@integrity-labs/agt-cli 0.28.637 → 0.28.638

package/dist/bin/agt.js

@@ -40,10 +40,10 @@ import {
   success,
   table,
   warn
-} from "../chunk-3JBA6RYG.js";
+} from "../chunk-LCKKFDCR.js";
 import {
   readDirectChatSessionState
-} from "../chunk-W4GVD3KK.js";
+} from "../chunk-MRQ2GRTN.js";
 import {
   AnchorSessionClient,
   CHANNEL_REGISTRY,
@@ -73,7 +73,7 @@ import {
   requiredMcpWildcard,
   resolveChannels,
   serializeManifestForSlackCli
-} from "../chunk-JT2JQUK3.js";
+} from "../chunk-MAZUBHNT.js";
 import "../chunk-XWVM4KPK.js";
 
 // src/bin/agt.ts
@@ -4909,7 +4909,7 @@ import { execFileSync, execSync } from "child_process";
 import { existsSync as existsSync10, realpathSync as realpathSync2 } from "fs";
 import chalk18 from "chalk";
 import ora16 from "ora";
-var cliVersion = true ? "0.28.637" : "dev";
+var cliVersion = true ? "0.28.638" : "dev";
 async function fetchLatestVersion() {
   const host2 = getHost();
   if (!host2) return null;
@@ -6089,7 +6089,7 @@ function handleError(err) {
 }
 
 // src/bin/agt.ts
-var cliVersion2 = true ? "0.28.637" : "dev";
+var cliVersion2 = true ? "0.28.638" : "dev";
 var program = new Command();
 program.name("agt").description("Augmented CLI \u2014 agent provisioning and management").version(cliVersion2).option("--json", "Emit machine-readable JSON output (suppress spinners and colors)").option("--skip-update-check", "Skip the automatic update check on startup");
 program.hook("preAction", async (thisCommand, actionCommand) => {

package/dist/{chunk-3JBA6RYG.js → chunk-LCKKFDCR.js}

@@ -6,8 +6,9 @@ import {
   buildRemoteMcpEntry,
   expandTemplateVars,
   isolationMode,
+  log,
   parseEnvIntegrations
-} from "./chunk-W4GVD3KK.js";
+} from "./chunk-MRQ2GRTN.js";
 import {
   BIND_FAILURE_QUARANTINE_THRESHOLD,
   INTEGRATIONS_SECTION_END,
@@ -54,7 +55,7 @@ import {
   resolveConnectivityProbe,
   worseConnectivityOutcome,
   wrapScheduledTaskPrompt
-} from "./chunk-JT2JQUK3.js";
+} from "./chunk-MAZUBHNT.js";
 import {
   parsePsRows
 } from "./chunk-XWVM4KPK.js";
@@ -1537,7 +1538,7 @@ function getAgentDir(codeName) {
   return resolveRealAgentPath(join2(getHomeDir(), ".augmented", codeName));
 }
 var migratedCodeNames = /* @__PURE__ */ new Set();
-function migrateLegacyClaudecodeDir(codeName, log) {
+function migrateLegacyClaudecodeDir(codeName, log2) {
   assertValidCodeName(codeName);
   if (migratedCodeNames.has(codeName))
     return;
@@ -1548,7 +1549,7 @@ function migrateLegacyClaudecodeDir(codeName, log) {
   }
   const newRoot = getAgentDir(codeName);
   const emit = (msg) => {
-    log?.(msg);
+    log2?.(msg);
   };
   try {
     const walkAndMigrate = (srcDir, destDir) => {
@@ -6072,7 +6073,7 @@ function exchangeFailureKind(err) {
 }
 
 // src/lib/api-client.ts
-var agtCliVersion = true ? "0.28.637" : "dev";
+var agtCliVersion = true ? "0.28.638" : "dev";
 var lastConfigHash = null;
 function setConfigHash(hash) {
   lastConfigHash = hash && hash.length > 0 ? hash : null;
@@ -6886,7 +6887,7 @@ function findMcpChildrenForAgent(args) {
   return matched;
 }
 function reapStaleMcpChildren(args) {
-  const { log, codeName, serverKeys, mcpJson, graceMs = 5e3, isolated = false } = args;
+  const { log: log2, codeName, serverKeys, mcpJson, graceMs = 5e3, isolated = false } = args;
   if (serverKeys.length === 0) return [];
   const runPs = args.runPs ?? (isolated ? () => execFileSync("docker", ["exec", `agt-${codeName}`, "ps", "-eo", "pid,ppid,args"], {
     encoding: "utf-8",
@@ -6929,7 +6930,7 @@ function reapStaleMcpChildren(args) {
   try {
     psOutput = runPs();
   } catch (err) {
-    log(`[stale-mcp-reaper] ps invocation failed for '${codeName}': ${err.message} \u2014 skipping reap`);
+    log2(`[stale-mcp-reaper] ps invocation failed for '${codeName}': ${err.message} \u2014 skipping reap`);
     return [];
   }
   const rows = parsePsRows(psOutput);
