@integrity-labs/agt-cli 0.19.22 → 0.19.25

package/dist/lib/manager-worker.js

@@ -22,7 +22,7 @@ import {
   resolveChannels,
   resolveDmTarget,
   wrapScheduledTaskPrompt
-} from "../chunk-TD4ZSQ74.js";
+} from "../chunk-EJKRQX3I.js";
 import {
   findTaskByTemplate,
   getProjectDir,
@@ -35,6 +35,7 @@ import {
   buildAllowedTools,
   getLastFailureContext,
   getProjectDir as getProjectDir2,
+  getSessionState,
   injectMessage,
   isAgentIdle,
   isSessionHealthy,
@@ -49,7 +50,7 @@ import {
   startPersistentSession,
   stopAllSessionsAndWait,
   stopPersistentSession
-} from "../chunk-3K3RO5NS.js";
+} from "../chunk-ZQNOWGHB.js";
 
 // src/lib/manager-worker.ts
 import { createHash as createHash2 } from "crypto";
@@ -295,6 +296,68 @@ function reapStaleMcpChildren(args) {
   return targets;
 }
 
+// src/lib/mcp-presence-reaper.ts
+import { execFileSync as execFileSync2 } from "child_process";
+var DEFAULT_COLD_START_GRACE_MS = 9e4;
+function findMissingMcpServers(args) {
+  const { rows, codeName, mcpJson } = args;
+  const declared = Object.keys(mcpJson?.mcpServers ?? {});
+  if (declared.length === 0) return [];
+  const missing = [];
+  for (const key of declared) {
+    const live = findMcpChildrenForAgent({
+      rows,
+      codeName,
+      serverKeys: [key],
+      mcpJson
+    });
+    if (live.length === 0) missing.push(key);
+  }
+  return missing;
+}
+function reapMissingMcpSessions(args) {
+  const {
+    log: log2,
+    codeName,
+    mcpJson,
+    sessionStartedAt,
+    stopSession,
+    graceMs = DEFAULT_COLD_START_GRACE_MS
+  } = args;
+  const now = args.now ?? Date.now;
+  const runPs = args.runPs ?? (() => execFileSync2("ps", ["-eo", "pid,ppid,args"], { encoding: "utf-8", timeout: 5e3 }));
+  if (!mcpJson?.mcpServers) {
+    return { missing: [], restarted: false, reason: "no-mcp-json" };
+  }
+  const declaredCount = Object.keys(mcpJson.mcpServers).length;
+  if (declaredCount === 0) {
+    return { missing: [], restarted: false, reason: "no-declared-servers" };
+  }
+  if (sessionStartedAt === null) {
+    return { missing: [], restarted: false, reason: "session-start-unknown" };
+  }
+  if (now() - sessionStartedAt < graceMs) {
+    return { missing: [], restarted: false, reason: "cold-start" };
+  }
+  let psOutput;
+  try {
+    psOutput = runPs();
+  } catch (err) {
+    log2(`[mcp-presence-reaper] ps invocation failed for '${codeName}': ${err.message} \u2014 skipping`);
+    return { missing: [], restarted: false, reason: "healthy" };
+  }
+  const rows = parsePsRows(psOutput);
+  const missing = findMissingMcpServers({ rows, codeName, mcpJson });
+  if (missing.length === 0) {
+    return { missing, restarted: false, reason: "healthy" };
+  }
+  log2(
+    `[mcp-presence-reaper] '${codeName}': declared MCP(s) [${missing.join(", ")}] have no live children \u2014 restarting session`
+  );
+  stopSession(codeName);
+  return { missing, restarted: true };
+}
+
 // src/lib/channel-restart-decision.ts
 function launchableChannelIds(channelConfigs) {
   if (!channelConfigs) return /* @__PURE__ */ new Set();
@@ -323,6 +386,22 @@ function decideChannelRestart(input) {
   return { restart: true, added, removed };
 }
 
+// src/lib/poll-backoff.ts
+var POLL_BACKOFF_BASE_MS = 1e3;
+var POLL_BACKOFF_MAX_MS = 6e4;
+function currentPollBackoffMs(consecutivePollFailures2) {
+  if (consecutivePollFailures2 <= 0) return 0;
