@integrity-labs/agt-cli 0.28.225 → 0.28.227

package/dist/{claude-pair-runtime-JFP2N723.js → claude-pair-runtime-CGFYM63W.js}

@@ -100,7 +100,7 @@ async function spawnPairSession(session) {
     return { ok: true };
   } catch {
   }
-  const { resolveClaudeBinary } = await import("./persistent-session-I4CQEDME.js");
+  const { resolveClaudeBinary } = await import("./persistent-session-VZ44GVA4.js");
   const claudeBin = resolveClaudeBinary();
   const pairEnv = {
     ...process.env,
@@ -373,4 +373,4 @@ export {
   startClaudePair,
   submitClaudePairCode
 };
-//# sourceMappingURL=claude-pair-runtime-JFP2N723.js.map
+//# sourceMappingURL=claude-pair-runtime-CGFYM63W.js.map

package/dist/lib/manager-worker.js

@@ -37,7 +37,7 @@ import {
   requireHost,
   safeWriteJsonAtomic,
   setConfigHash
-} from "../chunk-A36N2XXT.js";
+} from "../chunk-VUPYJ4WU.js";
 import {
   getProjectDir as getProjectDir2,
   getReadyTasks,
@@ -122,7 +122,7 @@ import {
   takeZombieDetection,
   transcriptActivityAgeSeconds,
   writeEgressAllowlist
-} from "../chunk-S43WVYAF.js";
+} from "../chunk-VV5IXUN6.js";
 import {
   reapOrphanChannelMcps
 } from "../chunk-XWVM4KPK.js";
@@ -1251,6 +1251,83 @@ function decideRestartGate(opts) {
   return "proceed";
 }
 
+// src/lib/model-api-error.ts
+function classifyHttpStatus(status) {
+  if (status === 529) return "overloaded_529";
+  if (status === 429) return "rate_limit_429";
+  if (status >= 500 && status <= 599) return "server_error_5xx";
+  return null;
+}
+function classifyModelApiError(input) {
+  if (typeof input.status === "number") {
+    const cls = classifyHttpStatus(input.status);
+    if (cls) return { errorClass: cls, httpStatus: input.status };
+  }
+  const err = input.error;
+  const rawMessage = input.message ?? (err instanceof Error ? err.message : typeof err === "string" ? err : "");
+  const errName = err instanceof Error ? err.name : "";
+  const haystack = `${errName} ${rawMessage}`.toLowerCase();
+  if (!haystack.trim()) return null;
+  const statusMatch = haystack.match(/returned\s+(\d{3})\b/);
+  if (statusMatch) {
+    const status = Number(statusMatch[1]);
+    const cls = classifyHttpStatus(status);
+    if (cls) return { errorClass: cls, httpStatus: status };
+  }
+  if (errName === "TimeoutError" || errName === "AbortError" || /\btimed?\s*out\b|\btimeout\b|\baborted\b|etimedout/.test(haystack)) {
+    return { errorClass: "timeout", httpStatus: null };
+  }
+  if (/econnreset|econnrefused|enotfound|eai_again|epipe|socket hang up|network error|fetch failed|und_err/.test(
+    haystack
+  )) {
+    return { errorClass: "connection_error", httpStatus: null };
+  }
+  if (/overloaded/.test(haystack)) {
+    return { errorClass: "overloaded_529", httpStatus: null };
+  }
+  return null;
+}
+var AGENT_API_ERROR_RE = /API\s*Error[:\s]+(?:(\d{3})\b|([A-Za-z][^\n]{0,60}))/gi;
+function scrapeAgentRuntimeModelApiError(paneTail) {
+  if (!paneTail) return null;
+  let last = null;
+  let m;
+  AGENT_API_ERROR_RE.lastIndex = 0;
+  while ((m = AGENT_API_ERROR_RE.exec(paneTail)) !== null) {
+    if (m[1]) {
+      const status = Number(m[1]);
+      const cls = classifyHttpStatus(status);
+      if (cls) last = { errorClass: cls, httpStatus: status };
+      continue;
+    }
+    const text = (m[2] ?? "").toLowerCase();
+    if (/overloaded/.test(text)) last = { errorClass: "overloaded_529", httpStatus: null };
+    else if (/\btimed?\s*out\b|\btimeout\b/.test(text)) last = { errorClass: "timeout", httpStatus: null };
+  }
+  return last;
+}
+async function reportModelApiError(opts) {
+  try {
+    const hostId = await getHostId();
+    if (!hostId) return;
+    await api.post("/host/model-api-error-events", {
+      host_id: hostId,
+      agent_code_name: opts.agentCodeName ?? null,
+      source: opts.source,
+      error_class: opts.errorClass,
+      http_status: opts.httpStatus ?? null,
+      model_id: opts.modelId ?? null,
+      // Content-free by contract, but bound it defensively.
