switchroom 0.18.13 → 0.18.15

package/telegram-plugin/dist/gateway/gateway.js

@@ -19458,7 +19458,7 @@ function renderFallbackAnnouncement(input) {
   const tz = input.tz ?? "UTC";
   const lines = [];
   const limitWord = input.oldQuota ? limitWordFor(input.oldQuota) : "quota";
-  const headerLimit = limitWord === "quota" ? "quota cap" : `${limitWord} limit`;
+  const headerLimit = input.cause === "rate-limit" ? "rate limit" : limitWord === "quota" ? "quota cap" : `${limitWord} limit`;
   if (!input.newLabel) {
     lines.push(`\uD83D\uDD34 **All accounts blocked \u00b7 ${headerLimit} on ${escapeMarkdown(input.oldLabel)}**`);
     lines.push("");
@@ -19479,8 +19479,8 @@ function renderFallbackAnnouncement(input) {
         lines.push("");
         lines.push(`Earliest recovery: \`${codeSpanSafe(earliest.label)}\` ` + `${formatAbsolute(earliest.at, tz)} (in ${formatRelative(earliest.at, now)})`);
       }
-    } else if (input.oldQuota) {
-      const recovery = recoveryAtFor(input.oldQuota);
+    } else {
+      const recovery = (input.oldQuota ? recoveryAtFor(input.oldQuota) : null) ?? input.parsedResetAt ?? null;
       if (recovery) {
         lines.push(`${escapeMarkdown(input.oldLabel)} recovers ${formatAbsolute(recovery, tz)} ` + `(in ${formatRelative(recovery, now)})`);
       }
@@ -19495,8 +19495,8 @@ function renderFallbackAnnouncement(input) {
   lines.push(`\`${codeSpanSafe(input.oldLabel)}\` \u2192 \`${codeSpanSafe(input.newLabel)}\``);
   lines.push(`Triggered by: agent **${escapeMarkdown(input.triggerAgent)}**`);
   lines.push("");
-  if (input.oldQuota) {
-    const recovery = recoveryAtFor(input.oldQuota);
+  {
+    const recovery = (input.oldQuota ? recoveryAtFor(input.oldQuota) : null) ?? input.parsedResetAt ?? null;
     if (recovery) {
       lines.push(`\`${codeSpanSafe(input.oldLabel)}\` recovers ` + `${formatAbsolute(recovery, tz)} (in ${formatRelative(recovery, now)})`);
     }
@@ -19661,7 +19661,7 @@ function decodeResponse2(line) {
   }
   return ResponseSchema2.parse(parsed);
 }
-var MAX_FRAME_BYTES2, PROTOCOL_VERSION = 1, ProviderNameSchema, GetCredentialsRequestSchema, ListStateRequestSchema, SetActiveRequestSchema, MarkExhaustedRequestSchema, RefreshAccountRequestSchema, AnthropicCredentialsSchema, GoogleCredentialsSchema, MicrosoftCredentialsSchema, ProviderCredentialsSchema, AddAccountRequestSchema, RmAccountRequestSchema, SetOverrideRequestSchema, ListGoogleAccountsRequestSchema, ListMicrosoftAccountsRequestSchema, ProbeQuotaRequestSchema, ClaimNotificationRequestSchema, RequestSchema2, GetCredentialsDataSchema, AccountStateSchema, AgentStateSchema, ConsumerStateSchema, ListStateDataSchema, SetActiveDataSchema, MarkExhaustedDataSchema, RefreshAccountDataSchema, AddAccountDataSchema, RmAccountDataSchema, SetOverrideDataSchema, ClaimNotificationDataSchema, GoogleAccountStateSchema, ListGoogleAccountsDataSchema, MicrosoftAccountStateSchema, ListMicrosoftAccountsDataSchema, ErrorBodySchema, SuccessResponseSchema, ErrorResponseSchema2, ResponseSchema2;
+var MAX_FRAME_BYTES2, PROTOCOL_VERSION = 1, ProviderNameSchema, GetCredentialsRequestSchema, ListStateRequestSchema, SetActiveRequestSchema, MarkExhaustedRequestSchema, MarkThrottledRequestSchema, RefreshAccountRequestSchema, AnthropicCredentialsSchema, GoogleCredentialsSchema, MicrosoftCredentialsSchema, ProviderCredentialsSchema, AddAccountRequestSchema, RmAccountRequestSchema, SetOverrideRequestSchema, ListGoogleAccountsRequestSchema, ListMicrosoftAccountsRequestSchema, ProbeQuotaRequestSchema, ClaimNotificationRequestSchema, RequestSchema2, GetCredentialsDataSchema, AccountStateSchema, AgentStateSchema, ConsumerStateSchema, ListStateDataSchema, SetActiveDataSchema, MarkExhaustedDataSchema, MarkThrottledDataSchema, RefreshAccountDataSchema, AddAccountDataSchema, RmAccountDataSchema, SetOverrideDataSchema, ClaimNotificationDataSchema, GoogleAccountStateSchema, ListGoogleAccountsDataSchema, MicrosoftAccountStateSchema, ListMicrosoftAccountsDataSchema, ErrorBodySchema, SuccessResponseSchema, ErrorResponseSchema2, ResponseSchema2;
 var init_protocol2 = __esm(() => {
   init_zod();
   MAX_FRAME_BYTES2 = 64 * 1024;
@@ -19690,6 +19690,12 @@ var init_protocol2 = __esm(() => {
     id: exports_external.string().min(1),
     until: exports_external.number().int().positive().optional()
   });
+  MarkThrottledRequestSchema = exports_external.object({
+    v: exports_external.literal(PROTOCOL_VERSION),
+    op: exports_external.literal("mark-throttled"),
+    id: exports_external.string().min(1),
+    until: exports_external.number().int().positive()
+  });
   RefreshAccountRequestSchema = exports_external.object({
     v: exports_external.literal(PROTOCOL_VERSION),
     op: exports_external.literal("refresh-account"),
@@ -19790,6 +19796,7 @@ var init_protocol2 = __esm(() => {
     ListStateRequestSchema,
     SetActiveRequestSchema,
     MarkExhaustedRequestSchema,