+  const expDelay = POLL_BACKOFF_BASE_MS * Math.pow(2, consecutivePollFailures2 - 1);
+  return Math.min(POLL_BACKOFF_MAX_MS, expDelay);
+}
+function classifyPollError(message) {
+  const msg = message.toLowerCase();
+  if (msg.includes("api unreachable") || /\b5\d\d\b/.test(msg) || msg.includes("internal server error") || msg.includes("bad gateway")) return "5xx";
+  if (msg.includes("exchange failed") || /\b401\b/.test(msg) || msg.includes("invalid token")) return "auth";
+  if (msg.includes("econn") || msg.includes("etimedout") || msg.includes("fetch failed")) return "network";
+  return "other";
+}
+
 // src/lib/integration-context-render.ts
 var PLUGIN_CONTEXT_PLACEHOLDER_RE = /\{\{\s*context\.([a-zA-Z_][a-zA-Z0-9_]*)\s*\}\}/g;
 var TEAM_OVERRIDES_HEADER = "## Team Overrides\n\n> **The following overrides anything you've read above.** If any rule here conflicts with earlier instructions in this skill, follow what is written here. These are user-supplied directives that take precedence over the plugin's default guidance.\n";
@@ -885,7 +964,7 @@ function saveChannelHashCache(source, configDir) {
 }
 
 // src/lib/channel-sweep.ts
-import { execFileSync as execFileSync2 } from "child_process";
+import { execFileSync as execFileSync3 } from "child_process";
 var CHANNEL_BASENAMES = [
   "slack-channel",
   "direct-chat-channel",
@@ -1048,7 +1127,7 @@ function resolveLiveAnchorPids(agentCodeNames) {
   for (const codeName of agentCodeNames) {
     const pids = /* @__PURE__ */ new Set();
     try {
-      const out = execFileSync2("tmux", ["list-panes", "-t", `agt-${codeName}`, "-F", "#{pane_pid}"], {
+      const out = execFileSync3("tmux", ["list-panes", "-t", `agt-${codeName}`, "-F", "#{pane_pid}"], {
         encoding: "utf-8",
         timeout: 2e3,
         stdio: ["ignore", "pipe", "ignore"]
@@ -1068,7 +1147,7 @@ async function sweepChannelProcesses(opts) {
   const kill = opts.killFn ?? defaultKill;
   let psOutput = "";
   try {
-    psOutput = execFileSync2("ps", ["eww", "-o", "pid=,ppid=,etime=,command="], {
+    psOutput = execFileSync3("ps", ["eww", "-o", "pid=,ppid=,etime=,command="], {
       encoding: "utf-8",
       timeout: 5e3,
       maxBuffer: 10 * 1024 * 1024
@@ -1105,7 +1184,7 @@ function defaultKillSignal(pid, signal) {
   }
 }
 function defaultPs() {
-  return execFileSync2("ps", ["eww", "-o", "pid=,ppid=,etime=,command="], {
+  return execFileSync3("ps", ["eww", "-o", "pid=,ppid=,etime=,command="], {
     encoding: "utf-8",
     timeout: 5e3,
     maxBuffer: 10 * 1024 * 1024
@@ -2060,10 +2139,10 @@ function clearAgentCaches(agentId, codeName) {
 var cachedFrameworkVersion = null;
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
-var agtCliVersion = true ? "0.19.22" : "dev";
-function resolveBrewPath(execFileSync3) {
+var agtCliVersion = true ? "0.19.25" : "dev";
+function resolveBrewPath(execFileSync4) {
   try {
-    const out = execFileSync3("which", ["brew"], { timeout: 5e3 }).toString().trim();
+    const out = execFileSync4("which", ["brew"], { timeout: 5e3 }).toString().trim();
     if (out) return out;
   } catch {
   }
@@ -2105,9 +2184,9 @@ async function ensureToolkitCli(toolkitSlug) {
   }
   const { binary, installer, package: pkg, script } = integration.cli_tool;
   const resolvedInstaller = installer ?? "manual";
-  const { execFileSync: execFileSync3, execSync } = await import("child_process");
