@integrity-labs/agt-cli 0.28.572 → 0.28.573

package/dist/bin/agt.js

@@ -40,7 +40,7 @@ import {
   success,
   table,
   warn
-} from "../chunk-NG2MFOSH.js";
+} from "../chunk-GW2JI7NO.js";
 import {
   AnchorSessionClient,
   CHANNEL_REGISTRY,
@@ -71,7 +71,7 @@ import {
   requiredMcpWildcard,
   resolveChannels,
   serializeManifestForSlackCli
-} from "../chunk-72K46XUZ.js";
+} from "../chunk-5Z6RJ3RX.js";
 import "../chunk-XWVM4KPK.js";
 
 // src/bin/agt.ts
@@ -4853,7 +4853,7 @@ 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 ? "0.28.572" : "dev";
+var cliVersion = true ? "0.28.573" : "dev";
 async function fetchLatestVersion() {
   const host2 = getHost();
   if (!host2) return null;
@@ -6028,7 +6028,7 @@ function handleError(err) {
 }
 
 // src/bin/agt.ts
-var cliVersion2 = true ? "0.28.572" : "dev";
+var cliVersion2 = true ? "0.28.573" : "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-72K46XUZ.js → chunk-5Z6RJ3RX.js}

@@ -6112,6 +6112,28 @@ var SCOPE_DEFICIT_ERROR_PATTERNS = [
   "write_content",
   "sales channel is not enabled"
 ];
+var QUOTA_EXHAUSTED_ERROR_PATTERNS = [
+  "resource_exhausted",
+  "resource has been exhausted",
+  "rate limit",
+  "rate_limit",
+  "ratelimit",
+  "ratescope",
+  "rate_scope",
+  "too many requests",
+  "quota exceeded",
+  "quota_exceeded",
+  "quotaexceeded",
+  "quota error",
+  "quotaerror",
+  "quota_error",
+  "exceeded your quota",
+  "out of quota",
+  "insufficient quota",
+  "check quota",
+  "usage limit",
+  "daily limit exceeded"
+];
 function isReadonlyToolDescriptor(t) {
   if (!t?.name)
     return false;
@@ -6168,9 +6190,47 @@ function isScopeDeficitError(message) {
     return true;
   return m.includes("scope") && ["doesn't have", "does not have", "not have the", "not granted", "lacks the", "not authorized for"].some((p2) => m.includes(p2));
 }
+function isQuotaExhaustedError(message) {
+  const m = message.toLowerCase();
+  if (QUOTA_EXHAUSTED_ERROR_PATTERNS.some((p2) => m.includes(p2)))
+    return true;
+  if (/\b429\s+client error/.test(m))
+    return true;
+  return /(status(?:_?code)?|http_?status(?:_code)?|mercury_last_http_status_code|["'\\]code)["'\\\s]*[:=]["'\\\s]*429\b/.test(m);
+}
+function extractRetryAfterSeconds(message) {
+  const m = message.toLowerCase();
+  const patterns = [
+    /retry\s+in\s+(\d+)\s*(?:s\b|sec\b|secs\b|second)/,
+    /retry[-_ ]?after["'\\\s]*[:=]["'\\\s]*(\d+)/,
+    /retry_?delay["'\\\s]*[:=]["'\\\s]*"?(\d+)s?/
+  ];
+  for (const re of patterns) {
+    const hit = re.exec(m);
+    if (!hit?.[1])
+      continue;
+    const seconds = Number(hit[1]);
+    if (Number.isFinite(seconds) && seconds > 0)
+      return seconds;
+  }
+  return null;
+}
+function formatRetryWindow(seconds) {
+  if (seconds < 90)
+    return `${seconds} seconds`;
+  const minutes = seconds / 60;
+  if (minutes < 90)
+    return `about ${Math.round(minutes)} minutes`;
+  const hours = minutes / 60;
+  if (hours < 24)
+    return `about ${Math.round(hours * 10) / 10} hours`;
+  return `about ${Math.round(hours / 24 * 10) / 10} days`;
+}
 function classifyToolCallFailure(text) {
   if (isAccountResolutionError(text))
     return "account";
+  if (isQuotaExhaustedError(text))
+    return "quota";
   if (isScopeDeficitError(text))
     return "scope";
   if (isUpstreamAuthError(text))
@@ -6298,6 +6358,19 @@ async function probeComposioMcpToolCall(config, fetchImpl = fetch) {
           details: baseDetails
         };
       }
+      if (kind === "quota") {
+        const retryAfterSeconds = extractRetryAfterSeconds(failureText);
+        const retryPhrase = retryAfterSeconds ? ` The provider says to retry in ${formatRetryWindow(retryAfterSeconds)}.` : "";
+        return {
+          status: "degraded",
+          message: `Live tool call '${toolName}' reached the provider and was refused: a quota or rate limit is exhausted. The connection itself is valid \u2014 do NOT reconnect, this resolves when the window resets.${retryPhrase} ${snippet}`,
+          details: {
+            ...baseDetails,
+            reason: "quota_exhausted",
+            ...retryAfterSeconds ? { retry_after_seconds: retryAfterSeconds } : {}
+          }
+        };
+      }
       if (kind === "scope") {
         return {
           status: "degraded",
@@ -6319,7 +6392,11 @@ async function probeComposioMcpToolCall(config, fetchImpl = fetch) {
           details: { ...baseDetails, reason: "site_unresolved" }
         };
       }
-      return { status: "ok", message: `Live tool call '${toolName}' resolved the account (tool error: ${snippet})`, details: baseDetails };
+      return {
+        status: "ok",
+        message: `Live tool call '${toolName}' resolved the account (tool error: ${snippet})`,
+        details: { ...baseDetails, tool_error: snippet }
+      };
     }
     return { status: "ok", message: `Live tool call '${toolName}' resolved the connected account`, details: baseDetails };
   } catch (err) {
@@ -10439,7 +10516,7 @@ var FLAG_REGISTRY = [
   },
   {
     key: "direct-chat-per-user",