+    MarkThrottledRequestSchema,
     RefreshAccountRequestSchema,
     AddAccountRequestSchema,
     RmAccountRequestSchema,
@@ -19809,6 +19816,7 @@ var init_protocol2 = __esm(() => {
     expiresAt: exports_external.number().optional(),
     exhausted: exports_external.boolean(),
     exhausted_until: exports_external.number().optional(),
+    throttled_until: exports_external.number().optional(),
     threshold_violations: exports_external.number().int().nonnegative().optional(),
     last_refreshed_at: exports_external.number().optional()
   });
@@ -19839,6 +19847,12 @@ var init_protocol2 = __esm(() => {
     rolled: exports_external.array(exports_external.string()),
     rolledTo: exports_external.string().nullable().optional()
   });
+  MarkThrottledDataSchema = exports_external.object({
+    account: exports_external.string(),
+    throttled_until: exports_external.number(),
+    escalated: exports_external.boolean(),
+    rolledTo: exports_external.string().nullable().optional()
+  });
   RefreshAccountDataSchema = exports_external.object({
     account: exports_external.string(),
     expiresAt: exports_external.number().optional()
@@ -20037,6 +20051,15 @@ class AuthBrokerClient {
     const data = await this.send(req);
     return data;
   }
+  async markThrottled(until) {
+    const data = await this.send({
+      v: PROTOCOL_VERSION,
+      id: randomUUID(),
+      op: "mark-throttled",
+      until
+    });
+    return data;
+  }
   async claimNotification(key, windowMs) {
     const data = await this.send({
       v: PROTOCOL_VERSION,
@@ -20474,7 +20497,6 @@ var init_schema = __esm(() => {
     enabled: exports_external.boolean().default(true).describe("Master switch for the per-agent Telegram gateway sidecar. " + "When false, start.sh skips the gateway supervise loop and the " + "agent boots without bot-token requirements (smoke-test + " + "offline-dev use case)."),
     plugin: exports_external.enum(["switchroom", "official"]).optional().describe("Which Telegram MCP plugin to load. Default is 'switchroom' \u2014 the " + "enhanced fork with streaming edits, reactions, history, and " + "access control. Set to 'official' for the upstream marketplace " + "plugin (basic send/receive only)."),
     format: exports_external.enum(["html", "markdownv2", "text"]).optional().describe("Default reply format passed to the plugin"),
-    rate_limit_ms: exports_external.number().optional().describe("Minimum delay between outgoing messages in ms"),
     stream_mode: exports_external.enum(["pty", "checklist"]).optional().describe("How live progress is streamed to Telegram during a turn. " + "'pty' (default) surfaces text snapshots of Claude Code's TUI \u2014 " + "compatible but can flicker as Ink re-renders. 'checklist' drives " + "a structured progress card from session-tail events \u2014 stable " + "order, per-tool status emojis, fires only on semantic transitions."),
     stream_throttle_ms: exports_external.number().int().nonnegative().optional().describe("Throttle window in ms between successive in-place stream edits " + "during a turn. Lower = more responsive stream, higher = fewer API " + "calls. Floored at 250 by draft-stream itself. Default 400 ms for DMs " + "and 1000 ms for groups/forums (respects Telegram's ~1 edit/sec/message " + "practical ceiling). Override per-agent if a particular agent needs " + "snappier or quieter streaming."),
     clear_status_on_completion: exports_external.boolean().optional().describe("When true, the live activity/status feed (the in-place 'what it's " + "doing' message \u2014 Reading X, Searching the web for Y, \u2026) is DELETED " + "when the turn's final answer lands, so only the reply remains. " + "Default false: the status message is left in the chat as a record " + "(its last step marked done) \u2014 no post-then-delete. Per-agent " + "override; cascades defaults \u2192 profile \u2192 agent (per-key)."),
@@ -28164,7 +28186,16 @@ function initHistory(stateDir, retentionDays = 30) {
     CREATE INDEX IF NOT EXISTS idx_messages_recent
       ON messages (chat_id, thread_id, ts DESC)
   `);
-  for (const column of ["reply_to_message_id INTEGER", "reply_to_text TEXT", "user_reaction TEXT"]) {
+  for (const column of [
+    "reply_to_message_id INTEGER",
+    "reply_to_text TEXT",
+    "user_reaction TEXT",
+    "forwarded_from TEXT",
+    "forwarded_from_type TEXT",
+    "forwarded_from_id TEXT",
+    "forwarded_date TEXT",
+    "forwarded_message_id INTEGER"
+  ]) {
     try {
       db.exec(`ALTER TABLE messages ADD COLUMN ${column}`);
     } catch (err) {
@@ -28264,10 +28295,10 @@ function recordInbound(args) {
     return;
   const stmt = requireDb().prepare(`
     INSERT OR REPLACE INTO messages
-      (chat_id, thread_id, message_id, role, user, user_id, ts, text, attachment_kind, group_id, reply_to_message_id, reply_to_text)
-    VALUES (?, ?, ?, 'user', ?, ?, ?, ?, ?, NULL, ?, ?)