@@ -6941,7 +6942,7 @@ function reapStaleMcpChildren(args) {
     const pkgMatch = argv.match(/(@[a-z0-9_-]+\/[a-z0-9_-]+|[a-z0-9_-]+-mcp-server|[a-z0-9_-]+-mcp\b)/i);
     return pkgMatch ? `${pkgMatch[1]} (pid ${pid})` : `pid ${pid}`;
   };
-  log(
+  log2(
     `[stale-mcp-reaper] '${codeName}': rotating ${targets.length} stale MCP child(ren) for [${serverKeys.join(", ")}]: ${targets.map(describe).join(", ")}`
   );
   for (const pid of targets) {
@@ -6963,7 +6964,7 @@ function reapStaleMcpChildren(args) {
       try {
         freshPsOutput = runPs();
       } catch (err) {
-        log(`[stale-mcp-reaper] '${codeName}': fresh ps for SIGKILL re-verify failed: ${err.message} \u2014 skipping SIGKILL pass`);
+        log2(`[stale-mcp-reaper] '${codeName}': fresh ps for SIGKILL re-verify failed: ${err.message} \u2014 skipping SIGKILL pass`);
         return;
       }
       const stillOwned = new Set(
@@ -6977,14 +6978,14 @@ function reapStaleMcpChildren(args) {
       );
       const stragglers = targets.filter((pid) => isAlive(pid) && stillOwned.has(pid));
       if (stragglers.length === 0) return;
-      log(
+      log2(
         `[stale-mcp-reaper] '${codeName}': ${stragglers.length} child(ren) survived SIGTERM; sending SIGKILL: ${stragglers.map(describe).join(", ")}`
       );
       for (const pid of stragglers) {
         killProcess(pid, "SIGKILL");
       }
     } catch (err) {
-      log(`[stale-mcp-reaper] '${codeName}': error in SIGKILL pass: ${err.message}`);
+      log2(`[stale-mcp-reaper] '${codeName}': error in SIGKILL pass: ${err.message}`);
     }
   }, graceMs).unref();
   return targets;
@@ -7053,7 +7054,7 @@ function findMissingMcpServers(args) {
 }
 function reapMissingMcpSessions(args) {
   const {
-    log,
+    log: log2,
     codeName,
     mcpJson,
     sessionStartedAt,
@@ -7109,7 +7110,7 @@ function reapMissingMcpSessions(args) {
   try {
     psOutput = runPs();
   } catch (err) {
-    log(`[mcp-presence-reaper] ps invocation failed for '${codeName}': ${err.message} \u2014 skipping`);
+    log2(`[mcp-presence-reaper] ps invocation failed for '${codeName}': ${err.message} \u2014 skipping`);
     return { missing: [], restarted: false, reason: "healthy" };
   }
   const rows = parsePsRows(psOutput);
@@ -7125,7 +7126,7 @@ function reapMissingMcpSessions(args) {
     try {
       onRotationRespawnOutcome(codeName, key, outcome);
     } catch (err) {
-      log(
+      log2(
         `[mcp-presence-reaper] onRotationRespawnOutcome callback threw for '${codeName}:${key}' (suppressed; reaper continues): ${err.message}`
       );
     }
@@ -7152,7 +7153,7 @@ function reapMissingMcpSessions(args) {
       escalated: false
     };
     if (priorStall && priorStall !== stall) {
-      log(
+      log2(
         `[mcp-presence-reaper] '${codeName}:${key}' discarding stall state from a previous rotation (rotation=${priorStall.rotationRecordedAt}, probes=${priorStall.probeAttempts}, escalated=${priorStall.escalated}) \u2014 this is a NEW rotation (rotation=${rotationRecordedAt}) and gets a full prompt-probe budget (ENG-8225)`
       );
     }
@@ -7170,14 +7171,14 @@ function reapMissingMcpSessions(args) {
       stall.escalated = true;
       rotationProbeExhausted.push(key);
       recordRotationOutcome(key, "never-respawned");
-      log(
+      log2(
         `[mcp-presence-reaper] '${codeName}:${key}' rotation outcome=never-respawned \u2014 child SIGTERM'd by stale-mcp-reaper never returned after ${stall.probeAttempts} prompt bind-probe(s); escalating (integration marked unhealthy + reaper_action alert) and releasing to the normal restart path (ENG-8225)`
       );
       if (onRotationRespawnGaveUp) {
         try {
           onRotationRespawnGaveUp(codeName, key, { attempts: stall.probeAttempts });
         } catch (err) {
-          log(
+          log2(
             `[mcp-presence-reaper] onRotationRespawnGaveUp callback threw for '${codeName}:${key}' (suppressed; reaper continues): ${err.message}`
