W23U7GQU.js");
         await api.post("/host/heartbeat", {
           host_id: hostId,
           agent_diagnostics: collectDiagnostics([codeName], quarantineEntriesFor, claudeMdSizeFor, spawnOutcomeForDiagnostics, pidPressureFor, forwardedToolsFor)
@@ -20049,7 +20049,7 @@ async function respawnAgentAfterMcpStop(codeName, reason) {
     }
     try {
       const hostId = await getHostId();
-      const { collectDiagnostics } = await import("../persistent-session-6YTEDS3H.js");
+      const { collectDiagnostics } = await import("../persistent-session-W23U7GQU.js");
       await api.post("/host/heartbeat", {
         host_id: hostId,
         agent_diagnostics: collectDiagnostics([codeName], quarantineEntriesFor, claudeMdSizeFor, spawnOutcomeForDiagnostics, pidPressureFor, forwardedToolsFor)
@@ -20685,7 +20685,7 @@ async function processClaudePairSessions(agents) {
     killPairSession,
     pairTmuxSession,
     finalizeClaudePairOnboarding
-  } = await import("../claude-pair-runtime-UTDZXJKO.js");
+  } = await import("../claude-pair-runtime-PUWAIK73.js");
   for (const pairId of pendingResp.cancelled_pair_ids ?? []) {
     log(`[claude-pair] sweeping orphan tmux session for pair ${pairId.slice(0, 8)}`);
     const killed = await killPairSession(pairTmuxSession(pairId));

package/dist/mcp/augmented-admin.js

@@ -21812,7 +21812,7 @@ server.registerTool(
 );
 server.registerTool(
   "debug_inspect_dead_letters",
-  { description: "List an agent's DEAD-LETTERED / quarantined inbound \u2014 customer messages that were parked on the host instead of delivered (so the agent never saw them), rather than silently lost. Use when a user says \"the agent never replied\" but nothing shows in the normal flow, or to audit what fell into the dead-letter path. Resolves the agent's current host and reads two on-disk stores over SSM: `stale` (markers moved aside on a wedge-respawn because the agent couldn't receive) and `cleared:<stamp>` (markers the clear_pending_inbound remedial action moved aside). Returns { agent_id, code_name, host, markers[], total, ssm_status } where each marker is { index, store, channel (slack|telegram), marker_name, received_at, undeliverable, discretionary, replay_count, conversation, message_ref } \u2014 routing metadata only, NEVER the message body. To re-inject one, pass its exact { channel, store, marker_name } to debug_replay_dead_letter. Fails closed for an unauthorized org; every call is audited as a cross-org host access. Pass { agent_id }.", inputSchema: inspectDeadLettersSchema },
+  { description: 'List an agent\'s DEAD-LETTERED / quarantined inbound \u2014 customer messages that were parked on the host instead of delivered (so the agent never saw them), rather than silently lost. Use when a user says "the agent never replied" but nothing shows in the normal flow, or to audit what fell into the dead-letter path. Resolves the agent\'s current host and reads two on-disk stores over SSM: `stale` (markers moved aside on a wedge-respawn because the agent couldn\'t receive) and `cleared:<stamp>` (markers the clear_pending_inbound remedial action moved aside). Returns { agent_id, code_name, host, markers[], total, ssm_status } where each marker is { index, store, channel (slack|telegram), marker_name, received_at, undeliverable, discretionary, replay_count, has_payload, conversation, message_ref } \u2014 routing metadata only, NEVER the message body. `has_payload` is a TRI-STATE (true | false | null) saying whether a replay payload exists, never the payload itself - `null` means this host did not report it (an agt-cli predating the field, or an unparseable marker) and must NOT be read as \\"no payload\\": it is what separates a marker that was eligible for replay and never re-pushed from one that was structurally unreplayable, and those need opposite fixes. `replay_count` is replays already PUSHED \u2014 0 means none was ever attempted (the marker write omits the field; the replay path increments it), so 0 is a fact rather than missing data and says nothing about the payload. To re-inject one, pass its exact { channel, store, marker_name } to debug_replay_dead_letter. Fails closed for an unauthorized org; every call is audited as a cross-org host access. Pass { agent_id }.', inputSchema: inspectDeadLettersSchema },
   async (args) => {
     try {
       const result = await client.inspectDeadLetters(args);

package/dist/mcp/direct-chat-channel.js

@@ -30289,10 +30289,19 @@ var UNDELIVERABLE_NOTICE_THROTTLE_MS = 5 * 60 * 1e3;
 function shouldPostUndeliverableNotice(lastNoticeAtMs, nowMs, throttleMs = UNDELIVERABLE_NOTICE_THROTTLE_MS) {