-    description: "Scope a STANDARD agent's Direct Chat to the user who had the conversation (ENG-8712). Today two teammates who open the same agent land in the same thread and read each other's messages: the /recent and per-session reads carry no user predicate unless the agent is the per-org system_support concierge. When on, both reads return only sessions the requester authored in PLUS sessions no user has authored in \u2014 the second arm is load-bearing, since every server-originated row (kanban and scheduled-task nudges, integration failures, HITL approvals, degradation alerts) is written as role=user with no user_id and would otherwise vanish from every console (ENG-8431). Does NOT make the thread private in the live direction: assistant replies still fan out on the per-TEAM realtime topic direct-chat-agent:{agentId}, so a teammate holding the drawer open still receives them. Narrowing that topic is follow-up work.",
+    description: "Scope a STANDARD agent's Direct Chat to the user who had the conversation (ENG-8712). Today two teammates who open the same agent land in the same thread and read each other's messages: the /recent and per-session reads carry no user predicate unless the agent is the per-org system_support concierge. When on, both reads return only sessions the requester authored in PLUS sessions no user has authored in \u2014 the second arm is load-bearing, since every server-originated row (kanban and scheduled-task nudges, integration failures, HITL approvals, degradation alerts) is written as role=user with no user_id and would otherwise vanish from every console (ENG-8431). Does NOT by itself make the thread private in the live direction: assistant replies fan out on the per-TEAM realtime topic direct-chat-agent:{agentId}, so a teammate holding the drawer open still receives them. That half is direct-chat-live-per-user (ENG-8760), which is a SEPARATE flag on purpose \u2014 turn both on for a private thread.",
     flagType: "boolean",
     // Declared safe value is `false` — today's team-wide thread. Server-side only
     // (the two reads in routes/agents.ts); the browser needs no knowledge of it,
@@ -10452,6 +10529,27 @@ var FLAG_REGISTRY = [
     // backfill in migration 20260812000006 is not reversed by the flag.
     defaultValue: false
   },
+  {
+    key: "direct-chat-live-per-user",
+    description: "The LIVE half of per-user Direct Chat (ENG-8760). direct-chat-per-user scoped the two REST reads; assistant replies kept fanning out on the per-TEAM realtime topic direct-chat-agent:{agentId}, whose RLS (20260804000003) has no session and no user predicate \u2014 so a teammate holding the drawer open still received another member's replies live, and only a reload hid them. When on, a reply in a session a human has authored in is broadcast to direct-chat-user:{agentId}:{userId} instead, whose policy (20260812000009) pins the third segment to auth.uid(). Sessions NO human has authored in \u2014 the agent's inbound work queue: kanban and scheduled-task nudges, integration failures, HITL approvals, degradation alerts \u2014 keep going to the per-TEAM topic, matching arm 2 of the REST read, or this would repeat ENG-8431 in the live lane. Also admits the system_support concierge to the per-user topic, which the per-TEAM topic denies outright (its live delivery is additive here, never narrowed).",
+    flagType: "boolean",
+    // Declared safe value is `false` — today's per-TEAM fan-out, byte for byte,
+    // for every agent kind including the concierge. Server-side only: the browser
+    // subscribes to BOTH topics unconditionally and needs no knowledge of the
+    // flag, so NOT public — which is also what keeps client and server from
+    // disagreeing about it. No envVar: an org/team-scoped rollout gate with no
+    // host-side consumer, and a declared envVar pinned in sst.config.ts would
+    // make every admin-UI override unreachable (ENG-8303).
+    //
+    // DELIBERATELY INDEPENDENT of direct-chat-per-user. The two are separable in
+    // both directions and neither ordering is unsafe: this one alone narrows live
+    // delivery below what the REST read still serves (a teammate can load the
+    // thread but stops receiving it live), and that one alone is the state
+    // ENG-8712 shipped. Coupling them into one switch would have removed the
+    // ability to soak the live change on an org that is already comfortable with
+    // the read change.
+    defaultValue: false
+  },
   {
     key: "onboarding-auto-deploy",
     description: `Auto-deploy the first agent during onboarding (ENG-7378): the agent edit page's Deploy & Test tab fires the existing deploy (draft -> active) automatically once the agent's host is fully deployed (active + manager heartbeating, not provisioning) AND Claude-authenticated (claude_auth_status=valid), replacing the manual "Deploy Agent" click in the recruit funnel. Only draft, non-system_support agents; a failed or timed-out attempt falls back to the manual button (no auto-retry loop).`,
@@ -15832,4 +15930,4 @@ export {
   stopAllSessionsAndWait,
   getProjectDir
 };
-//# sourceMappingURL=chunk-72K46XUZ.js.map
+//# sourceMappingURL=chunk-5Z6RJ3RX.js.map