+  const { execFileSync: execFileSync4, execSync } = await import("child_process");
   try {
-    execFileSync3("which", [binary], { timeout: 5e3, stdio: "pipe" });
+    execFileSync4("which", [binary], { timeout: 5e3, stdio: "pipe" });
     toolkitCliEnsured.add(toolkitSlug);
     toolkitCliRetryAfter.delete(toolkitSlug);
     toolkitCliFailureCount.delete(toolkitSlug);
@@ -2128,14 +2207,14 @@ async function ensureToolkitCli(toolkitSlug) {
         return;
       }
       log(`[toolkit-install] ${toolkitSlug}: installing via npm (${pkg})\u2026`);
-      execFileSync3("npm", ["install", "-g", pkg], { timeout: 18e4, stdio: "pipe" });
+      execFileSync4("npm", ["install", "-g", pkg], { timeout: 18e4, stdio: "pipe" });
     } else if (resolvedInstaller === "brew") {
       if (!pkg) {
         log(`[toolkit-install] ${toolkitSlug}: installer=brew but no package declared`);
         toolkitCliEnsured.add(toolkitSlug);
         return;
       }
-      const brewPath = resolveBrewPath(execFileSync3);
+      const brewPath = resolveBrewPath(execFileSync4);
       if (!brewPath) {
         log(`[toolkit-install] ${toolkitSlug}: installer=brew but Homebrew not available \u2014 install manually: brew install ${pkg}`);
         toolkitCliEnsured.add(toolkitSlug);
@@ -2145,9 +2224,9 @@ async function ensureToolkitCli(toolkitSlug) {
       const isRoot = typeof process.getuid === "function" && process.getuid() === 0;
       log(`[toolkit-install] ${toolkitSlug}: installing via brew (${pkg})\u2026`);
       if (isRoot) {
-        execFileSync3("sudo", ["-u", "ec2-user", "-H", brewPath, "install", pkg], { timeout: 18e4, stdio: "pipe", cwd: "/tmp" });
+        execFileSync4("sudo", ["-u", "ec2-user", "-H", brewPath, "install", pkg], { timeout: 18e4, stdio: "pipe", cwd: "/tmp" });
       } else {
-        execFileSync3(brewPath, ["install", pkg], { timeout: 18e4, stdio: "pipe" });
+        execFileSync4(brewPath, ["install", pkg], { timeout: 18e4, stdio: "pipe" });
       }
     } else if (resolvedInstaller === "script") {
       if (!script) {
@@ -2167,7 +2246,7 @@ async function ensureToolkitCli(toolkitSlug) {
     process.env.PATH = `${brewBinDir}:${process.env.PATH ?? ""}`;
   }
   try {
-    execFileSync3("which", [binary], { timeout: 5e3, stdio: "pipe" });
+    execFileSync4("which", [binary], { timeout: 5e3, stdio: "pipe" });
     log(`[toolkit-install] ${toolkitSlug}: installed \u2014 ${binary} now on PATH`);
     toolkitCliEnsured.add(toolkitSlug);
     toolkitCliRetryAfter.delete(toolkitSlug);
@@ -2220,8 +2299,8 @@ async function ensureFrameworkBinary(frameworkId) {
   if (frameworkId !== "claude-code") return;
   if (frameworkBinaryChecked.has(frameworkId)) return;
   frameworkBinaryChecked.add(frameworkId);
-  const { execFileSync: execFileSync3 } = await import("child_process");
-  const brewPath = resolveBrewPath(execFileSync3);
+  const { execFileSync: execFileSync4 } = await import("child_process");
+  const brewPath = resolveBrewPath(execFileSync4);
   if (!brewPath) {
     log("Homebrew not found (no `brew` on PATH, no /home/linuxbrew/.linuxbrew/bin/brew). Cannot auto-install Claude Code. Install manually: https://claude.ai/download");
     return;
@@ -2237,7 +2316,7 @@ async function ensureFrameworkBinary(frameworkId) {
   let claudeExists = existsSync3("/home/linuxbrew/.linuxbrew/bin/claude");
   if (!claudeExists) {
     try {
