@integrity-labs/agt-cli 0.27.113 → 0.27.115

package/dist/lib/manager-worker.js

@@ -17,7 +17,7 @@ import {
   provisionStopHook,
   requireHost,
   safeWriteJsonAtomic
-} from "../chunk-UHG6DSOX.js";
+} from "../chunk-EOMWSV4B.js";
 import {
   getProjectDir as getProjectDir2,
   getReadyTasks,
@@ -47,6 +47,7 @@ import {
   readPaneLogTail,
   resetRestartCount,
   resolveClaudeBinary,
+  rotateSessionForWedge,
   sanitizeMcpJson,
   sendToAgent,
   sessionTranscriptDir,
@@ -55,7 +56,7 @@ import {
   stopPersistentSession,
   takeWatchdogGiveUpCount,
   takeZombieDetection
-} from "../chunk-RN3OMHW7.js";
+} from "../chunk-IUN5LBTQ.js";
 import {
   KANBAN_CHECK_COMMAND,
   SUPPRESS_SENTINEL,
@@ -3076,6 +3077,53 @@ function clearAgentState(agentId, codeName) {
   return channelCacheMutated;
 }
 
+// src/lib/wedge-detection.ts
+var DEFAULTS = {
+  inboundWaitSeconds: 120,
+  inboundHardWaitSeconds: 300,
+  paneStaleSeconds: 120,
+  minCycles: 3
+};
+function parseMode(raw) {
+  const v = (raw ?? "").trim().toLowerCase();
+  return v === "shadow" || v === "enforce" ? v : "off";
+}
+function parsePositiveInt(raw, fallback, floor) {
+  const n = raw ? Number.parseInt(raw, 10) : NaN;
+  return Number.isInteger(n) && n >= floor ? n : fallback;
+}
+function resolveWedgeConfig(env = process.env) {
+  const inboundWaitSeconds = parsePositiveInt(
+    env.AGT_WEDGE_INBOUND_WAIT_SECONDS,
+    DEFAULTS.inboundWaitSeconds,
+    30
+  );
+  const inboundHardWaitSeconds = Math.max(
+    inboundWaitSeconds,
+    parsePositiveInt(env.AGT_WEDGE_INBOUND_HARD_WAIT_SECONDS, DEFAULTS.inboundHardWaitSeconds, 30)
+  );
+  return {
+    mode: parseMode(env.AGT_WEDGE_RESTART_MODE),
+    inboundWaitSeconds,
+    inboundHardWaitSeconds,
+    paneStaleSeconds: parsePositiveInt(env.AGT_WEDGE_PANE_STALE_SECONDS, DEFAULTS.paneStaleSeconds, 30),
+    minCycles: parsePositiveInt(env.AGT_WEDGE_MIN_CYCLES, DEFAULTS.minCycles, 2)
+  };
+}
+function isWedgeCandidateCycle(signals, config2) {
+  const inboundAge = signals.pendingInboundOldestAgeSeconds;
+  if (inboundAge === null) return false;
+  if (inboundAge < config2.inboundWaitSeconds) return false;
+  if (inboundAge >= config2.inboundHardWaitSeconds) return true;
+  const paneAge = signals.paneActivityAgeSeconds;
+  const paneAdvancing = paneAge !== null && paneAge < config2.paneStaleSeconds;
+  return !paneAdvancing;
+}
+function decideWedgeRestart(input, config2) {
+  if (!isWedgeCandidateCycle(input, config2)) return "none";
+  return input.consecutiveWedgeCycles >= config2.minCycles ? "wedged" : "none";
+}
+
 // src/lib/restart-flags.ts
 import { existsSync as existsSync4, mkdirSync as mkdirSync3, readdirSync as readdirSync3, readFileSync as readFileSync8, renameSync as renameSync2, rmSync, writeFileSync as writeFileSync3 } from "fs";
 import { homedir as homedir3 } from "os";
@@ -3789,6 +3837,7 @@ var autoResumeInFlight = /* @__PURE__ */ new Set();
 var AUTO_RESUME_SELF_WINDOW_MS = 12e4;
 var autoResumeLoggedSkips = /* @__PURE__ */ new Map();
 var autoResumeStandDowns = /* @__PURE__ */ new Set();
+var killPausedCodeNames = /* @__PURE__ */ new Set();
 function maybeAutoResume(agent) {
   const codeName = agent.code_name;
   if (autoResumeInFlight.has(codeName)) return;
@@ -3875,6 +3924,7 @@ function cancelPendingSessionRestart(codeName) {
 }
 var RESTART_DEFER_RECHECK_MS = 6e4;
 var lastInboundMs = /* @__PURE__ */ new Map();
+var consecutiveWedgeCycles = /* @__PURE__ */ new Map();
 function noteInbound(codeName) {
   lastInboundMs.set(codeName, Date.now());
 }
