@integrity-labs/agt-cli 0.28.104 → 0.28.106

package/dist/{chunk-WNN5WT42.js → chunk-O4OYAFTZ.js}

@@ -3,7 +3,7 @@ import {
   formatMissingVar,
   isClaudeFastMode,
   probeMcpEnvSubstitution
-} from "./chunk-5DYG42FL.js";
+} from "./chunk-D22IMWAZ.js";
 import {
   reapOrphanChannelMcps
 } from "./chunk-XWVM4KPK.js";
@@ -1565,4 +1565,4 @@ export {
   stopAllSessionsAndWait,
   getProjectDir
 };
-//# sourceMappingURL=chunk-WNN5WT42.js.map
+//# sourceMappingURL=chunk-O4OYAFTZ.js.map

package/dist/{claude-pair-runtime-JP3TEGDV.js → claude-pair-runtime-6DZEDQAG.js}

@@ -100,7 +100,7 @@ async function spawnPairSession(session) {
     return { ok: true };
   } catch {
   }
-  const { resolveClaudeBinary } = await import("./persistent-session-6CU3BIHP.js");
+  const { resolveClaudeBinary } = await import("./persistent-session-RMRG5HXI.js");
   const claudeBin = resolveClaudeBinary();
   const pairEnv = {
     ...process.env,
@@ -373,4 +373,4 @@ export {
   startClaudePair,
   submitClaudePairCode
 };
-//# sourceMappingURL=claude-pair-runtime-JP3TEGDV.js.map
+//# sourceMappingURL=claude-pair-runtime-6DZEDQAG.js.map

