@integrity-labs/agt-cli 0.28.84 → 0.28.86

package/dist/{chunk-XDZFMTY5.js → chunk-OZWBVM7M.js}

@@ -3,7 +3,7 @@ import {
   formatMissingVar,
   isClaudeFastMode,
   probeMcpEnvSubstitution
-} from "./chunk-3KLQA3SC.js";
+} from "./chunk-E5TGFEDQ.js";
 import {
   reapOrphanChannelMcps
 } from "./chunk-XWVM4KPK.js";
@@ -1395,4 +1395,4 @@ export {
   stopAllSessionsAndWait,
   getProjectDir
 };
-//# sourceMappingURL=chunk-XDZFMTY5.js.map
+//# sourceMappingURL=chunk-OZWBVM7M.js.map

package/dist/{claude-pair-runtime-6XYMSC6B.js → claude-pair-runtime-O5PHMHMF.js}

@@ -100,7 +100,7 @@ async function spawnPairSession(session) {
     return { ok: true };
   } catch {
   }
-  const { resolveClaudeBinary } = await import("./persistent-session-U6P5I6TT.js");
+  const { resolveClaudeBinary } = await import("./persistent-session-ZAKXWRX7.js");
   const claudeBin = resolveClaudeBinary();
   const pairEnv = {
     ...process.env,
@@ -373,4 +373,4 @@ export {
   startClaudePair,
   submitClaudePairCode
 };
-//# sourceMappingURL=claude-pair-runtime-6XYMSC6B.js.map
+//# sourceMappingURL=claude-pair-runtime-O5PHMHMF.js.map

