@integrity-labs/agt-cli 0.28.717 → 0.28.718

package/dist/{claude-pair-runtime-KWBCSCZO.js → claude-pair-runtime-PDG74SAT.js}

@@ -100,7 +100,7 @@ async function spawnPairSession(session) {
     return { ok: true };
   } catch {
   }
-  const { resolveClaudeBinary } = await import("./persistent-session-GB3D5OIA.js");
+  const { resolveClaudeBinary } = await import("./persistent-session-2AWYIUCX.js");
   const claudeBin = resolveClaudeBinary();
   const pairEnv = {
     ...process.env,
@@ -373,4 +373,4 @@ export {
   startClaudePair,
   submitClaudePairCode
 };
-//# sourceMappingURL=claude-pair-runtime-KWBCSCZO.js.map
+//# sourceMappingURL=claude-pair-runtime-PDG74SAT.js.map

package/dist/lib/manager-worker.js

@@ -57,7 +57,7 @@ import {
   safeWriteJsonAtomic,
   setConfigHash,
   tripClass
-} from "../chunk-XSBRGOGL.js";
+} from "../chunk-EEZHSNQ2.js";
 import {
   getProjectDir as getProjectDir2,
   getReadyTasks,
@@ -115,7 +115,7 @@ import {
   takeZombieDetection,
   toOpencodeModel,
   writeEgressAllowlist
-} from "../chunk-JN4D7534.js";
+} from "../chunk-FNALM4HG.js";
 import {
   ACCOUNT_ENFORCEMENT_MARKER_FILENAME,
   AnchorSessionClient,
@@ -200,7 +200,7 @@ import {
   subagentActivityAgeSeconds,
   sumTranscriptUsageInWindow,
   transcriptActivityAgeSeconds
-} from "../chunk-KVR5KNED.js";
+} from "../chunk-7LH3SVAD.js";
 import {
   reapOrphanChannelMcps
 } from "../chunk-XWVM4KPK.js";
@@ -10641,13 +10641,58 @@ function refreshSlackRestartContextHints(codeNames, opts = {}) {
 