+      detail: opts.detail ? opts.detail.slice(0, 500) : null,
+      occurred_at: opts.occurredAt ?? (/* @__PURE__ */ new Date()).toISOString()
+    });
+  } catch (err) {
+    opts.log?.(
+      `[model-api-error] report failed (${opts.source}/${opts.errorClass}): ${err instanceof Error ? err.message : String(err)}`
+    );
+  }
+}
+
 // src/lib/claude-pid-tracker.ts
 import { existsSync, readFileSync as readFileSync5 } from "fs";
 function readPidFile(path) {
@@ -2179,6 +2256,7 @@ async function maybeEvaluateConversations(args) {
     } catch (err) {
       backendError = toHealthErrorMessage(err);
       log2(`[conversation-eval] ${codeName}: scoring failed: ${backendError}`);
+      args.onModelCallError?.(err);
       continue;
     }
     if (!verdict) {
@@ -2394,6 +2472,7 @@ async function maybeExtractMemories(args) {
       candidates = parseCandidates(out);
     } catch (err) {
       log2(`[memory-extract] ${codeName}: extraction failed: ${err.message}`);
+      args.onModelCallError?.(err);
       continue;
     }
     if (candidates === null) {
@@ -6055,6 +6134,11 @@ async function runAgentConnectivityProbes(agent, integrations, projectDir) {
     );
   }
 }
+var sessionToolRebindCooldownAt = /* @__PURE__ */ new Map();
+var SESSION_TOOL_REBIND_COOLDOWN_MS = 10 * 60 * 1e3;
+function sessionToolRebindKey(codeName, serverKey) {
+  return `${codeName}\0${serverKey}`;
+}
 async function runAgentSessionToolBindProbes(agent, integrations, projectDir) {
   if (integrations.length === 0) return;
   const intervalSec = Number(process.env.AGT_CONNECTIVITY_PROBE_INTERVAL_SECONDS) || 3600;
@@ -6074,6 +6158,36 @@ async function runAgentSessionToolBindProbes(agent, integrations, projectDir) {
       `Session-tool-bind probe for '${agent.code_name}': probed=${result.probed} reported=${result.reports.length} due=${result.due}`
     );
   }
+  if (result && result.rebindCandidates.length > 0 && hostFlagStore().getBoolean("session-tool-rebind")) {
+    let mcpJsonForRebind = null;
+    try {
+      mcpJsonForRebind = JSON.parse(readFileSync14(join16(projectDir, ".mcp.json"), "utf-8"));
+    } catch {
+      mcpJsonForRebind = null;
+    }
+    const isolated = isolationMode(agent.code_name) === "docker";
+    const now = Date.now();
+    for (const candidate of result.rebindCandidates) {
+      const keysToReap = candidate.serverKeys.filter((k) => {
+        const last = sessionToolRebindCooldownAt.get(sessionToolRebindKey(agent.code_name, k));
+        return last === void 0 || now - last >= SESSION_TOOL_REBIND_COOLDOWN_MS;
+      });
+      if (keysToReap.length === 0) continue;
+      reapStaleMcpChildren({
+        log,
+        codeName: agent.code_name,
+        serverKeys: keysToReap,
+        mcpJson: mcpJsonForRebind,
+        isolated
+      });
+      for (const k of keysToReap) {
+        sessionToolRebindCooldownAt.set(sessionToolRebindKey(agent.code_name, k), now);
+      }
+      log(
+        `Session-tool rebind for '${agent.code_name}': ${candidate.status} integration=${candidate.integration_id} reaped MCP child(ren) [${keysToReap.join(", ")}] for respawn`
+      );
+    }
+  }
 }
 function stopPersistentSessionAndForgetMcpBaseline(codeName, breakerReason, gateReason = breakerReason) {
   const gate = restartGateFor(codeName, gateReason);
@@ -6242,6 +6356,7 @@ function clearAgentCaches(agentId, codeName) {
   claudeSchedulerStates.delete(codeName);
   claudeTaskConcurrency.delete(codeName);
   lastMcpFailedBannerCount.delete(codeName);
+  lastModelApiErrorSig.delete(codeName);