+      (chat_id, thread_id, message_id, role, user, user_id, ts, text, attachment_kind, group_id, reply_to_message_id, reply_to_text, forwarded_from, forwarded_from_type, forwarded_from_id, forwarded_date, forwarded_message_id)
+    VALUES (?, ?, ?, 'user', ?, ?, ?, ?, ?, NULL, ?, ?, ?, ?, ?, ?, ?)
   `);
-  stmt.run(args.chat_id, args.thread_id ?? null, args.message_id, args.user ?? null, args.user_id ?? null, args.ts, redact(args.text), args.attachment_kind ?? null, args.reply_to_message_id ?? null, args.reply_to_text != null ? redact(args.reply_to_text) : args.reply_to_text ?? null);
+  stmt.run(args.chat_id, args.thread_id ?? null, args.message_id, args.user ?? null, args.user_id ?? null, args.ts, redact(args.text), args.attachment_kind ?? null, args.reply_to_message_id ?? null, args.reply_to_text != null ? redact(args.reply_to_text) : args.reply_to_text ?? null, args.forwarded_from != null ? redact(args.forwarded_from) : null, args.forwarded_from_type ?? null, args.forwarded_from_id ?? null, args.forwarded_date ?? null, args.forwarded_message_id ?? null);
 }
 function recordOutbound(args) {
   if (args.message_ids.length === 0)
@@ -36143,7 +36174,7 @@ function renderFallbackAnnouncement2(input) {
   const tz = input.tz ?? "UTC";
   const lines = [];
   const limitWord = input.oldQuota ? limitWordFor2(input.oldQuota) : "quota";
-  const headerLimit = limitWord === "quota" ? "quota cap" : `${limitWord} limit`;
+  const headerLimit = input.cause === "rate-limit" ? "rate limit" : limitWord === "quota" ? "quota cap" : `${limitWord} limit`;
   if (!input.newLabel) {
     lines.push(`\uD83D\uDD34 **All accounts blocked \u00b7 ${headerLimit} on ${escapeMarkdown(input.oldLabel)}**`);
     lines.push("");
@@ -36164,8 +36195,8 @@ function renderFallbackAnnouncement2(input) {
         lines.push("");
         lines.push(`Earliest recovery: \`${codeSpanSafe(earliest.label)}\` ` + `${formatAbsolute2(earliest.at, tz)} (in ${formatRelative2(earliest.at, now)})`);
       }
-    } else if (input.oldQuota) {
-      const recovery = recoveryAtFor2(input.oldQuota);
+    } else {
+      const recovery = (input.oldQuota ? recoveryAtFor2(input.oldQuota) : null) ?? input.parsedResetAt ?? null;
       if (recovery) {
         lines.push(`${escapeMarkdown(input.oldLabel)} recovers ${formatAbsolute2(recovery, tz)} ` + `(in ${formatRelative2(recovery, now)})`);
       }
@@ -36180,8 +36211,8 @@ function renderFallbackAnnouncement2(input) {
   lines.push(`\`${codeSpanSafe(input.oldLabel)}\` \u2192 \`${codeSpanSafe(input.newLabel)}\``);
   lines.push(`Triggered by: agent **${escapeMarkdown(input.triggerAgent)}**`);
   lines.push("");
-  if (input.oldQuota) {
-    const recovery = recoveryAtFor2(input.oldQuota);
+  {
+    const recovery = (input.oldQuota ? recoveryAtFor2(input.oldQuota) : null) ?? input.parsedResetAt ?? null;
     if (recovery) {
       lines.push(`\`${codeSpanSafe(input.oldLabel)}\` recovers ` + `${formatAbsolute2(recovery, tz)} (in ${formatRelative2(recovery, now)})`);
     }
@@ -39275,6 +39306,133 @@ function buildExtraAttachmentMeta(resolved) {
   return out;
 }
 
+// steering.ts
+function escapeXmlAttribute(s) {
+  return s.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;").replace(/'/g, "&apos;");
+}
+
+// gateway/forward-origin.ts
+var FORWARDED_FROM_NAME_MAX = 100;
+function truncateName(name) {
+  return name.length > FORWARDED_FROM_NAME_MAX ? name.slice(0, FORWARDED_FROM_NAME_MAX - 1) + "\u2026" : name;
+}
+function personName(parts) {
+  const name = [parts.first_name, parts.last_name].filter((p) => typeof p === "string" && p.length > 0).join(" ");
+  const handle = parts.username ? `(@${parts.username})` : "";
+  return [name, handle].filter((p) => p.length > 0).join(" ");
+}
+function chatTitle(chat) {
+  if (chat.title && chat.title.length > 0) {
+    const handle = chat.username ? `(@${chat.username})` : "";
+    return [chat.title, handle].filter((p) => p.length > 0).join(" ");
+  }
+  return personName(chat);
+}
+function parseForwardOrigin(origin) {
+  if (origin == null || typeof origin !== "object")
+    return;
+  const date = typeof origin.date === "number" ? origin.date : undefined;
+  switch (origin.type) {
+    case "user": {
+      const u = origin.sender_user;
+      if (u == null || typeof u !== "object")
+        return;
+      const name = personName(u);
+      if (name.length === 0) {