package/dist/lib/manager-worker.js

@@ -28,7 +28,7 @@ import {
   requireHost,
   safeWriteJsonAtomic,
   setConfigHash
-} from "../chunk-EIXW7L6A.js";
+} from "../chunk-3EZHAJGB.js";
 import {
   getProjectDir as getProjectDir2,
   getReadyTasks,
@@ -66,7 +66,7 @@ import {
   takeWatchdogGiveUpCount,
   takeZombieDetection,
   transcriptActivityAgeSeconds
-} from "../chunk-XDZFMTY5.js";
+} from "../chunk-OZWBVM7M.js";
 import {
   FLAGS_SCHEMA_VERSION,
   FLAG_REGISTRY,
@@ -100,7 +100,7 @@ import {
   resolveDmTarget,
   sumTranscriptUsageInWindow,
   wrapScheduledTaskPrompt
-} from "../chunk-3KLQA3SC.js";
+} from "../chunk-E5TGFEDQ.js";
 import {
   parsePsRows,
   reapOrphanChannelMcps
@@ -2263,6 +2263,8 @@ var WINDOW_PAD_MS = 5 * 6e4;
 var MAX_TURN_CHARS = 1500;
 var MAX_TRANSCRIPT_CHARS = 6e3;
 var DEFAULT_CLAUDE_EVAL_MODEL = "claude-haiku-4-5-20251001";
+var DEFAULT_ANTHROPIC_MESSAGES_URL = "https://api.anthropic.com/v1/messages";
+var ANTHROPIC_API_VERSION = "2023-06-01";
 var DEFAULT_LOCAL_EVAL_URL = "http://localhost:11434/v1/chat/completions";
 var DEFAULT_LOCAL_EVAL_MODEL = "gemma4:12b";
 var EVAL_TIMEOUT_MS = 12e4;
@@ -2299,6 +2301,87 @@ async function runLocalEvalChat(prompt, opts) {
   }
   return text;
 }
+async function runAnthropicMessages(prompt, opts) {
+  if (!opts.apiKey) {
+    throw new Error("anthropic-api eval requires an API key");
+  }
+  const doFetch = opts.fetchImpl ?? fetch;
+  const res = await doFetch(opts.url ?? DEFAULT_ANTHROPIC_MESSAGES_URL, {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      "x-api-key": opts.apiKey,
+      "anthropic-version": ANTHROPIC_API_VERSION
+    },
+    body: JSON.stringify({
+      model: opts.model,
+      max_tokens: opts.maxTokens ?? 512,
+      temperature: 0,
+      messages: [{ role: "user", content: prompt }]
+    }),
+    signal: AbortSignal.timeout(opts.timeoutMs ?? EVAL_TIMEOUT_MS)
+  });
+  if (!res.ok) {
+    throw new Error(`anthropic messages api returned ${res.status}`);
+  }
+  const data = await res.json();
+  const text = (data.content ?? []).filter((b) => b?.type === "text" && typeof b.text === "string").map((b) => b.text).join("").trim();
+  if (!text) {
+    throw new Error("anthropic messages api returned no text content");
+  }
+  return text;
+}
+function selectEvalBackend(opts) {
+  const kind = (opts.kind ?? "").trim().toLowerCase();
+  const log2 = opts.log ?? (() => {
+  });
+  if (kind === "local") {
+    const { url, model, apiKey } = opts.local;
+    const safeUrl = (() => {
+      try {
+        const parsed = new URL(url);
+        return `${parsed.origin}${parsed.pathname}`;
+      } catch {
+        return "[invalid-url]";
+      }
+    })();
+    log2(`[conversation-eval] backend=local url=${safeUrl} model=${model}`);
+    return { model, run: (prompt) => runLocalEvalChat(prompt, { url, model, apiKey }) };
+  }
+  if (kind === "" || kind === "claude-p") {
+    log2(`[conversation-eval] backend=claude-p model=${opts.claudeModel}`);
+    return { model: opts.claudeModel, run: (prompt) => opts.claudePRunner(prompt, opts.claudeModel) };
+  }
+  if (kind === "anthropic-api") {
+    if (!opts.anthropicApiKey) {
+      log2(
+        "[conversation-eval] backend=anthropic-api but no API key (AGT_CONV_EVAL_ANTHROPIC_API_KEY / ANTHROPIC_API_KEY); evaluation disabled"
+      );
+      return {
+        model: "anthropic-api-missing-key",
+        run: async () => {
+          throw new Error(
+            "anthropic-api eval requires AGT_CONV_EVAL_ANTHROPIC_API_KEY or ANTHROPIC_API_KEY"
+          );
+        }
+      };
+    }
+    log2(`[conversation-eval] backend=anthropic-api model=${opts.claudeModel}`);
+    return {
+      model: opts.claudeModel,
+      run: (prompt) => runAnthropicMessages(prompt, { apiKey: opts.anthropicApiKey, model: opts.claudeModel })
+    };
+  }
+  log2(
+    `[conversation-eval] invalid AGT_CONV_EVAL_BACKEND='${kind}' - expected 'anthropic-api', 'claude-p', or 'local'; evaluation disabled`
+  );
+  return {
+    model: "invalid-conversation-eval-backend",
+    run: async () => {
+      throw new Error(`Unsupported AGT_CONV_EVAL_BACKEND='${kind}'. Expected 'anthropic-api', 'claude-p', or 'local'.`);
+    }
+  };
+}
 var state3 = /* @__PURE__ */ new Map();
 function channelRefTokens(channelRef) {
   return channelRef.split(":").slice(1).filter((p) => p && p !== "dm");
@@ -6803,7 +6886,7 @@ var cachedMaintenanceWindow = null;
 var lastVersionCheckAt = 0;
 var VERSION_CHECK_INTERVAL_MS = 5 * 60 * 1e3;
 var lastResponsivenessProbeAt = 0;
-var agtCliVersion = true ? "0.28.84" : "dev";
+var agtCliVersion = true ? "0.28.86" : "dev";
 function resolveBrewPath(execFileSync4) {
   try {
     const out = execFileSync4("which", ["brew"], { timeout: 5e3 }).toString().trim();
@@ -7423,37 +7506,40 @@ function memoryExtractionEnabled() {
   return hostFlagStore().getBoolean("memory-extraction");
 }
 var conversationEvalBackend = null;
+var conversationEvalBackendSig = null;