   return lastNoticeAtMs == null || nowMs - lastNoticeAtMs >= throttleMs;
 }
-function undeliverableNoticeText(replay = channelReplayEnabled()) {
-  return noticePromiseIsHonest(replay) ? "\u23F3 I can't get to this right now \u2014 no need to resend; I'll pick it up automatically as soon as I'm free." : "\u23F3 I can't get to this right now. Please resend in a few minutes \u2014 I may not see this one.";
+function undeliverableNoticeText(replay) {
+  const honest = noticePromiseIsHonest(replay);
+  return honest ? "\u23F3 I can't get to this right now \u2014 no need to resend; I'll pick it up automatically as soon as I'm free." : "\u23F3 I can't get to this right now. Please resend in a few minutes \u2014 I may not see this one.";
+}
+function undeliverableNoticeTextForRepeat(i) {
+  if (i.priorNotices <= 0) return undeliverableNoticeText(i.replay);
+  if (i.priorNotices === 1) {
+    return noticePromiseIsHonest(i.replay) ? "\u23F3 Still stuck on this one \u2014 the automatic pickup I mentioned has not happened. Please resend when you can." : "\u23F3 Still stuck on this one, and I still have not picked it up. Resending is the reliable way through.";
+  }
+  return null;
 }
 function noticePromiseIsHonest(replay) {
+  if (replay === void 0) return false;
   if (typeof replay === "boolean") return replay;
   return markerReplayable({
     enabled: replay.enabled ?? channelReplayEnabled(),
@@ -30383,6 +30392,13 @@ function channelReplayEnabled() {
     defaultValue: true
   });
 }
+function combineMarkerReplayFacts(markers) {
+  if (markers.length === 0) return null;
+  return {
+    hasPayload: markers.every((m) => m.hasPayload),
+    discretionary: markers.some((m) => m.discretionary === true)
+  };
+}
 function markerReplayable(i) {
   if (!i.enabled) return false;
   if (!i.hasPayload) return false;
@@ -36674,7 +36690,7 @@ var FLAG_REGISTRY = [
   },
   {
     key: "auto-resume-enabled",
-    description: "DO NOT ARM YET (ENG-8395): this flag has no `since` set until a follow-up bisects it from the published agt-cli dist, and an unset `since` makes the flip-reach estimate report FULL reach for a reader that is host-side \u2014 it will tell you a flip reached hosts that have no code to honour it. Arm only once that follow-up has landed. Bounded BLIND auto-resume for circuit-breaker pauses (ENG-6088): at most one unattended resume per trip cycle, fired once the trip driver has been quiet for the configured window and only for provisioning-reload trips (the ENG-7577 class gate \u2014 a crash-class trip still needs an operator). Superseded wherever `resume-reconciler` is on: that path proves the dependency recovered instead of assuming it, and the blind timer is its flag-off fallback. Boolean gate; ships dark. ENG-8395 promoted this from a raw host env var because AGT_AUTO_RESUME_ENABLED lives in the host env file, so its grain is per-HOST \u2014 one flip arms every agent that host runs (agt-aws-1 carries 14), which made the canary-then-fleet rollout its own documentation prescribes impossible. As a registry flag it resolves per org/team/host/AGENT, so one agent can genuinely be cut over first. Read host-side by the manager (agt-cli), not centrally.",
+    description: "Bounded BLIND auto-resume for circuit-breaker pauses (ENG-6088): at most one unattended resume per trip cycle, fired once the trip driver has been quiet for the configured window and only for provisioning-reload trips (the ENG-7577 class gate \u2014 a crash-class trip still needs an operator). Superseded wherever `resume-reconciler` is on: that path proves the dependency recovered instead of assuming it, and the blind timer is its flag-off fallback. Boolean gate; ships dark. ENG-8395 promoted this from a raw host env var because AGT_AUTO_RESUME_ENABLED lives in the host env file, so its grain is per-HOST \u2014 one flip arms every agent that host runs (agt-aws-1 carries 14), which made the canary-then-fleet rollout its own documentation prescribes impossible. As a registry flag it resolves per org/team/host/AGENT, so one agent can genuinely be cut over first. Read host-side by the manager (agt-cli), not centrally.",
     flagType: "boolean",
     // Declared safe value is `false` — identical to today's behaviour on a host
     // that has never set the env var. Fail-safe direction on a flag-DB read
@@ -36691,16 +36707,41 @@ var FLAG_REGISTRY = [
     // An unattended resume is a real availability change — the agent starts
     // spawning sessions again with no human in the loop — so arming it is a
     // deliberate, confirmed write (ADR-0022 sensitive-flag confirm).