-      execFileSync3("which", ["claude"], { timeout: 5e3 });
+      execFileSync4("which", ["claude"], { timeout: 5e3 });
       claudeExists = true;
     } catch {
     }
@@ -2329,16 +2408,16 @@ async function checkAndUpdateCli() {
   }
 }
 async function checkAndUpdateCliViaBrew() {
-  const { execFileSync: execFileSync3 } = await import("child_process");
-  const brewPath = resolveBrewPath(execFileSync3);
+  const { execFileSync: execFileSync4 } = await import("child_process");
+  const brewPath = resolveBrewPath(execFileSync4);
   if (!brewPath) return;
   try {
-    execFileSync3(brewPath, ["update", "--quiet"], { timeout: 6e4, stdio: "pipe" });
+    execFileSync4(brewPath, ["update", "--quiet"], { timeout: 6e4, stdio: "pipe" });
   } catch (err) {
     log(`[self-update] brew update failed (continuing with stale cache): ${err.message}`);
   }
   try {
-    const outdated = execFileSync3(brewPath, ["outdated", "--json=v2"], {
+    const outdated = execFileSync4(brewPath, ["outdated", "--json=v2"], {
       timeout: 3e4,
       encoding: "utf-8"
     });
@@ -2349,7 +2428,7 @@ async function checkAndUpdateCliViaBrew() {
       const latest = agtOutdated.current_version ?? "unknown";
       log(`[self-update] agt CLI update available: ${installed} \u2192 ${latest}. Upgrading via brew...`);
       try {
-        execFileSync3(brewPath, ["upgrade", "integrity-labs/tap/agt"], {
+        execFileSync4(brewPath, ["upgrade", "integrity-labs/tap/agt"], {
           timeout: 12e4,
           stdio: "pipe"
         });
@@ -2368,7 +2447,7 @@ async function checkAndUpdateCliViaBrew() {
   }
 }
 async function checkAndUpdateCliViaNpm() {
-  const { execFileSync: execFileSync3 } = await import("child_process");
+  const { execFileSync: execFileSync4 } = await import("child_process");
   if (agtCliVersion === "dev") return;
   let latest;
   try {
@@ -2417,7 +2496,7 @@ async function checkAndUpdateCliViaNpm() {
     "--registry=https://registry.npmjs.org"
   ];
   try {
-    execFileSync3(cmd, args, { timeout: 18e4, stdio: "pipe" });
+    execFileSync4(cmd, args, { timeout: 18e4, stdio: "pipe" });
     log(`[self-update] agt CLI upgraded to ${latest}. Scheduling manager restart so the new binary takes effect.`);
     restartAfterUpgrade = true;
     pendingUpgradeVersion = latest;
@@ -2895,6 +2974,7 @@ Automatic restart failed: ${err.message}`
     }
   }
 }
+var consecutivePollFailures = 0;
 async function pollCycle() {
   if (!config) return;
   if (restartAfterUpgrade) return;
@@ -2963,7 +3043,7 @@ async function pollCycle() {
     }
     try {
       const { detectHostSecurity } = await import("../host-security-6PDFG7F5.js");
-      const { collectDiagnostics } = await import("../persistent-session-M2GVL6Z6.js");
+      const { collectDiagnostics } = await import("../persistent-session-J2K3JFJQ.js");
       const diagCodeNames = [...persistentSessionAgents];
       const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames) : void 0;
       let tailscaleHostname;
@@ -3193,17 +3273,23 @@ async function pollCycle() {
       pollCount: state.pollCount + 1,
       agents: agentStates
     };
+    if (consecutivePollFailures > 0) {
+      log(`[poll-backoff] recovered after ${consecutivePollFailures} failure(s), resuming normal interval`);
+      consecutivePollFailures = 0;
+    }
     send({ type: "state-update", state });
   } catch (err) {