+        if (typeof u.id !== "number")
+          return;
+        return { name: String(u.id), type: "user", id: u.id, date };
+      }
+      return {
+        name: truncateName(name),
+        type: "user",
+        ...typeof u.id === "number" ? { id: u.id } : {},
+        date
+      };
+    }
+    case "hidden_user": {
+      const name = typeof origin.sender_user_name === "string" ? origin.sender_user_name : "";
+      if (name.length === 0)
+        return;
+      return { name: truncateName(name), type: "hidden_user", date };
+    }
+    case "chat": {
+      const c = origin.sender_chat;
+      if (c == null || typeof c !== "object")
+        return;
+      const name = chatTitle(c);
+      if (name.length === 0) {
+        if (typeof c.id !== "number")
+          return;
+        return { name: String(c.id), type: "chat", id: c.id, date };
+      }
+      return {
+        name: truncateName(name),
+        type: "chat",
+        ...typeof c.id === "number" ? { id: c.id } : {},
+        date
+      };
+    }
+    case "channel": {
+      const c = origin.chat;
+      if (c == null || typeof c !== "object")
+        return;
+      const name = chatTitle(c);
+      const messageId = typeof origin.message_id === "number" ? origin.message_id : undefined;
+      if (name.length === 0) {
+        if (typeof c.id !== "number")
+          return;
+        return { name: String(c.id), type: "channel", id: c.id, date, messageId };
+      }
+      return {
+        name: truncateName(name),
+        type: "channel",
+        ...typeof c.id === "number" ? { id: c.id } : {},
+        date,
+        messageId
+      };
+    }
+    default:
+      return;
+  }
+}
+function forwardOriginKey(o) {
+  return o.id != null ? `${o.type}:${o.id}` : `${o.type}:${o.name}`;
+}
+function dedupeForwardOrigins(origins) {
+  const seen = new Set;
+  const out = [];
+  for (const o of origins) {
+    if (o == null)
+      continue;
+    const key = forwardOriginKey(o);
+    if (seen.has(key))
+      continue;
+    seen.add(key);
+    out.push(o);
+  }
+  return out;
+}
+function buildForwardOriginMeta(origins) {
+  const out = {};
+  origins.forEach((o, i) => {
+    const suffix = i === 0 ? "" : `_${i + 1}`;
+    out[`forwarded_from${suffix}`] = escapeXmlAttribute(truncateName(o.name));
+    out[`forwarded_from_type${suffix}`] = o.type;
+    if (o.id != null)
+      out[`forwarded_from_id${suffix}`] = String(o.id);
+    if (o.date != null) {
+      out[`forwarded_date${suffix}`] = new Date(o.date * 1000).toISOString();
+    }
+  });
+  return out;
+}
+function forwardOriginDateIso(o) {
+  if (o?.date == null)
+    return null;
+  return new Date(o.date * 1000).toISOString();
+}
+
 // status-reactions.ts
 var TELEGRAM_REACTION_WHITELIST = new Set([
   "\uD83D\uDC4D",
@@ -42019,6 +42177,7 @@ function createAuthBrokerClient() {
     listState: () => broker.listState(),
     setActive: (label) => broker.setActive(label),
     markExhausted: (until) => broker.markExhausted(until),
+    markThrottled: (until) => broker.markThrottled(until),
     rmAccount: (label) => broker.rmAccount(label),
     refreshAccount: (label) => broker.refreshAccount(label),
     setOverride: (agent, account) => broker.setOverride(agent, account),
@@ -54665,9 +54824,9 @@ function renderSafe(doc, source, maxLen = RICH_MESSAGE_MAX_CHARS) {
 var PLAIN_TEXT_MAX_CHARS = 4096;
 function parseRichRenderEnabled(raw) {
   if (raw == null)
-    return false;
+    return true;
   const v = raw.trim().toLowerCase();
-  return v === "1" || v === "true" || v === "on" || v === "yes";
+  return !(v === "0" || v === "false" || v === "off" || v === "no");
 }
 function richRenderEnabled(env = process.env) {
   return parseRichRenderEnabled(env.SWITCHROOM_RICH_RENDER);
@@ -62755,6 +62914,7 @@ function createAuthBrokerClient2() {
     listState: () => broker.listState(),
     setActive: (label) => broker.setActive(label),
     markExhausted: (until) => broker.markExhausted(until),
+    markThrottled: (until) => broker.markThrottled(until),
     rmAccount: (label) => broker.rmAccount(label),
     refreshAccount: (label) => broker.refreshAccount(label),
     setOverride: (agent, account) => broker.setOverride(agent, account),
@@ -63161,6 +63321,15 @@ class AuthBrokerClient2 {
     const data = await this.send(req);
     return data;
   }
+  async markThrottled(until) {
+    const data = await this.send({