   memoryFileHashes.delete(agentId);
   lastDownloadHash.delete(agentId);
   lastLocalFileHash.delete(agentId);
@@ -6258,7 +6373,7 @@ var agentRestartTimezoneInputs = /* @__PURE__ */ new Map();
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
 var lastResponsivenessProbeAt = 0;
-var agtCliVersion = true ? "0.28.225" : "dev";
+var agtCliVersion = true ? "0.28.227" : "dev";
 function resolveBrewPath(execFileSync2) {
   try {
     const out = execFileSync2("which", ["brew"], { timeout: 5e3 }).toString().trim();
@@ -6980,6 +7095,22 @@ function hostFlagStore() {
   }
   return _hostFlagStore;
 }
+function reportManagerModelCallError(source, err, modelId) {
+  try {
+    if (!hostFlagStore().getBoolean("model-api-error-reporting")) return;
+    const classified = classifyModelApiError({ error: err });
+    if (!classified) return;
+    void reportModelApiError({
+      source,
+      errorClass: classified.errorClass,
+      httpStatus: classified.httpStatus,
+      modelId,
+      detail: err instanceof Error ? err.message : String(err),
+      log
+    });
+  } catch {
+  }
+}
 function channelQuarantineMode() {
   return hostFlagStore().getString("channel-quarantine-mode");
 }
@@ -7167,7 +7298,7 @@ async function pollCycle() {
     }
     try {
       const { detectHostSecurity } = await import("../host-security-6PDFG7F5.js");
-      const { collectDiagnostics } = await import("../persistent-session-I4CQEDME.js");
+      const { collectDiagnostics } = await import("../persistent-session-VZ44GVA4.js");
       const diagCodeNames = [...agentState.persistentSessionAgents];
       const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames) : void 0;
       let tailscaleHostname;
@@ -7315,7 +7446,7 @@ async function pollCycle() {
       const {
         collectResponsivenessProbes,
         getResponsivenessIntervalMs
-      } = await import("../responsiveness-probe-MIUMXKV4.js");
+      } = await import("../responsiveness-probe-DLIVXYKX.js");
       const probeIntervalMs = getResponsivenessIntervalMs();
       if (now - lastResponsivenessProbeAt > probeIntervalMs) {
         const probeCodeNames = [...agentState.persistentSessionAgents];
@@ -7347,7 +7478,7 @@ async function pollCycle() {
           collectResponsivenessProbes,
           livePendingInboundOldestAgeSeconds,
           parkPendingInbound
-        } = await import("../responsiveness-probe-MIUMXKV4.js");
+        } = await import("../responsiveness-probe-DLIVXYKX.js");
         const { getProjectDir: wedgeProjectDir } = await import("../claude-scheduler-FATCLHDM.js");
         const wedgeNow = /* @__PURE__ */ new Date();
         const liveAgents = agentState.persistentSessionAgents;
@@ -7480,11 +7611,27 @@ async function pollCycle() {
       const { scrapeMcpFailedBannerCount } = await import("../pane-mcp-banner-scraper-JA437JIB.js");
       const observations = [];
       const pendingCacheCommits = [];
+      const modelApiErrorReportingOn = hostFlagStore().getBoolean("model-api-error-reporting");
       for (const codeName of agentState.persistentSessionAgents) {
         const agentId = agentState.codeNameToAgentId.get(codeName);
         if (!agentId) continue;
         const tail = readPaneLogTail(codeName, 40);
         if (!tail) continue;
+        if (modelApiErrorReportingOn) {
+          const observedErr = scrapeAgentRuntimeModelApiError(tail);
+          const sig = observedErr ? `${observedErr.errorClass}:${observedErr.httpStatus ?? ""}` : "";