-    sensitive: true
-    // ENG-8395 AC5: `since` is deliberately UNSET at merge. It must be bisected
-    // from the PUBLISHED agt-cli dist once the artifact carrying the host-side
-    // reader exists, never guessed — auto-resume.ts is host-side, so an
-    // under-`since` host has no consumer for this flag and a guessed value makes
-    // the flip-reach modal promise reach it does not have (the defect caught on
-    // ENG-8344 / PR #4014). Precedent for the bisect method: slack-hot-thread-guard
-    // = 0.28.339, slack-scheduled-channel-guard = 0.28.421. Setting it later is
-    // safe: `since` is excluded from projectDefinition, so it does not roll
+    sensitive: true,
+    // ENG-10034: bisected from the PUBLISHED agt-cli dists, not from git history —
+    // auto-resume.ts is host-side, so what matters is the first ARTIFACT a host
+    // could be running that contains a consumer for this flag.
+    //
+    // METHOD, so it can be re-run rather than trusted: download each candidate
+    // tarball from the npm registry and grep its contents for the flag-specific
+    // wiring. Binary search over the 22 versions published either side of
+    // ENG-8395's merge gave a clean one-release boundary, eight minutes wide:
+    //
+    //   0.28.506  2026-08-03T22:29:22Z  readAutoResumeEnvOverride ✗  AUTO_RESUME_FLAG_KEY ✗  'auto-resume-enabled' ✗
+    //   0.28.507  2026-08-03T22:37:00Z  readAutoResumeEnvOverride ✓  AUTO_RESUME_FLAG_KEY ✓  'auto-resume-enabled' ✓
+    //
+    // All three symbols are checked rather than one. `getBooleanForAgent` alone
+    // would have been the wrong probe — it is present in BOTH builds, because the
+    // generic resolver long predates this flag. A `since` bisected on a symbol the
+    // flag does not depend on is a number that looks rigorous and is not.
+    //
+    // ENG-8395 AC5 left this unset on purpose and said why: an unset `since` makes
+    // the flip-reach estimate report FULL reach for a host-side reader, promising
+    // reach it does not have (the defect caught on ENG-8344 / PR #4014). The
+    // follow-up it was waiting on was never filed, so the flag sat un-armable for
+    // a month and ENG-6088's bounded auto-resume stayed dark fleet-wide — a
+    // customer agent sat breaker-paused 2h08m on 2026-09-05 with the recovery path
+    // silently disabled. That is what an administrative blocker recorded only in a
+    // code comment costs.
+    //
+    // Same method as its precedents: slack-hot-thread-guard = 0.28.339,
+    // slack-scheduled-channel-guard = 0.28.421 (ENG-8149). Setting it now is safe:
+    // `since` is excluded from projectDefinition, so it does not roll
     // FLAGS_SCHEMA_VERSION.
+    //
+    // This does NOT arm the flag. defaultValue stays false and `sensitive` stays
+    // true; arming is a separate, deliberate decision that belongs to ENG-6066.