           );
         }
@@ -7194,7 +7195,7 @@ function reapMissingMcpSessions(args) {
           maxAttempts: maxRotationPromptProbes
         });
       } catch (err) {
-        log(
+        log2(
           `[mcp-presence-reaper] onRotationRespawnStalled callback threw for '${codeName}:${key}' (suppressed; falling back to the pre-ENG-8225 restart path): ${err.message}`
         );
         armed = false;
@@ -7207,17 +7208,17 @@ function reapMissingMcpSessions(args) {
     stall.probeAttempts = attempt;
     stall.lastProbeAt = now();
     rotationProbeArmed.push(key);
-    log(
+    log2(
       `[mcp-presence-reaper] '${codeName}:${key}' rotation grace expired (${rotationGraceMs}ms) with no live child and no tool call to respawn it \u2014 armed prompt bind-probe ${attempt}/${maxRotationPromptProbes} (ENG-7429) instead of restarting the session (ENG-8225)`
     );
   }
   if (rotationGraced.length > 0) {
-    log(
+    log2(
       `[mcp-presence-reaper] '${codeName}': skipping [${rotationGraced.join(", ")}] within rotation grace window (${rotationGraceMs}ms) \u2014 child(ren) just SIGTERM'd by stale-mcp-reaper, awaiting respawn on next tool call (ENG-5344)`
     );
   }
   if (rotationProbeWaiting.length > 0) {
-    log(
+    log2(
       `[mcp-presence-reaper] '${codeName}': skipping [${rotationProbeWaiting.join(", ")}] \u2014 prompt bind-probe already armed for the stalled rotation, waiting out the ${rotationPromptProbeIntervalMs}ms re-arm interval (ENG-8225)`
     );
   }
@@ -7240,7 +7241,7 @@ function reapMissingMcpSessions(args) {
       key,
       viaProbe ? "respawned-after-prompt-probe" : "respawned-on-tool-call"
     );
-    log(
+    log2(
       viaProbe ? `[mcp-presence-reaper] '${codeName}:${key}' rotation outcome=respawned-after-prompt-probe \u2014 child is live again after ${stall?.probeAttempts} prompt bind-probe(s)${stall?.escalated ? " (recovered AFTER escalation \u2014 the earlier never-respawned verdict stands for that window)" : ""}, no session restart needed (ENG-8225)` : `[mcp-presence-reaper] '${codeName}:${key}' rotation outcome=respawned-on-tool-call \u2014 child returned on the agent's own next tool call, ENG-5344's assumption held (ENG-8225)`
     );
     clearStaleRotation(codeName, [key]);
@@ -7308,7 +7309,7 @@ function reapMissingMcpSessions(args) {
       genuinelyDead.push(key);
       if (currentGenerationExternallyStarted && !state.deadChannelLogged) {
         state.deadChannelLogged = true;
-        log(
+        log2(
           `[mcp-presence-reaper] '${codeName}:${key}' continuously missing ${Math.round((nowMs - state.continuouslyMissingSince) / 6e4)}m (> ${Math.round(deadChannelMs / 6e4)}m) \u2014 genuinely dead, overriding storm-awareness so it can reach give-up/quarantine instead of churning (ENG-8062)`
         );
       }
@@ -7334,7 +7335,7 @@ function reapMissingMcpSessions(args) {
       state.gaveUp = true;
       if (!state.gaveUpLogged) {
         const cause = overAttemptsCap ? `${MAX_PRESENCE_RESTART_ATTEMPTS} consecutive failed restarts \u2014 declared but never recovers (likely orphaned config; see ENG-5279)` : `${restartsInWindow} restarts within ${Math.round(BACKOFF_WINDOW_MS / 6e4)}m \u2014 persistently flapping (crash-loop; ENG-6480)`;
-        log(
+        log2(
           `[mcp-presence-reaper] giving up on '${codeName}:${key}' after ${cause}`
         );
         state.gaveUpLogged = true;
@@ -7342,7 +7343,7 @@ function reapMissingMcpSessions(args) {
           try {
             onGiveUp(codeName, key);
           } catch (err) {
-            log(
+            log2(
               `[mcp-presence-reaper] onGiveUp callback threw for '${codeName}:${key}' (suppressed; reaper continues): ${err.message}`
             );
           }
@@ -7354,21 +7355,21 @@ function reapMissingMcpSessions(args) {
           state.quarantineLogged = true;