 // src/lib/persistent-session-stuck-tracker.ts
 var DEFAULT_STUCK_WARN_THRESHOLD = 3;
+var DEFAULT_SPAWN_LOOP_WARN_THRESHOLD = 3;
 var PersistentSessionStuckTracker = class {
   counts = /* @__PURE__ */ new Map();
   warned = /* @__PURE__ */ new Set();
   threshold;
-  constructor(threshold = DEFAULT_STUCK_WARN_THRESHOLD) {
+  /**
+   * ENG-9614 — the second streak, held separately because it answers a
+   * DIFFERENT question and must not inherit the first one's reset rule.
+   *
+   * `counts` above resets whenever `spawnAttempted` is true, which is correct
+   * for its own question (ENG-5116/Vigil: unhealthy and nobody even TRIED).
+   * But it makes the tracker structurally blind to the inverse: the scratch/tmp
+   * collision attempted a spawn on EVERY tick, so `counts` was deleted on every
+   * tick, `shouldWarn` could never fire, and an agent looped unrecoverably for
+   * 3m48s without this class emitting a single line.
+   *
+   * Folding both into one counter was considered and rejected - a shared
+   * counter would have to pick one reset rule, and either choice makes the
+   * other question unanswerable. Two questions, two counters.
+   */
+  spawnLoopCounts = /* @__PURE__ */ new Map();
+  spawnLoopWarned = /* @__PURE__ */ new Set();
+  spawnLoopThreshold;
+  constructor(threshold = DEFAULT_STUCK_WARN_THRESHOLD, spawnLoopThreshold = DEFAULT_SPAWN_LOOP_WARN_THRESHOLD) {
     if (threshold < 1) throw new Error("stuck-tracker threshold must be >= 1");
+    if (spawnLoopThreshold < 1) throw new Error("stuck-tracker spawn-loop threshold must be >= 1");
     this.threshold = threshold;
+    this.spawnLoopThreshold = spawnLoopThreshold;
+  }
+  /**
+   * Advance the spawn-loop streak for one tick.
+   *
+   * Counts only when a spawn was attempted AND the session is still unhealthy.
+   * Health alone resets it - an agent that came up is not looping, however it
+   * got there. A tick with no spawn attempt neither counts nor resets: that is
+   * the OTHER failure, already owned by `counts`, and letting it clear this
+   * streak would let the two shapes alternate and hide each other.
+   */
+  recordSpawnLoop(codeName, sessionHealthy, spawnAttempted) {
+    if (sessionHealthy) {
+      this.spawnLoopCounts.delete(codeName);
+      this.spawnLoopWarned.delete(codeName);
+      return { ticks: 0, shouldWarn: false };
+    }
+    if (!spawnAttempted) {
+      return { ticks: this.spawnLoopCounts.get(codeName) ?? 0, shouldWarn: false };
+    }
+    const next = (this.spawnLoopCounts.get(codeName) ?? 0) + 1;
+    this.spawnLoopCounts.set(codeName, next);
+    const crossed = next >= this.spawnLoopThreshold && !this.spawnLoopWarned.has(codeName);
+    if (crossed) this.spawnLoopWarned.add(codeName);
+    return { ticks: next, shouldWarn: crossed };
   }
   /**
    * Record one tick for one agent. Returns whether the caller should log
@@ -10658,6 +10703,7 @@ var PersistentSessionStuckTracker = class {
    */
   record(input) {
     const { codeName, sessionHealthy, spawnAttempted } = input;
+    const spawnLoop = this.recordSpawnLoop(codeName, sessionHealthy, spawnAttempted);
     const wasStuck = (this.counts.get(codeName) ?? 0) >= this.threshold;
     if (sessionHealthy || spawnAttempted) {
       this.counts.delete(codeName);
@@ -10665,7 +10711,9 @@ var PersistentSessionStuckTracker = class {
       return {
         consecutiveTicks: 0,
         shouldWarn: false,
-        recovered: wasStuck && wasWarned
+        recovered: wasStuck && wasWarned,
+        spawnLoopTicks: spawnLoop.ticks,
+        shouldWarnSpawnLoop: spawnLoop.shouldWarn
       };
     }
     const next = (this.counts.get(codeName) ?? 0) + 1;
@@ -10675,17 +10723,25 @@ var PersistentSessionStuckTracker = class {
     return {
       consecutiveTicks: next,
       shouldWarn: crossed,
-      recovered: false
+      recovered: false,
+      spawnLoopTicks: spawnLoop.ticks,
+      shouldWarnSpawnLoop: spawnLoop.shouldWarn
     };
   }
   /** Test/debug helper: peek at an agent's current streak. */
   peek(codeName) {
     return this.counts.get(codeName) ?? 0;
   }
+  /** Test/debug helper: peek at an agent's current spawn-loop streak. */
+  peekSpawnLoop(codeName) {
+    return this.spawnLoopCounts.get(codeName) ?? 0;
+  }
   /** Test helper: clear all state. */
   reset() {
     this.counts.clear();
     this.warned.clear();
+    this.spawnLoopCounts.clear();
+    this.spawnLoopWarned.clear();
   }
 };
 