+    since: "0.28.507"
   },
   {
     key: "host-poll-rate-alarm",
@@ -39671,6 +39712,7 @@ await mcp.connect(new StdioServerTransport());
 var processedIds = /* @__PURE__ */ new Set();
 var lastBusyAckNoticeAt = /* @__PURE__ */ new Map();
 var lastUndeliverableNoticeAt = /* @__PURE__ */ new Map();
+var undeliverableNoticeCount = /* @__PURE__ */ new Map();
 var lastTurnFailedNoticeAt = /* @__PURE__ */ new Map();
 async function postDirectChatNoticeMessage(sessionId, content) {
   try {
@@ -39705,14 +39747,47 @@ function postDirectChatBusyAckNotice(sessionId) {
     }
   })();
 }
+function directChatSessionReplayFacts(sessionId) {
+  if (!DIRECT_CHAT_PENDING_INBOUND_DIR) return null;
+  const found = [];
+  try {
+    clearDirectChatPendingMarkersForSession(
+      DIRECT_CHAT_PENDING_INBOUND_DIR,
+      sessionId,
+      (fullPath) => {
+        try {
+          const marker = JSON.parse(readFileSync11(fullPath, "utf-8"));
+          found.push({ hasPayload: Boolean(marker.payload), discretionary: marker.discretionary });
+        } catch {
+          found.push({ hasPayload: false });
+        }
+      }
+    );
+  } catch {
+    return null;
+  }
+  const facts = combineMarkerReplayFacts(found);
+  return facts ? { enabled: channelReplayEnabled(), ...facts } : null;
+}
 function postDirectChatUndeliverableNotice(sessionId) {
   const now = Date.now();
   if (!shouldPostUndeliverableNotice(lastUndeliverableNoticeAt.get(sessionId), now)) {
     return;
   }
   lastUndeliverableNoticeAt.set(sessionId, now);
+  const text = undeliverableNoticeTextForRepeat({
+    replay: directChatSessionReplayFacts(sessionId) ?? void 0,
+    priorNotices: undeliverableNoticeCount.get(sessionId) ?? 0
+  });
+  if (text === null) {
+    lastUndeliverableNoticeAt.delete(sessionId);
+    return;
+  }
   void (async () => {
-    const ok = await postDirectChatNoticeMessage(sessionId, undeliverableNoticeText());
+    const ok = await postDirectChatNoticeMessage(sessionId, text);
+    if (ok) {
+      undeliverableNoticeCount.set(sessionId, (undeliverableNoticeCount.get(sessionId) ?? 0) + 1);
+    }
     if (!ok) {
       lastUndeliverableNoticeAt.delete(sessionId);
       process.stderr.write(

package/dist/mcp/index.js

@@ -28403,7 +28403,7 @@ var FLAG_REGISTRY = [
   },
   {
     key: "auto-resume-enabled",
-    description: "DO NOT ARM YET (ENG-8395): this flag has no `since` set until a follow-up bisects it from the published agt-cli dist, and an unset `since` makes the flip-reach estimate report FULL reach for a reader that is host-side \u2014 it will tell you a flip reached hosts that have no code to honour it. Arm only once that follow-up has landed. Bounded BLIND auto-resume for circuit-breaker pauses (ENG-6088): at most one unattended resume per trip cycle, fired once the trip driver has been quiet for the configured window and only for provisioning-reload trips (the ENG-7577 class gate \u2014 a crash-class trip still needs an operator). Superseded wherever `resume-reconciler` is on: that path proves the dependency recovered instead of assuming it, and the blind timer is its flag-off fallback. Boolean gate; ships dark. ENG-8395 promoted this from a raw host env var because AGT_AUTO_RESUME_ENABLED lives in the host env file, so its grain is per-HOST \u2014 one flip arms every agent that host runs (agt-aws-1 carries 14), which made the canary-then-fleet rollout its own documentation prescribes impossible. As a registry flag it resolves per org/team/host/AGENT, so one agent can genuinely be cut over first. Read host-side by the manager (agt-cli), not centrally.",
+    description: "Bounded BLIND auto-resume for circuit-breaker pauses (ENG-6088): at most one unattended resume per trip cycle, fired once the trip driver has been quiet for the configured window and only for provisioning-reload trips (the ENG-7577 class gate \u2014 a crash-class trip still needs an operator). Superseded wherever `resume-reconciler` is on: that path proves the dependency recovered instead of assuming it, and the blind timer is its flag-off fallback. Boolean gate; ships dark. ENG-8395 promoted this from a raw host env var because AGT_AUTO_RESUME_ENABLED lives in the host env file, so its grain is per-HOST \u2014 one flip arms every agent that host runs (agt-aws-1 carries 14), which made the canary-then-fleet rollout its own documentation prescribes impossible. As a registry flag it resolves per org/team/host/AGENT, so one agent can genuinely be cut over first. Read host-side by the manager (agt-cli), not centrally.",
     flagType: "boolean",
     // Declared safe value is `false` — identical to today's behaviour on a host
     // that has never set the env var. Fail-safe direction on a flag-DB read
@@ -28420,16 +28420,41 @@ var FLAG_REGISTRY = [
     // An unattended resume is a real availability change — the agent starts
     // spawning sessions again with no human in the loop — so arming it is a
     // deliberate, confirmed write (ADR-0022 sensitive-flag confirm).