+      v: PROTOCOL_VERSION,
+      id: randomUUID4(),
+      op: "mark-throttled",
+      until
+    });
+    return data;
+  }
   async claimNotification(key, windowMs) {
     const data = await this.send({
       v: PROTOCOL_VERSION,
@@ -64205,7 +64374,7 @@ function parseSteerPrefix(body) {
     return { steering: false, body };
   return { steering: true, body: m[2].trim() };
 }
-function escapeXmlAttribute(s) {
+function escapeXmlAttribute2(s) {
   return s.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;").replace(/'/g, "&apos;");
 }
 function decodeXmlEntities(s) {
@@ -64216,13 +64385,13 @@ function formatPriorAssistantPreview(text4, maxChars = 200) {
   const decoded = decodeXmlEntities(stripped);
   const collapsed = decoded.replace(/\s+/g, " ").trim();
   const truncated = collapsed.length > maxChars ? collapsed.slice(0, maxChars) : collapsed;
-  return escapeXmlAttribute(truncated);
+  return escapeXmlAttribute2(truncated);
 }
 function formatReplyToText(text4, maxChars) {
   if (text4 == null)
     return;
   const truncated = text4.length > maxChars ? text4.slice(0, maxChars - 1) + "\u2026" : text4;
-  return escapeXmlAttribute(truncated);
+  return escapeXmlAttribute2(truncated);
 }
 
 // gateway/auto-classify-mid-turn.ts
@@ -64471,6 +64640,26 @@ var transientUpstreamSignals = [
   "would exceed your account\u2019s rate limit",
   "would exceed your account's rate limit"
 ];
+var litellmProxyLocal429Signals = [
+  "deployment over user-defined ratelimit",
+  "model rate limit exceeded. tpm limit",
+  "model rate limit exceeded. rpm limit",
+  "deployment over defined rpm limit",
+  "no deployments available for selected model",
+  "litellm rate limit handler",
+  "crossed tpm / rpm",
+  "max parallel request limit reached"
+];
+var litellmV3LimiterSignalPair = ["rate limit exceeded for ", "limit type:"];
+function isLitellmProxyLocal429(text4) {
+  if (typeof text4 !== "string" || text4.length === 0)
+    return false;
+  const sample = text4.length > 16384 ? text4.slice(0, 16384) : text4;
+  const lower = sample.toLowerCase();
+  if (litellmProxyLocal429Signals.some((s) => lower.includes(s)))
+    return true;
+  return litellmV3LimiterSignalPair.every((s) => lower.includes(s));
+}
 function detectModelUnavailable(stderr) {
   if (typeof stderr !== "string" || stderr.length === 0)
     return null;
@@ -64480,6 +64669,10 @@ function detectModelUnavailable(stderr) {
     const resetAt = parseResetTime(sample);
     return resetAt !== undefined ? { kind: "overload", resetAt, raw: stderr } : { kind: "overload", raw: stderr };
   }
+  if (isLitellmProxyLocal429(sample)) {
+    const resetAt = parseResetTime(sample);
+    return resetAt !== undefined ? { kind: "overload", resetAt, raw: stderr } : { kind: "overload", raw: stderr };
+  }
   const quotaSignals = [
     "out of extra usage",
     "extra usage",
@@ -64684,6 +64877,8 @@ function formatReason(d, now) {
       return `quota exhausted${reset2}`;
     case "overload":
       return `model overloaded${reset2}`;
+    case "rate_limited":
+      return `account rate-limited${reset2}`;
     case "network":
       return "network unreachable";
   }
@@ -64703,6 +64898,433 @@ function resolveModelUnavailableFromOperatorEvent(ev) {
   return detectModelUnavailable(detail);
 }
 
+// throttle-tier.ts
+init_card_format();
+init_quota_check();
+
+// model-unavailable.ts
+init_quota_check();
+init_card_format();
+var transientUpstreamSignals2 = [
+  "not your usage limit",
+  "not your account",
+  "not your account's",
+  "temporarily limiting requests",
+  "temporarily rate",
+  "server is temporarily",
+  "would exceed your account\u2019s rate limit",
+  "would exceed your account's rate limit"
+];
+function isTransientUpstreamSignal(text4) {
+  if (typeof text4 !== "string" || text4.length === 0)
+    return false;
+  const sample = text4.length > 16384 ? text4.slice(0, 16384) : text4;
+  const lower = sample.toLowerCase();
+  return transientUpstreamSignals2.some((s) => lower.includes(s));
+}
+var litellmProxyLocal429Signals2 = [
+  "deployment over user-defined ratelimit",
+  "model rate limit exceeded. tpm limit",
+  "model rate limit exceeded. rpm limit",
+  "deployment over defined rpm limit",
+  "no deployments available for selected model",
+  "litellm rate limit handler",
+  "crossed tpm / rpm",
+  "max parallel request limit reached"