@@ -4173,7 +4223,7 @@ var cachedMaintenanceWindow = null;
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
 var lastResponsivenessProbeAt = 0;
-var agtCliVersion = true ? "0.27.113" : "dev";
+var agtCliVersion = true ? "0.27.115" : "dev";
 function resolveBrewPath(execFileSync4) {
   try {
     const out = execFileSync4("which", ["brew"], { timeout: 5e3 }).toString().trim();
@@ -5366,7 +5416,7 @@ async function pollCycle() {
     }
     try {
       const { detectHostSecurity } = await import("../host-security-6PDFG7F5.js");
-      const { collectDiagnostics } = await import("../persistent-session-E6YLH6TX.js");
+      const { collectDiagnostics } = await import("../persistent-session-PIR45QFL.js");
       const diagCodeNames = [...agentState.persistentSessionAgents];
       const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames) : void 0;
       let tailscaleHostname;
@@ -5453,12 +5503,12 @@ async function pollCycle() {
       const {
         collectResponsivenessProbes,
         getResponsivenessIntervalMs
-      } = await import("../responsiveness-probe-TDHX6JKG.js");
+      } = await import("../responsiveness-probe-FLQZ7E5H.js");
       const probeIntervalMs = getResponsivenessIntervalMs();
       if (now - lastResponsivenessProbeAt > probeIntervalMs) {
         const probeCodeNames = [...agentState.persistentSessionAgents];
         if (probeCodeNames.length > 0) {
-          const { takeAcpxExecFailureCount, creditAcpxExecFailureCount } = await import("../persistent-session-E6YLH6TX.js");
+          const { takeAcpxExecFailureCount, creditAcpxExecFailureCount } = await import("../persistent-session-PIR45QFL.js");
           const drainedGiveUps = /* @__PURE__ */ new Map();
           const drainedAcpxFailures = /* @__PURE__ */ new Map();
           const probes = collectResponsivenessProbes(probeCodeNames).map((p) => {
@@ -5485,6 +5535,58 @@ async function pollCycle() {
     } catch (err) {
       log(`[responsiveness-probe] collection failed: ${err.message}`);
     }
+    try {
+      const wedgeConfig = resolveWedgeConfig();
+      if (wedgeConfig.mode !== "off") {
+        const { collectResponsivenessProbes } = await import("../responsiveness-probe-FLQZ7E5H.js");
+        const liveAgents = agentState.persistentSessionAgents;
+        for (const tracked of consecutiveWedgeCycles.keys()) {
+          if (!liveAgents.has(tracked)) consecutiveWedgeCycles.delete(tracked);
+        }
+        for (const probe of collectResponsivenessProbes([...liveAgents])) {
+          const codeName = probe.code_name;
+          if (!isSessionHealthy(codeName)) {
+            consecutiveWedgeCycles.delete(codeName);
+            continue;
+          }
+          const signals = {
+            paneActivityAgeSeconds: probe.pane_activity_age_seconds,
+            pendingInboundOldestAgeSeconds: probe.pending_inbound_oldest_age_seconds ?? null
+          };
+          if (!isWedgeCandidateCycle(signals, wedgeConfig)) {
+            consecutiveWedgeCycles.delete(codeName);
+            continue;
+          }
+          const streak = (consecutiveWedgeCycles.get(codeName) ?? 0) + 1;
+          consecutiveWedgeCycles.set(codeName, streak);
+          if (decideWedgeRestart({ ...signals, consecutiveWedgeCycles: streak }, wedgeConfig) !== "wedged") {
+            continue;
+          }
+          const detail = `agent=${codeName} paneAge=${signals.paneActivityAgeSeconds}s inboundAge=${signals.pendingInboundOldestAgeSeconds}s cycles=${streak}`;
+          if (wedgeConfig.mode === "shadow") {