-    sensitive: true
-    // ENG-8395 AC5: `since` is deliberately UNSET at merge. It must be bisected
-    // from the PUBLISHED agt-cli dist once the artifact carrying the host-side
-    // reader exists, never guessed — auto-resume.ts is host-side, so an
-    // under-`since` host has no consumer for this flag and a guessed value makes
-    // the flip-reach modal promise reach it does not have (the defect caught on
-    // ENG-8344 / PR #4014). Precedent for the bisect method: slack-hot-thread-guard
-    // = 0.28.339, slack-scheduled-channel-guard = 0.28.421. Setting it later is
-    // safe: `since` is excluded from projectDefinition, so it does not roll
+    sensitive: true,
+    // ENG-10034: bisected from the PUBLISHED agt-cli dists, not from git history —
+    // auto-resume.ts is host-side, so what matters is the first ARTIFACT a host
+    // could be running that contains a consumer for this flag.
+    //
+    // METHOD, so it can be re-run rather than trusted: download each candidate
+    // tarball from the npm registry and grep its contents for the flag-specific
+    // wiring. Binary search over the 22 versions published either side of
+    // ENG-8395's merge gave a clean one-release boundary, eight minutes wide:
+    //
+    //   0.28.506  2026-08-03T22:29:22Z  readAutoResumeEnvOverride ✗  AUTO_RESUME_FLAG_KEY ✗  'auto-resume-enabled' ✗
+    //   0.28.507  2026-08-03T22:37:00Z  readAutoResumeEnvOverride ✓  AUTO_RESUME_FLAG_KEY ✓  'auto-resume-enabled' ✓
+    //
+    // All three symbols are checked rather than one. `getBooleanForAgent` alone
+    // would have been the wrong probe — it is present in BOTH builds, because the
+    // generic resolver long predates this flag. A `since` bisected on a symbol the
+    // flag does not depend on is a number that looks rigorous and is not.
+    //
+    // ENG-8395 AC5 left this unset on purpose and said why: an unset `since` makes
+    // the flip-reach estimate report FULL reach for a host-side reader, promising
+    // reach it does not have (the defect caught on ENG-8344 / PR #4014). The
+    // follow-up it was waiting on was never filed, so the flag sat un-armable for
+    // a month and ENG-6088's bounded auto-resume stayed dark fleet-wide — a
+    // customer agent sat breaker-paused 2h08m on 2026-09-05 with the recovery path
+    // silently disabled. That is what an administrative blocker recorded only in a
+    // code comment costs.
+    //
+    // Same method as its precedents: slack-hot-thread-guard = 0.28.339,
+    // slack-scheduled-channel-guard = 0.28.421 (ENG-8149). Setting it now is safe:
+    // `since` is excluded from projectDefinition, so it does not roll
     // FLAGS_SCHEMA_VERSION.
+    //
+    // This does NOT arm the flag. defaultValue stays false and `sensitive` stays
+    // true; arming is a separate, deliberate decision that belongs to ENG-6066.
+    since: "0.28.507"
   },
   {
     key: "host-poll-rate-alarm",

package/dist/mcp/origami.js

@@ -42356,7 +42356,7 @@ var FLAG_REGISTRY = [
   },
   {
     key: "auto-resume-enabled",
-    description: "DO NOT ARM YET (ENG-8395): this flag has no `since` set until a follow-up bisects it from the published agt-cli dist, and an unset `since` makes the flip-reach estimate report FULL reach for a reader that is host-side \u2014 it will tell you a flip reached hosts that have no code to honour it. Arm only once that follow-up has landed. Bounded BLIND auto-resume for circuit-breaker pauses (ENG-6088): at most one unattended resume per trip cycle, fired once the trip driver has been quiet for the configured window and only for provisioning-reload trips (the ENG-7577 class gate \u2014 a crash-class trip still needs an operator). Superseded wherever `resume-reconciler` is on: that path proves the dependency recovered instead of assuming it, and the blind timer is its flag-off fallback. Boolean gate; ships dark. ENG-8395 promoted this from a raw host env var because AGT_AUTO_RESUME_ENABLED lives in the host env file, so its grain is per-HOST \u2014 one flip arms every agent that host runs (agt-aws-1 carries 14), which made the canary-then-fleet rollout its own documentation prescribes impossible. As a registry flag it resolves per org/team/host/AGENT, so one agent can genuinely be cut over first. Read host-side by the manager (agt-cli), not centrally.",