           if (quarantineMode === "enforce") {
             quarantined.push(key);
-            log(
+            log2(
               `[mcp-presence-reaper] QUARANTINE '${codeName}:${key}' \u2014 optional channel dead past restart budget + ${quarantineDwellMs}ms dwell; dropping from provisioned set so the dead channel stops restarting the whole session (ENG-5932)`
             );
             if (onQuarantine) {
               try {
                 onQuarantine(codeName, key);
               } catch (err) {
-                log(
+                log2(
                   `[mcp-presence-reaper] onQuarantine callback threw for '${codeName}:${key}' (suppressed; reaper continues): ${err.message}`
                 );
               }
             }
           } else {
             wouldQuarantine.push(key);
-            log(
+            log2(
               `[mcp-presence-reaper] SHADOW would-quarantine '${codeName}:${key}' \u2014 optional channel dead past restart budget + ${quarantineDwellMs}ms dwell; no action taken (ENG-5932 shadow mode)`
             );
           }
@@ -7390,7 +7391,7 @@ function reapMissingMcpSessions(args) {
   if (active.length === 0) {
     const reason = backingOff.length > 0 ? "backoff" : "all-keys-over-cap";
     if (backingOff.length > 0) {
-      log(
+      log2(
         `[mcp-presence-reaper] '${codeName}': holding off restart for [${backingOff.join(", ")}] \u2014 escalating backoff interval not yet elapsed (ENG-6480)${givenUp.length > 0 ? ` (given up: [${givenUp.join(", ")}])` : ""}`
       );
     }
@@ -7411,7 +7412,7 @@ function reapMissingMcpSessions(args) {
     };
   }
   const givenUpSuffix = givenUp.length > 0 ? ` (skipping over-cap: [${givenUp.join(", ")}])` : "";
-  log(
+  log2(
     `[mcp-presence-reaper] '${codeName}': declared MCP(s) [${active.join(", ")}] have no live children \u2014 restarting session${givenUpSuffix}`
   );
   if (onRestart) {
@@ -7419,13 +7420,13 @@ function reapMissingMcpSessions(args) {
       const maybePromise = onRestart(codeName, { activeKeys: active, givenUpKeys: givenUp });
       if (maybePromise && typeof maybePromise.then === "function") {
         void maybePromise.then(void 0, (err) => {
-          log(
+          log2(
             `[mcp-presence-reaper] onRestart callback rejected for '${codeName}' (suppressed; restart proceeds): ${err.message}`
           );
         });
       }
     } catch (err) {
-      log(
+      log2(
         `[mcp-presence-reaper] onRestart callback threw for '${codeName}' (suppressed; restart proceeds): ${err.message}`
       );
     }
@@ -8861,6 +8862,63 @@ function provision(input, frameworkId = "claude-code") {
   };
 }
 
+// src/lib/startup-telemetry.ts
+var SLOW_STARTUP_WARN_MS = 6e4;
+function shouldWarnSlowStartup(elapsedMs, opts = {}) {
+  if (!Number.isFinite(elapsedMs) || elapsedMs < 0) return false;
+  const threshold = opts.thresholdMs ?? SLOW_STARTUP_WARN_MS;
+  if (!Number.isFinite(threshold) || threshold < 0) return false;
+  return elapsedMs >= threshold;
+}
+function formatStartupElapsed(elapsedMs) {
+  if (!Number.isFinite(elapsedMs) || elapsedMs < 0) return "unknown";
+  const seconds = elapsedMs / 1e3;
+  if (seconds < 60) return `${seconds.toFixed(1)}s`;
+  return `${seconds.toFixed(1)}s (${(seconds / 60).toFixed(1)} min)`;
+}
+function describeStartupCompletion(elapsedMs, opts = {}) {
+  const elapsed = formatStartupElapsed(elapsedMs);
+  if (!shouldWarnSlowStartup(elapsedMs, opts)) {
+    return { warn: false, message: `[startup] ready in ${elapsed}` };
+  }
+  const threshold = opts.thresholdMs ?? SLOW_STARTUP_WARN_MS;
+  return {
+    warn: true,
+    message: `[startup] SLOW BOOT \u2014 ready in ${elapsed}, over the ${formatStartupElapsed(threshold)} threshold. MCP bundles are deployed during startup, so agents spawning in this window can come up with no MCP tools (ENG-9129). Check for a build/install running on the host.`