+            if (streak === wedgeConfig.minCycles) {
+              log(`[wedge] SHADOW would force-fresh respawn ${detail}`);
+            }
+            continue;
+          }
+          try {
+            const newId = rotateSessionForWedge(codeName);
+            try {
+              const { execSync: es } = await import("child_process");
+              es(`tmux kill-session -t agt-${codeName} 2>/dev/null`, { stdio: "ignore" });
+            } catch {
+            }
+            stopPersistentSessionAndForgetMcpBaseline(codeName);
+            consecutiveWedgeCycles.delete(codeName);
+            log(`[wedge] forced fresh respawn ${detail} \u2192 new session ${newId} (transcript preserved)`);
+          } catch (err) {
+            log(`[wedge] force-fresh respawn failed for ${codeName}: ${err.message}`);
+          }
+        }
+      }
+    } catch (err) {
+      log(`[wedge] detection failed: ${err.message}`);
+    }
     try {
       const { scrapeMcpFailedBannerCount } = await import("../pane-mcp-banner-scraper-JA437JIB.js");
       const observations = [];
@@ -5818,6 +5920,16 @@ async function processAgent(agent, agentStates) {
     if (previousKnownStatus !== agent.status) {
       log(`Agent '${agent.code_name}' is ${agent.status}, skipping provisioning`);
     }
+    if (agent.kill_switch) {
+      killPausedCodeNames.add(agent.code_name);
+      if (previousKnownStatus !== agent.status) {
+        log(
+          `[kill-switch] '${agent.code_name}' halted by ${agent.kill_switch.scope ?? "unknown"} kill switch` + (agent.kill_switch.reason ? ` (reason: ${agent.kill_switch.reason})` : "") + ` source=${agent.kill_switch.source ?? "unknown"}`
+        );
+      }
+    } else if (killPausedCodeNames.delete(agent.code_name)) {
+      log(`[kill-switch] '${agent.code_name}' kill switch cleared while still ${agent.status} (no resume \u2014 paused for another reason)`);
+    }
     await stopGatewayIfRunning(agent.code_name, adapter);
     stopPersistentSessionAndForgetMcpBaseline(agent.code_name);
     try {
@@ -5826,7 +5938,7 @@ async function processAgent(agent, agentStates) {
       log(`Killed tmux session for paused agent '${agent.code_name}'`);
     } catch {
     }
-    if (agent.status === "paused") {
+    if (agent.status === "paused" && !agent.kill_switch) {
       maybeAutoResume(agent);
     }
     agentStates.push({
@@ -5932,6 +6044,16 @@ async function processAgent(agent, agentStates) {
         log(`[auto-resume] Cleared auto-resume marker for '${agent.code_name}' on operator resume \u2014 credit re-armed (ENG-6088)`);
       }
     }
+    if (previousStatus === "paused" && agent.status === "active" && killPausedCodeNames.delete(agent.code_name)) {
+      log(`[kill-switch] '${agent.code_name}' kill switch cleared \u2014 resuming session`);
+      void injectMessage(
+        agent.code_name,
+        "system",
+        "The kill switch that halted you has been cleared. Your session is resuming.",
+        { task_name: "kill-switch-resume" },
+        log
+      ).catch(() => false);
+    }
     let channelCacheMutated = false;
     for (const key of agentState.knownChannelConfigHashes.keys()) {
       if (key.startsWith(`${agent.agent_id}:`)) {
@@ -9834,7 +9956,7 @@ async function processClaudePairSessions(agents) {
     killPairSession,
     pairTmuxSession,
     finalizeClaudePairOnboarding
-  } = await import("../claude-pair-runtime-KVJ4U436.js");
+  } = await import("../claude-pair-runtime-UX7GWW3O.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));