+];
+var litellmV3LimiterSignalPair2 = ["rate limit exceeded for ", "limit type:"];
+function isLitellmProxyLocal4292(text4) {
+  if (typeof text4 !== "string" || text4.length === 0)
+    return false;
+  const sample = text4.length > 16384 ? text4.slice(0, 16384) : text4;
+  const lower = sample.toLowerCase();
+  if (litellmProxyLocal429Signals2.some((s) => lower.includes(s)))
+    return true;
+  return litellmV3LimiterSignalPair2.every((s) => lower.includes(s));
+}
+function parseLitellmLimitDetail(text4, parseTimeNow = new Date) {
+  const empty2 = { limitType: null, limit: null, currentUsage: null, resetAtMs: null };
+  if (typeof text4 !== "string" || text4.length === 0)
+    return empty2;
+  const sample = text4.length > 16384 ? text4.slice(0, 16384) : text4;
+  const lower = sample.toLowerCase();
+  let limitType = null;
+  let limit = null;
+  const eqLimit = lower.match(/\b([tr]pm)[ _]limit[=:]\s*(\d+)/);
+  if (eqLimit) {
+    limitType = eqLimit[1];
+    limit = Number(eqLimit[2]);
+  }
+  if (limitType == null) {
+    const v3Type = lower.match(/limit type:\s*(tokens|requests|max_parallel_requests)/);
+    if (v3Type)
+      limitType = v3Type[1];
+  }
+  if (limit == null) {
+    const v3Limit = lower.match(/current limit:\s*(\d+)/);
+    if (v3Limit)
+      limit = Number(v3Limit[1]);
+  }
+  let currentUsage = null;
+  const usage = lower.match(/current usage=(\d+)/);
+  if (usage)
+    currentUsage = Number(usage[1]);
+  let resetAtMs = null;
+  const resetsAt = sample.match(/limit resets at:\s*(\d{4}-\d{2}-\d{2}) (\d{2}:\d{2}:\d{2}) UTC/i);
+  if (resetsAt) {
+    const d = new Date(`${resetsAt[1]}T${resetsAt[2]}Z`);
+    if (!Number.isNaN(d.getTime()))
+      resetAtMs = d.getTime();
+  }
+  if (resetAtMs == null) {
+    const tryAgain = lower.match(/try again in\s+(\d+(?:\.\d+)?)\s*seconds/);
+    if (tryAgain) {
+      const secs = Number(tryAgain[1]);
+      if (Number.isFinite(secs) && secs > 0 && secs < 7 * 24 * 3600) {
+        resetAtMs = parseTimeNow.getTime() + Math.round(secs * 1000);
+      }
+    }
+  }
+  return {
+    limitType,
+    limit: limit != null && Number.isFinite(limit) ? limit : null,
+    currentUsage: currentUsage != null && Number.isFinite(currentUsage) ? currentUsage : null,
+    resetAtMs
+  };
+}
+function parseResetTime2(text4, parseTimeNow = new Date) {
+  const lower = text4.toLowerCase();
+  const retryAfter = lower.match(/retry[\s-]*after[:\s]+(\d+)\s*(seconds?|s\b|minutes?|m\b|hours?|h\b)?/);
+  if (retryAfter) {
+    const n = Number(retryAfter[1]);
+    if (Number.isFinite(n) && n > 0 && n < 7 * 24 * 3600) {
+      const unit = (retryAfter[2] ?? "seconds").toLowerCase();
+      const ms = unit.startsWith("h") ? n * 3600000 : unit.startsWith("m") ? n * 60000 : n * 1000;
+      return new Date(parseTimeNow.getTime() + ms);
+    }
+  }
+  const relReset = lower.match(/resets?\s+in\s+([0-9hms\s]+)/);
+  if (relReset) {
+    const ms = parseRelativeDuration2(relReset[1]);
+    if (ms != null)
+      return new Date(parseTimeNow.getTime() + ms);
+  }
+  const iso = text4.match(/\b\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(?:\.\d+)?(?:Z|[+-]\d{2}:?\d{2})?\b/);
+  if (iso) {
+    const d = new Date(iso[0]);
+    if (!Number.isNaN(d.getTime()))
+      return d;
+  }
+  const calReset = text4.match(/resets?\s+(?:at\s+)?([A-Z][a-z]{2,8}\s+\d{1,2}(?:,?\s*(?:\d{1,2}(?::\d{2})?\s*(?:am|pm|AM|PM)?))?)/);
+  if (calReset) {
+    const candidate = `${calReset[1]} ${parseTimeNow.getUTCFullYear()}`;
+    const d = new Date(candidate);
+    if (!Number.isNaN(d.getTime()))
+      return d;
+  }
+  const timeOnly = text4.match(/resets?\s+(?:at\s+)?(?!(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)[a-z]*\b)(\d{1,2})(?::(\d{2}))?\s*(am|pm)?\s*(?:\(([^)]+)\))?/i);
+  if (timeOnly) {
+    const d = resolveNextWallClock2(Number(timeOnly[1]), timeOnly[2] ? Number(timeOnly[2]) : 0, timeOnly[3]?.toLowerCase(), timeOnly[4]?.trim(), parseTimeNow);
+    if (d != null)
+      return d;
+  }
+  return;
+}
+function resolveNextWallClock2(hour12or24, minute, ampm, tz, nowDate) {
+  let hour = hour12or24;
+  if (ampm === "pm" && hour < 12)
+    hour += 12;
+  if (ampm === "am" && hour === 12)
+    hour = 0;
+  if (!Number.isFinite(hour) || hour > 23 || hour < 0)
+    return;
+  if (!Number.isFinite(minute) || minute > 59 || minute < 0)
+    return;
+  const nowMs2 = nowDate.getTime();