+    description: "Bounded BLIND auto-resume for circuit-breaker pauses (ENG-6088): at most one unattended resume per trip cycle, fired once the trip driver has been quiet for the configured window and only for provisioning-reload trips (the ENG-7577 class gate \u2014 a crash-class trip still needs an operator). Superseded wherever `resume-reconciler` is on: that path proves the dependency recovered instead of assuming it, and the blind timer is its flag-off fallback. Boolean gate; ships dark. ENG-8395 promoted this from a raw host env var because AGT_AUTO_RESUME_ENABLED lives in the host env file, so its grain is per-HOST \u2014 one flip arms every agent that host runs (agt-aws-1 carries 14), which made the canary-then-fleet rollout its own documentation prescribes impossible. As a registry flag it resolves per org/team/host/AGENT, so one agent can genuinely be cut over first. Read host-side by the manager (agt-cli), not centrally.",
     flagType: "boolean",
     // Declared safe value is `false` — identical to today's behaviour on a host
     // that has never set the env var. Fail-safe direction on a flag-DB read
@@ -42373,16 +42373,41 @@ var FLAG_REGISTRY = [
     // An unattended resume is a real availability change — the agent starts
     // spawning sessions again with no human in the loop — so arming it is a
     // deliberate, confirmed write (ADR-0022 sensitive-flag confirm).
-    sensitive: true
-    // ENG-8395 AC5: `since` is deliberately UNSET at merge. It must be bisected
-    // from the PUBLISHED agt-cli dist once the artifact carrying the host-side
-    // reader exists, never guessed — auto-resume.ts is host-side, so an
-    // under-`since` host has no consumer for this flag and a guessed value makes
-    // the flip-reach modal promise reach it does not have (the defect caught on
-    // ENG-8344 / PR #4014). Precedent for the bisect method: slack-hot-thread-guard
-    // = 0.28.339, slack-scheduled-channel-guard = 0.28.421. Setting it later is
-    // safe: `since` is excluded from projectDefinition, so it does not roll
+    sensitive: true,
+    // ENG-10034: bisected from the PUBLISHED agt-cli dists, not from git history —
+    // auto-resume.ts is host-side, so what matters is the first ARTIFACT a host
+    // could be running that contains a consumer for this flag.
+    //
+    // METHOD, so it can be re-run rather than trusted: download each candidate
+    // tarball from the npm registry and grep its contents for the flag-specific
+    // wiring. Binary search over the 22 versions published either side of
+    // ENG-8395's merge gave a clean one-release boundary, eight minutes wide:
+    //
+    //   0.28.506  2026-08-03T22:29:22Z  readAutoResumeEnvOverride ✗  AUTO_RESUME_FLAG_KEY ✗  'auto-resume-enabled' ✗
+    //   0.28.507  2026-08-03T22:37:00Z  readAutoResumeEnvOverride ✓  AUTO_RESUME_FLAG_KEY ✓  'auto-resume-enabled' ✓
+    //
+    // All three symbols are checked rather than one. `getBooleanForAgent` alone
+    // would have been the wrong probe — it is present in BOTH builds, because the
+    // generic resolver long predates this flag. A `since` bisected on a symbol the
+    // flag does not depend on is a number that looks rigorous and is not.
+    //
+    // ENG-8395 AC5 left this unset on purpose and said why: an unset `since` makes
+    // the flip-reach estimate report FULL reach for a host-side reader, promising
+    // reach it does not have (the defect caught on ENG-8344 / PR #4014). The
+    // follow-up it was waiting on was never filed, so the flag sat un-armable for
+    // a month and ENG-6088's bounded auto-resume stayed dark fleet-wide — a
+    // customer agent sat breaker-paused 2h08m on 2026-09-05 with the recovery path
+    // silently disabled. That is what an administrative blocker recorded only in a
+    // code comment costs.
+    //
+    // Same method as its precedents: slack-hot-thread-guard = 0.28.339,
+    // slack-scheduled-channel-guard = 0.28.421 (ENG-8149). Setting it now is safe:
+    // `since` is excluded from projectDefinition, so it does not roll
     // FLAGS_SCHEMA_VERSION.
+    //
+    // This does NOT arm the flag. defaultValue stays false and `sensitive` stays
+    // true; arming is a separate, deliberate decision that belongs to ENG-6066.
+    since: "0.28.507"
   },
   {
     key: "host-poll-rate-alarm",