@@ -12001,7 +12057,7 @@ 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-TDZDQXBU.js");
+  const { newestPendingInboundActivityMtimeMs } = await import("../responsiveness-probe-JQS4WA2L.js");
   const newest = newestPendingInboundActivityMtimeMs(dirname10(paneLogPath(codeName)));
   if (newest === null) return null;
   return Math.max(0, Math.floor((Date.now() - newest) / 1e3));
@@ -12649,7 +12705,7 @@ var agentRestartTimezoneInputs = /* @__PURE__ */ new Map();
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
 var lastResponsivenessProbeAt = 0;
-var agtCliVersion = true ? "0.28.717" : "dev";
+var agtCliVersion = true ? "0.28.718" : "dev";
 function resolveBrewPath(execFileSync3) {
   try {
     const out = execFileSync3("which", ["brew"], { timeout: 5e3 }).toString().trim();
@@ -14176,7 +14232,7 @@ function flushRestartedAgentDiagnostics(hostId, codeNames) {
   if (codeNames.length === 0) return;
   void (async () => {
     try {
-      const { collectDiagnostics } = await import("../persistent-session-GB3D5OIA.js");
+      const { collectDiagnostics } = await import("../persistent-session-2AWYIUCX.js");
       await api.post("/host/heartbeat", {
         host_id: hostId,
         agent_diagnostics: collectDiagnostics(codeNames, quarantineEntriesFor, claudeMdSizeFor, spawnOutcomeForDiagnostics, pidPressureFor, forwardedToolsFor)
@@ -14294,7 +14350,7 @@ async function pollCycleInner() {
     }
     try {
       const { detectHostSecurity } = await import("../host-security-6PDFG7F5.js");
-      const { collectDiagnostics } = await import("../persistent-session-GB3D5OIA.js");
+      const { collectDiagnostics } = await import("../persistent-session-2AWYIUCX.js");
       const diagCodeNames = [...agentState.persistentSessionAgents];
       const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames, quarantineEntriesFor, claudeMdSizeFor, spawnOutcomeForDiagnostics, pidPressureFor, forwardedToolsFor) : void 0;
       let tailscaleHostname;
@@ -14463,7 +14519,7 @@ async function pollCycleInner() {
         collectPanelessActivityProbes,
         getResponsivenessIntervalMs,
         occupancyQualificationClassifications
-      } = await import("../responsiveness-probe-TDZDQXBU.js");
+      } = await import("../responsiveness-probe-JQS4WA2L.js");
       const probeIntervalMs = getResponsivenessIntervalMs();
       if (now - lastResponsivenessProbeAt > probeIntervalMs) {
         const probeCodeNames = [...agentState.persistentSessionAgents];
@@ -14573,7 +14629,7 @@ async function pollCycleInner() {
           collectResponsivenessProbes,
           livePendingInboundOldestAgeSeconds,
           parkPendingInbound
-        } = await import("../responsiveness-probe-TDZDQXBU.js");
+        } = await import("../responsiveness-probe-JQS4WA2L.js");
         const { getProjectDir: wedgeProjectDir } = await import("../scheduler-engine-NDP36U7O.js");
         const wedgeNow = /* @__PURE__ */ new Date();
         const liveAgents = agentState.persistentSessionAgents;
@@ -14729,7 +14785,7 @@ async function pollCycleInner() {
     }
     try {
       const { scrapeMcpFailedBannerCount } = await import("../pane-mcp-banner-scraper-JA437JIB.js");
-      const { probeSessionAuth } = await import("../session-auth-dead-CPY4UZI7.js");
+      const { probeSessionAuth } = await import("../session-auth-dead-VBMZW4GG.js");
       const observations = [];
       const pendingCacheCommits = [];
       const modelApiErrorReportingOn = hostFlagStore().getBoolean("model-api-error-reporting");
@@ -17150,6 +17206,12 @@ async function processAgent(agent, agentStates, managedToolkits) {
     } else if (stuck.recovered) {
       log(`[persistent-session-stuck] agent=${agent.code_name} recovered from stuck state`);
     }
+    if (stuck.shouldWarnSpawnLoop) {
+      const failure = getLastFailureContext(agent.code_name);
+      log(
+        `[persistent-session-spawn-loop] WARN: agent=${agent.code_name} respawned ${stuck.spawnLoopTicks} consecutive ticks and is still unhealthy (last_decision=${psResult.decision} signature=${failure.signature} restart_count=${failure.restartCount}) \u2014 the spawn is failing in a way a retry cannot fix; see the pane log tail for the cause (ENG-9614)`
+      );
+    }
     if (agentFw === "claude-code") {
       try {
         checkMcpConfigDriftAndScheduleRestart(agent.code_name, getProjectDir(agent.code_name));
@@ -18554,7 +18616,7 @@ 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-GB3D5OIA.js");
+        const { collectDiagnostics } = await import("../persistent-session-2AWYIUCX.js");
         await api.post("/host/heartbeat", {
           host_id: hostId,
           agent_diagnostics: collectDiagnostics([codeName], quarantineEntriesFor, claudeMdSizeFor, spawnOutcomeForDiagnostics, pidPressureFor, forwardedToolsFor)
@@ -18605,7 +18667,7 @@ async function respawnAgentAfterMcpStop(codeName, reason) {
     }
     try {
       const hostId = await getHostId();
-      const { collectDiagnostics } = await import("../persistent-session-GB3D5OIA.js");
+      const { collectDiagnostics } = await import("../persistent-session-2AWYIUCX.js");
       await api.post("/host/heartbeat", {
         host_id: hostId,
         agent_diagnostics: collectDiagnostics([codeName], quarantineEntriesFor, claudeMdSizeFor, spawnOutcomeForDiagnostics, pidPressureFor, forwardedToolsFor)
@@ -19237,7 +19299,7 @@ async function processClaudePairSessions(agents) {
     killPairSession,
     pairTmuxSession,
     finalizeClaudePairOnboarding
-  } = await import("../claude-pair-runtime-KWBCSCZO.js");
+  } = await import("../claude-pair-runtime-PDG74SAT.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));