+  const base = new Date(nowMs2);
+  for (let dayOffset = 0;dayOffset <= 2; dayOffset++) {
+    const dateParts = tzDateParts2(new Date(nowMs2 + dayOffset * 86400000), tz);
+    if (dateParts == null)
+      return;
+    const epoch = wallClockToEpoch2(dateParts.year, dateParts.month, dateParts.day, hour, minute, tz);
+    if (epoch != null && epoch > nowMs2)
+      return new Date(epoch);
+  }
+  return;
+}
+function tzDateParts2(d, tz) {
+  if (!tz) {
+    return { year: d.getUTCFullYear(), month: d.getUTCMonth(), day: d.getUTCDate() };
+  }
+  try {
+    const fmt = new Intl.DateTimeFormat("en-US", {
+      timeZone: tz,
+      year: "numeric",
+      month: "2-digit",
+      day: "2-digit"
+    });
+    const parts = Object.fromEntries(fmt.formatToParts(d).filter((p) => p.type !== "literal").map((p) => [p.type, p.value]));
+    return {
+      year: Number(parts.year),
+      month: Number(parts.month) - 1,
+      day: Number(parts.day)
+    };
+  } catch {
+    return null;
+  }
+}
+function wallClockToEpoch2(year, month, day, hour, minute, tz) {
+  const asUtc = Date.UTC(year, month, day, hour, minute, 0);
+  if (!tz)
+    return asUtc;
+  try {
+    const fmt = new Intl.DateTimeFormat("en-US", {
+      timeZone: tz,
+      year: "numeric",
+      month: "2-digit",
+      day: "2-digit",
+      hour: "2-digit",
+      minute: "2-digit",
+      second: "2-digit",
+      hour12: false
+    });
+    const parts = Object.fromEntries(fmt.formatToParts(new Date(asUtc)).filter((p) => p.type !== "literal").map((p) => [p.type, p.value]));
+    const shown = Date.UTC(Number(parts.year), Number(parts.month) - 1, Number(parts.day), Number(parts.hour) % 24, Number(parts.minute), Number(parts.second));
+    const offset = shown - asUtc;
+    return asUtc - offset;
+  } catch {
+    return null;
+  }
+}
+function parseRelativeDuration2(s) {
+  let total = 0;
+  let matched = false;
+  const re = /(\d+)\s*(h|hours?|m|minutes?|s|seconds?)/g;
+  let m;
+  while ((m = re.exec(s)) != null) {
+    matched = true;
+    const n = Number(m[1]);
+    const unit = m[2].toLowerCase();
+    if (unit.startsWith("h"))
+      total += n * 3600000;
+    else if (unit.startsWith("m"))
+      total += n * 60000;
+    else
+      total += n * 1000;
+  }
+  return matched && total > 0 ? total : null;
+}
+
+// throttle-tier.ts
+var accountScopedThrottleSignals = [
+  "would exceed your account's rate limit",
+  "would exceed your account\u2019s rate limit",
+  "not your account"
+];
+function isAccountScopedThrottle(text4) {
+  if (typeof text4 !== "string" || text4.length === 0)
+    return false;
+  const sample = text4.length > 16384 ? text4.slice(0, 16384) : text4;
+  const lower = sample.toLowerCase();
+  return accountScopedThrottleSignals.some((s) => lower.includes(s));
+}
+function classify429Detail(text4) {
+  if (isAccountScopedThrottle(text4))
+    return "account-scoped";
+  if (isLitellmProxyLocal4292(text4))
+    return "litellm-local";
+  return "generic-transient";
+}
+function build429ClassifiedMetric(opts) {
+  const detail = typeof opts.detail === "string" ? opts.detail : "";
+  const litellm = parseLitellmLimitDetail(detail, new Date(opts.now));
+  const anthropicResetMs = parseResetTime2(detail, new Date(opts.now))?.getTime() ?? null;
+  const resetAtMs = anthropicResetMs ?? litellm.resetAtMs;
+  return {
+    kind: "rate_limit_429_classified",
+    agent: opts.agent,
+    classification: opts.classification,
+    action: opts.action,
+    reset_at_ms: resetAtMs,
+    reset_in_ms: resetAtMs != null ? Math.max(0, resetAtMs - opts.now) : null,
+    limit_type: litellm.limitType,
+    limit: litellm.limit,
+    current_usage: litellm.currentUsage
+  };
+}
+var THROTTLE_RETRY_IN_PLACE_MAX_MS_DEFAULT = 5 * 60000;
+var THROTTLE_DEFAULT_WAIT_MS = 60000;
+function throttleRetryInPlaceMaxMs(env = process.env) {
+  const raw = env.SWITCHROOM_THROTTLE_RETRY_IN_PLACE_MAX_MS;
+  if (raw == null || raw === "")
+    return THROTTLE_RETRY_IN_PLACE_MAX_MS_DEFAULT;
+  const n = Number(raw);
+  return Number.isFinite(n) && n > 0 ? n : THROTTLE_RETRY_IN_PLACE_MAX_MS_DEFAULT;
+}
+function decideThrottleTier(opts) {
+  const { detail, now, thresholdMs } = opts;
+  if (!isAccountScopedThrottle(detail))
+    return { action: "none" };
+  const resetAt = parseResetTime2(detail, new Date(now));
+  const resetAtMs = resetAt?.getTime();