package/dist/lib/manager-worker.js

@@ -28,7 +28,7 @@ import {
   requireHost,
   safeWriteJsonAtomic,
   setConfigHash
-} from "../chunk-DO6ZJ2SC.js";
+} from "../chunk-5VPPNSKR.js";
 import {
   getProjectDir as getProjectDir2,
   getReadyTasks,
@@ -71,7 +71,7 @@ import {
   takeZombieDetection,
   transcriptActivityAgeSeconds,
   writeEgressAllowlist
-} from "../chunk-WNN5WT42.js";
+} from "../chunk-O4OYAFTZ.js";
 import {
   FLAGS_SCHEMA_VERSION,
   FLAG_REGISTRY,
@@ -96,6 +96,7 @@ import {
   isParseError,
   isResolveError,
   isSelfCompletion,
+  laneTagFragment,
   parseDeliveryTarget,
   parseEnvIntegrations,
   parseTranscriptUsage,
@@ -106,7 +107,7 @@ import {
   resolveDmTarget,
   sumTranscriptUsageInWindow,
   wrapScheduledTaskPrompt
-} from "../chunk-5DYG42FL.js";
+} from "../chunk-D22IMWAZ.js";
 import {
   parsePsRows,
   reapOrphanChannelMcps
@@ -2558,7 +2559,9 @@ async function maybeEvaluateConversations(args) {
     );
     pending = resp?.conversations ?? [];
   } catch (err) {
-    log2(`[conversation-eval] ${codeName}: pending fetch failed: ${err.message}`);
+    const msg = toHealthErrorMessage(err);
+    log2(`[conversation-eval] ${codeName}: pending fetch failed: ${msg}`);
+    await reportBackendHealth(api2, log2, codeName, { ok: false, error: `pending fetch: ${msg}` });
     return;
   }
   if (pending.length === 0) return;
@@ -2566,22 +2569,24 @@ async function maybeEvaluateConversations(args) {
   const allTurns = readRecentTurns(dir, nowMs);
   if (allTurns.length === 0) {
     for (const conv of pending) {
-      await reportSkip(api2, agentId, conv.conversation_id, log2, codeName);
+      await reportSkip(api2, agentId, conv.conversation_id, "no_transcript", log2, codeName);
     }
     return;
   }
+  let backendSucceeded = false;
+  let backendError = null;
   for (const conv of pending) {
     const tokens = channelRefTokens(conv.channel_ref);
     const windowStart = Date.parse(conv.started_at) - WINDOW_PAD_MS;
     const windowEnd = Date.parse(conv.last_message_at) + WINDOW_PAD_MS;
     const turns = reconstructConversation(allTurns, tokens, windowStart, windowEnd);
     if (turns.length === 0) {
-      await reportSkip(api2, agentId, conv.conversation_id, log2, codeName);
+      await reportSkip(api2, agentId, conv.conversation_id, "not_reconstructable", log2, codeName);
       continue;
     }
     const transcript = renderTranscript(turns);
     if (!transcript.trim()) {
-      await reportSkip(api2, agentId, conv.conversation_id, log2, codeName);
+      await reportSkip(api2, agentId, conv.conversation_id, "empty_transcript", log2, codeName);
       continue;
     }
     let verdict;
@@ -2589,13 +2594,16 @@ async function maybeEvaluateConversations(args) {
       const out = await backend.run(buildEvalPrompt(conv.channel, transcript));
       verdict = parseVerdict(out);
     } catch (err) {
-      log2(`[conversation-eval] ${codeName}: scoring failed: ${err.message}`);
+      backendError = toHealthErrorMessage(err);
+      log2(`[conversation-eval] ${codeName}: scoring failed: ${backendError}`);
       continue;
     }
     if (!verdict) {
+      backendError = "unparseable verdict";
       log2(`[conversation-eval] ${codeName}: unparseable verdict for ${conv.conversation_id.slice(0, 8)}`);
       continue;
     }
+    backendSucceeded = true;
     try {
       await api2.post("/host/conversations/evaluation", {
         agent_id: agentId,
@@ -2612,18 +2620,42 @@ async function maybeEvaluateConversations(args) {
       log2(`[conversation-eval] ${codeName}: report failed: ${err.message}`);
     }
   }
+  if (backendSucceeded) {
+    await reportBackendHealth(api2, log2, codeName, { ok: true });
+  } else if (backendError) {
+    await reportBackendHealth(api2, log2, codeName, { ok: false, error: backendError, model: backend.model });
+  }
 }
-async function reportSkip(api2, agentId, conversationId, log2, codeName) {
+async function reportSkip(api2, agentId, conversationId, reason, log2, codeName) {
   try {
     await api2.post("/host/conversations/evaluation", {
       agent_id: agentId,
       conversation_id: conversationId,
-      skipped: true
+      skipped: true,
+      // The data-shaped reason this conversation couldn't be scored. The API
+      // stores it alongside the eval_attempts bump so the eventual give-up is
+      // explainable on the reporting surface.
+      failure_reason: reason
     });
   } catch (err) {
     log2(`[conversation-eval] ${codeName}: skip report failed: ${err.message}`);
   }
 }
+function toHealthErrorMessage(err) {
+  const raw = err instanceof Error ? err.message : typeof err === "string" ? err : "unknown backend error";
+  return raw.replace(/\s+/g, " ").replace(/(Bearer\s+)[^\s]+/gi, "$1[redacted]").replace(/([?&](?:api[_-]?key|token|secret|key)=)[^&\s]+/gi, "$1[redacted]").replace(/:\/\/[^/\s]+@/g, "://[redacted]@").trim().slice(0, 300);
+}
+async function reportBackendHealth(api2, log2, codeName, health) {
+  try {
+    await api2.post("/host/conversations/eval-backend-health", {
+      ok: health.ok,
+      error: health.error,
+      model: health.model
+    });
+  } catch (err) {
+    log2(`[conversation-eval] ${codeName}: backend-health report failed: ${err.message}`);
+  }
+}
 function readRecentTurns(dir, nowMs) {
   let entries;
   try {
@@ -6961,7 +6993,7 @@ var cachedMaintenanceWindow = null;
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
 var lastResponsivenessProbeAt = 0;
-var agtCliVersion = true ? "0.28.104" : "dev";
+var agtCliVersion = true ? "0.28.106" : "dev";
 function resolveBrewPath(execFileSync4) {
   try {
     const out = execFileSync4("which", ["brew"], { timeout: 5e3 }).toString().trim();
@@ -8072,7 +8104,7 @@ async function pollCycle() {
     }
     try {
       const { detectHostSecurity } = await import("../host-security-6PDFG7F5.js");
-      const { collectDiagnostics } = await import("../persistent-session-6CU3BIHP.js");
+      const { collectDiagnostics } = await import("../persistent-session-RMRG5HXI.js");
       const diagCodeNames = [...agentState.persistentSessionAgents];
       const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames) : void 0;
       let tailscaleHostname;
@@ -8173,7 +8205,7 @@ async function pollCycle() {
       const {
         collectResponsivenessProbes,
         getResponsivenessIntervalMs
-      } = await import("../responsiveness-probe-GLZZZ2OK.js");
+      } = await import("../responsiveness-probe-2KVIALGI.js");
       const probeIntervalMs = getResponsivenessIntervalMs();
       if (now - lastResponsivenessProbeAt > probeIntervalMs) {
         const probeCodeNames = [...agentState.persistentSessionAgents];
@@ -8205,7 +8237,7 @@ async function pollCycle() {
           collectResponsivenessProbes,
           livePendingInboundOldestAgeSeconds,
           parkPendingInbound
-        } = await import("../responsiveness-probe-GLZZZ2OK.js");
+        } = await import("../responsiveness-probe-2KVIALGI.js");
         const { getProjectDir: wedgeProjectDir } = await import("../claude-scheduler-FATCLHDM.js");
         const wedgeNow = /* @__PURE__ */ new Date();
         const liveAgents = agentState.persistentSessionAgents;
@@ -11314,7 +11346,7 @@ async function processDirectChatMessage(agent, msg) {
       }
     }
     const escapeXml = (value) => value.replaceAll("&", "&amp;").replaceAll("<", "&lt;").replaceAll(">", "&gt;").replaceAll('"', "&quot;").replaceAll("'", "&apos;");
-    const channelEnvelope = `<channel source="direct-chat" session_id="${escapeXml(msg.session_id)}" user="webapp">
+    const channelEnvelope = `<channel source="direct-chat" session_id="${escapeXml(msg.session_id)}" user="webapp" ${laneTagFragment("conversational", true)}>
 ${escapeXml(msg.content)}
 </channel>`;
     const delivered = await injectMessage(
@@ -11733,7 +11765,7 @@ async function processClaudePairSessions(agents) {
     killPairSession,
     pairTmuxSession,
     finalizeClaudePairOnboarding
-  } = await import("../claude-pair-runtime-JP3TEGDV.js");
+  } = await import("../claude-pair-runtime-6DZEDQAG.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));