+          if (observedErr && sig !== (lastModelApiErrorSig.get(codeName) ?? "")) {
+            void reportModelApiError({
+              source: "agent_runtime",
+              errorClass: observedErr.errorClass,
+              httpStatus: observedErr.httpStatus,
+              agentCodeName: codeName,
+              detail: `agent pane: ${observedErr.errorClass}`,
+              log
+            });
+          }
+          lastModelApiErrorSig.set(codeName, sig);
+        }
         const observed = scrapeMcpFailedBannerCount(tail);
         const count = observed ?? 0;
         const last = lastMcpFailedBannerCount.get(codeName);
@@ -7602,7 +7749,7 @@ async function pollCycle() {
       if (restartedCodeNames.length > 0) {
         void (async () => {
           try {
-            const { collectDiagnostics } = await import("../persistent-session-I4CQEDME.js");
+            const { collectDiagnostics } = await import("../persistent-session-VZ44GVA4.js");
             const freshDiagnostics = collectDiagnostics(restartedCodeNames);
             await api.post("/host/heartbeat", {
               host_id: hostId,
@@ -7833,21 +7980,25 @@ async function processAgent(agent, agentStates) {
       logMeteredScorerSuppressed();
     }
     if (!scorerSuppressed) {
+      const evalBackend = resolveConversationEvalBackend();
       void maybeEvaluateConversations({
         api,
-        backend: resolveConversationEvalBackend(),
+        backend: evalBackend,
         codeName: agent.code_name,
         agentId: agent.agent_id,
-        log
+        log,
+        onModelCallError: (err) => reportManagerModelCallError("conversation_eval", err, evalBackend.model ?? null)
       });
     }
     if (memoryExtractionEnabled() && !scorerSuppressed) {
+      const memBackend = resolveConversationEvalBackend();
       void maybeExtractMemories({
         api,
-        backend: resolveConversationEvalBackend(),
+        backend: memBackend,
         codeName: agent.code_name,
         agentId: agent.agent_id,
-        log
+        log,
+        onModelCallError: (err) => reportManagerModelCallError("memory_extraction", err, memBackend.model ?? null)
       });
     }
   }
@@ -9592,6 +9743,7 @@ In progress for ${age} minutes \u2014 auto-failed`).catch(() => {
   });
 }
 var lastMcpFailedBannerCount = /* @__PURE__ */ new Map();
+var lastModelApiErrorSig = /* @__PURE__ */ new Map();
 var persistentSessionStuckTracker = new PersistentSessionStuckTracker();
 var claudeAuthTupleBySession = /* @__PURE__ */ new Map();
 var egressAllowlistBySession = /* @__PURE__ */ new Map();
@@ -10100,7 +10252,7 @@ async function handleRestartDoorbell(agentId, requestedAt, restartReason) {
     void api.post("/host/restart-ack", { host_id: hostId, agent_id: agentId, restart_requested_at: requestedAt }).catch((err) => log(`[restart-lane] ack failed for '${codeName}': ${err.message}`));
     void (async () => {
       try {
-        const { collectDiagnostics } = await import("../persistent-session-I4CQEDME.js");
+        const { collectDiagnostics } = await import("../persistent-session-VZ44GVA4.js");
         await api.post("/host/heartbeat", {
           host_id: hostId,
           agent_diagnostics: collectDiagnostics([codeName])
@@ -10494,7 +10646,7 @@ async function processClaudePairSessions(agents) {
     killPairSession,
     pairTmuxSession,
     finalizeClaudePairOnboarding
-  } = await import("../claude-pair-runtime-JFP2N723.js");
+  } = await import("../claude-pair-runtime-CGFYM63W.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));