     state.errorCount++;
     const message = err.message;
     log(`Poll error: ${message}`);
     send({ type: "error", message });
-    if (message.includes("exchange failed") || message.includes("401")) {
-      log("Fatal auth error, exiting for watchdog restart");
-      send({ type: "shutdown" });
-      process.exit(1);
-    }
+    consecutivePollFailures += 1;
+    const scheduledDelayMs = Math.max(
+      config?.intervalMs ?? 0,
+      currentPollBackoffMs(consecutivePollFailures)
+    );
+    const errorClass = classifyPollError(message);
+    log(`[poll-backoff] attempt=${consecutivePollFailures} class=${errorClass} next_retry_ms=${scheduledDelayMs}`);
   }
 }
 async function getOrCacheRegisteredAgents(adapter, profile) {
@@ -4136,6 +4222,24 @@ async function processAgent(agent, agentStates) {
     } catch (err) {
       log(`[hot-reload] .mcp.json drift check failed for '${agent.code_name}': ${err.message}`);
     }
+    try {
+      const sess = getSessionState(agent.code_name);
+      let mcpJsonParsed = null;
+      try {
+        const mcpPath = join4(getProjectDir2(agent.code_name), ".mcp.json");
+        mcpJsonParsed = JSON.parse(readFileSync3(mcpPath, "utf-8"));
+      } catch {
+      }
+      reapMissingMcpSessions({
+        log,
+        codeName: agent.code_name,
+        mcpJson: mcpJsonParsed,
+        sessionStartedAt: sess?.startedAt ?? null,
+        stopSession: (codeName) => stopPersistentSessionAndForgetMcpBaseline(codeName)
+      });
+    } catch (err) {
+      log(`[mcp-presence-reaper] presence check failed for '${agent.code_name}': ${err.message}`);
+    }
   } else if (agentFw === "claude-code" && tasks.length > 0) {
     await syncAndCheckClaudeScheduler(agent, tasks, boardItems, refreshData);
   } else if (frameworkAdapter.syncScheduledTasks && gatewayRunning && gatewayPort) {
@@ -4951,6 +5055,11 @@ ${truncateForLog(ctx.tail)}` : `; pane_tail_hash=sha256:${createHash2("sha256").
     } catch (err) {
       log(`[persistent-session] Failed to provision isolation hook for '${codeName}': ${err.message}`);
     }
+    const sessionRunResult = await startRun({
+      agent_id: agent.agent_id,
+      source_type: "system",
+      metadata: { type: "persistent_session_boot" }
+    });
     startPersistentSession({
       codeName,
       agentId: agent.agent_id,
@@ -4962,6 +5071,7 @@ ${truncateForLog(ctx.tail)}` : `; pane_tail_hash=sha256:${createHash2("sha256").
       apiHost: requireHost(),
       claudeAuthMode,
       anthropicApiKey,
+      runId: sessionRunResult.run_id,
       log
     });
     persistentSessionAgents.add(codeName);
@@ -6299,7 +6409,7 @@ async function processClaudePairSessions(agents) {
     killPairSession,
     pairTmuxSession,
     finalizeClaudePairOnboarding
-  } = await import("../claude-pair-runtime-UF4OMFCA.js");
+  } = await import("../claude-pair-runtime-LI2ISCVI.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));
@@ -6810,13 +6920,15 @@ function scheduleNext() {
     restartNow();
     return;
   }
+  const backoffMs = currentPollBackoffMs(consecutivePollFailures);
+  const delayMs = Math.max(config.intervalMs, backoffMs);
   pollTimer = setTimeout(() => {
     if (restartAfterUpgrade) {
       restartNow();
       return;
     }
     void pollCycle().then(scheduleNext);
-  }, config.intervalMs);
+  }, delayMs);
 }
 async function killAllAgtTmuxSessions() {
   try {