+  };
+}
+var SLOW_MODULE_LOAD_WARN_MS = 1e4;
+function describeWorkerModuleLoad(elapsedMs, opts = {}) {
+  const elapsed = formatStartupElapsed(elapsedMs);
+  const threshold = opts.thresholdMs ?? SLOW_MODULE_LOAD_WARN_MS;
+  if (!shouldWarnSlowStartup(elapsedMs, { thresholdMs: threshold })) {
+    return { warn: false, message: `[startup] worker module loaded in ${elapsed}` };
+  }
+  return {
+    warn: true,
+    message: `[startup] SLOW MODULE LOAD \u2014 worker module took ${elapsed} to import, over the ${formatStartupElapsed(threshold)} threshold. The manager had already written its pidfile and reported itself started, so it looked healthy for this whole window while doing nothing \u2014 and MCP bundles are deployed after this point, so agents spawning meanwhile come up with no MCP tools (ENG-9129). Check for a build, install or filesystem rebuild running on the host.`
+  };
+}
+function describeRejection(err) {
+  if (err instanceof Error) return err.stack ?? err.message;
+  if (typeof err === "object" && err !== null) {
+    try {
+      const json = JSON.stringify(err);
+      if (json && json !== "{}") return json;
+    } catch {
+    }
+  }
+  try {
+    return String(err);
+  } catch {
+    return `<unprintable ${typeof err} rejection>`;
+  }
+}
+function describeWorkerModuleFailure(err) {
+  const detail = describeRejection(err);
+  return `[startup] FATAL \u2014 the manager worker module failed to load, so this process will never poll. The pidfile it already wrote would otherwise make it look like a healthy manager forever (ENG-9129). Exiting so the supervisor can react. Cause: ${detail}`;
+}
+
 // src/lib/self-update-pin.ts
 function isValidPinVersion(v) {
   return PIN_VERSION_RE.test(v);
@@ -8946,15 +9004,23 @@ function findOtherManagerPids(pgrepImpl = defaultPgrep, selfPid = process.pid) {
   let out;
   try {
     out = pgrepImpl();
-  } catch {
+  } catch (err) {
+    const signal = err?.signal;
+    if (signal) {
+      log(
+        `[startup] duplicate-manager scan timed out after ${PGREP_SCAN_TIMEOUT_MS}ms (killed by ${signal}) \u2014 proceeding without it, since the pidfile check is the primary guard. A process-table read this slow means the host is heavily loaded, which is also when a boot is most likely to stall (ENG-9129).`
+      );
+    }
     return [];
   }
   return out.split("\n").map((line) => parseInt(line.trim(), 10)).filter((pid) => !isNaN(pid) && pid !== selfPid);
 }
+var PGREP_SCAN_TIMEOUT_MS = 1e4;
 function defaultPgrep() {
   return execFileSync3("pgrep", ["-f", "agt manager start"], {
     encoding: "utf-8",
-    stdio: ["ignore", "pipe", "ignore"]
+    stdio: ["ignore", "pipe", "ignore"],
+    timeout: PGREP_SCAN_TIMEOUT_MS
   });
 }
 function readStateFile(configDir) {
@@ -9030,11 +9096,18 @@ function startWatchdog(opts) {
     );
   }
   writePidFile(configDir, process.pid);
+  const moduleLoadStartedAt = process.uptime() * 1e3;
+  log("[startup] loading worker module \u2014 nothing polls, logs or deploys MCP bundles until this resolves");
   void import("./lib/manager-worker.js").then(({ startManager }) => {
+    const { message } = describeWorkerModuleLoad(process.uptime() * 1e3 - moduleLoadStartedAt);
+    log(message);
     startManager({
       intervalMs: opts.intervalMs,
       configDir
     });
+  }).catch((err) => {
+    log(describeWorkerModuleFailure(err));
+    process.exit(1);
   });
   process.on("exit", () => {
     removePidFile(configDir);
@@ -9723,6 +9796,7 @@ export {
   buildConnectivityProbeDeps,
   gatherSessionToolBindProbe,
   provision,
+  describeStartupCompletion,
   executeConnectivityProbe,
   isValidPinVersion,
   resolveEffectivePinRaw,
@@ -9740,4 +9814,4 @@ export {
   managerInstallSystemUnitCommand,
   managerUninstallSystemUnitCommand
 };
-//# sourceMappingURL=chunk-3JBA6RYG.js.map
+//# sourceMappingURL=chunk-LCKKFDCR.js.map