 function resolveConversationEvalBackend() {
-  if (conversationEvalBackend) return conversationEvalBackend;
-  const kind = (process.env["AGT_CONV_EVAL_BACKEND"] ?? "claude-p").trim().toLowerCase();
-  if (kind === "local") {
-    const url = process.env["AGT_CONV_EVAL_LOCAL_URL"]?.trim() || DEFAULT_LOCAL_EVAL_URL;
-    const model = process.env["AGT_CONV_EVAL_LOCAL_MODEL"]?.trim() || DEFAULT_LOCAL_EVAL_MODEL;
-    const apiKey = process.env["AGT_CONV_EVAL_LOCAL_API_KEY"]?.trim() || void 0;
-    const safeUrl = (() => {
-      try {
-        const parsed = new URL(url);
-        return `${parsed.origin}${parsed.pathname}`;
-      } catch {
-        return "[invalid-url]";
+  const kind = hostFlagStore().getString("conversation-eval-backend");
+  const claudeModel = process.env["AGT_CONV_EVAL_CLAUDE_MODEL"]?.trim() || DEFAULT_CLAUDE_EVAL_MODEL;
+  const anthropicApiKey = process.env["AGT_CONV_EVAL_ANTHROPIC_API_KEY"]?.trim() || process.env["ANTHROPIC_API_KEY"]?.trim() || "";
+  const local = {
+    url: process.env["AGT_CONV_EVAL_LOCAL_URL"]?.trim() || DEFAULT_LOCAL_EVAL_URL,
+    model: process.env["AGT_CONV_EVAL_LOCAL_MODEL"]?.trim() || DEFAULT_LOCAL_EVAL_MODEL,
+    // feature-gate-allow: model tunable, not a gate (matches the _MODE substring of _MODEL)
+    apiKey: process.env["AGT_CONV_EVAL_LOCAL_API_KEY"]?.trim() || void 0
+  };
+  const sig = `${kind}|${claudeModel}|${anthropicApiKey ? "key" : "nokey"}|${local.url}|${local.model}|${local.apiKey ? "lkey" : "nolkey"}`;
+  if (conversationEvalBackend && conversationEvalBackendSig === sig) return conversationEvalBackend;
+  try {
+    conversationEvalBackend = selectEvalBackend({
+      kind,
+      claudeModel,
+      anthropicApiKey,
+      local,
+      // The manager owns the spawn helpers; selectEvalBackend stays transport-free.
+      claudePRunner: runEvalClaude,
+      log
+    });
+  } catch (err) {
+    const message = err.message;
+    log(`[conversation-eval] backend unavailable, eval disabled: ${message}`);
+    conversationEvalBackend = {
+      model: `${kind || "default"}:unavailable`,
+      run: async () => {
+        throw new Error(message);
       }
-    })();
-    log(`[conversation-eval] backend=local url=${safeUrl} model=${model}`);
-    conversationEvalBackend = { model, run: (prompt) => runLocalEvalChat(prompt, { url, model, apiKey }) };
-    return conversationEvalBackend;
-  }
-  if (kind === "" || kind === "claude-p") {
-    const model = process.env["AGT_CONV_EVAL_CLAUDE_MODEL"]?.trim() || DEFAULT_CLAUDE_EVAL_MODEL;
-    conversationEvalBackend = { model, run: (prompt) => runEvalClaude(prompt, model) };
-    return conversationEvalBackend;
+    };
   }
-  log(`[conversation-eval] invalid AGT_CONV_EVAL_BACKEND='${kind}' \u2014 expected 'claude-p' or 'local'; evaluation disabled`);
-  conversationEvalBackend = {
-    model: "invalid-conversation-eval-backend",
-    run: async () => {
-      throw new Error(`Unsupported AGT_CONV_EVAL_BACKEND='${kind}'. Expected 'claude-p' or 'local'.`);
-    }
-  };
+  conversationEvalBackendSig = sig;
   return conversationEvalBackend;
 }
 function loadGatewayPorts() {
@@ -7911,7 +7997,7 @@ async function pollCycle() {
     }
     try {
       const { detectHostSecurity } = await import("../host-security-6PDFG7F5.js");
-      const { collectDiagnostics } = await import("../persistent-session-U6P5I6TT.js");
+      const { collectDiagnostics } = await import("../persistent-session-ZAKXWRX7.js");
       const diagCodeNames = [...agentState.persistentSessionAgents];
       const agentDiagnostics = diagCodeNames.length > 0 ? collectDiagnostics(diagCodeNames) : void 0;
       let tailscaleHostname;
@@ -8012,7 +8098,7 @@ async function pollCycle() {
       const {
         collectResponsivenessProbes,
         getResponsivenessIntervalMs
-      } = await import("../responsiveness-probe-VKIJY4IC.js");
+      } = await import("../responsiveness-probe-NKH4VYGR.js");
       const probeIntervalMs = getResponsivenessIntervalMs();
       if (now - lastResponsivenessProbeAt > probeIntervalMs) {
         const probeCodeNames = [...agentState.persistentSessionAgents];
@@ -8044,7 +8130,7 @@ async function pollCycle() {
           collectResponsivenessProbes,
           livePendingInboundOldestAgeSeconds,
           parkPendingInbound
-        } = await import("../responsiveness-probe-VKIJY4IC.js");
+        } = await import("../responsiveness-probe-NKH4VYGR.js");
         const { getProjectDir: wedgeProjectDir } = await import("../claude-scheduler-FATCLHDM.js");
         const wedgeNow = /* @__PURE__ */ new Date();
         const liveAgents = agentState.persistentSessionAgents;
@@ -11513,7 +11599,7 @@ async function processClaudePairSessions(agents) {
     killPairSession,
     pairTmuxSession,
     finalizeClaudePairOnboarding
-  } = await import("../claude-pair-runtime-6XYMSC6B.js");
+  } = await import("../claude-pair-runtime-O5PHMHMF.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));