+  if (resetAtMs == null || !Number.isFinite(resetAtMs) || resetAtMs <= now) {
+    return {
+      action: "throttle",
+      throttledUntilMs: now + THROTTLE_DEFAULT_WAIT_MS,
+      resetParsed: false
+    };
+  }
+  if (resetAtMs - now <= thresholdMs) {
+    return { action: "throttle", throttledUntilMs: resetAtMs, resetParsed: true };
+  }
+  return { action: "failover", resetAtMs };
+}
+var THROTTLE_NOTICE_COOLDOWN_MS = 10 * 60000;
+
+// throttle-tier.ts
+init_card_format();
+init_quota_check();
+var THROTTLE_RETRY_IN_PLACE_MAX_MS_DEFAULT2 = 5 * 60000;
+var THROTTLE_NOTICE_COOLDOWN_MS2 = 10 * 60000;
+function evaluateThrottleNotice(prev, account, now, cooldownMs = THROTTLE_NOTICE_COOLDOWN_MS2) {
+  const last = prev.lastSentAtMsByAccount[account] ?? 0;
+  if (now - last >= cooldownMs) {
+    return {
+      send: true,
+      next: {
+        lastSentAtMsByAccount: {
+          ...prev.lastSentAtMsByAccount,
+          [account]: now
+        }
+      }
+    };
+  }
+  return { send: false, next: prev };
+}
+function renderThrottleNotice(opts) {
+  const now = opts.now ?? new Date;
+  const resetStr = formatResetRelative(new Date(opts.throttledUntilMs), now);
+  const acct = opts.account ? `\`${escapeMarkdown(opts.account)}\`` : "the active account";
+  const lines = [
+    `\uD83D\uDEA6 **Rate-limited, staying put** \u2014 ${acct} hit a transient rate limit on **${escapeMarkdown(opts.agent)}**.`,
+    `This is a short throttle, not a quota wall \u2014 ${opts.resetParsed ? resetStr : `no reset given, retrying in ~60s`}.`,
+    `_Staying on ${acct}; no failover needed. The turn retries automatically after the reset._`
+  ];
+  return lines.join(`
+`);
+}
+function renderThrottleEscalationNotice(opts) {
+  const acct = opts.account ? `\`${escapeMarkdown(opts.account)}\`` : "the active account";
+  const head = `\u26d4\ufe0f **Rate limit was actually a wall** \u2014 repeated 429s on ${acct} ` + `(trigger: **${escapeMarkdown(opts.agent)}**) were corroborated by a live quota probe.`;
+  const tail = opts.rolledTo ? `Marked exhausted and rolled to \`${escapeMarkdown(opts.rolledTo)}\`.` : `Marked exhausted \u2014 no fallback account had quota (all blocked). ` + `Use \`/auth add <label>\` to attach another subscription.`;
+  return `${head}
+${tail}`;
+}
+
+// gateway/throttle-tier-wiring.ts
+var THROTTLE_RETRY_NUDGE_SLACK_MS = 5000;
+var THROTTLE_RETRY_NUDGE_JITTER_MAX_MS = 30000;
+function createThrottleTierRunner(deps) {
+  const now = deps.now ?? (() => Date.now());
+  const schedule = deps.schedule ?? ((fn, ms) => {
+    const t = setTimeout(fn, ms);
+    if (typeof t.unref === "function")
+      t.unref();
+    return { cancel: () => clearTimeout(t) };
+  });
+  const jitterMs = deps.jitterMs ?? (() => Math.floor(Math.random() * THROTTLE_RETRY_NUDGE_JITTER_MAX_MS));
+  let noticeState = { lastSentAtMsByAccount: {} };
+  let pendingNudge = null;
+  async function broadcastDeduped(client3, keyPrefix, account, markdown) {
+    for (const chatId of deps.listNoticeChats()) {
+      let granted = true;
+      if (client3 && account) {
+        try {
+          granted = (await client3.claimNotification(`${keyPrefix}:${account}:${chatId}`, THROTTLE_NOTICE_COOLDOWN_MS2)).granted;
+        } catch {
+          granted = true;
+        }
+      }
+      if (granted) {
+        deps.sendNotice(chatId, markdown);
+      } else {
+        deps.log(`[throttle-tier] notice suppressed (fleet claim) chat=${chatId}`);
+      }
+    }
+  }
+  function nudgeResume(reason, armedAtMs) {
+    const newest = deps.newestActiveTurnStartedAtMs();
+    if (deps.turnInFlight()) {
+      if (newest != null && newest > armedAtMs) {
+        deps.log(`[throttle-tier] resume skipped (superseded by a live newer turn) reason=${reason}`);
+        return;
+      }
+      deps.log(`[throttle-tier] resume deferred to turn-complete reason=${reason}`);
+      deps.deferRestartToTurnComplete(deps.agentName, reason);
+      return;
+    }
+    const verdict = deps.resumeDecide(newest);
+    if (verdict === "resume") {
+      deps.log(`[throttle-tier] resuming dead turn via self-restart reason=${reason}`);
+      deps.restartNow(deps.agentName, reason);
+    } else {
+      deps.log(`[throttle-tier] resume suppressed (${verdict}) reason=${reason}`);
+    }
+  }
+  async function fire(triggerAgent, throttledUntilMs, resetParsed) {
+    const armedAtMs = now();
+    let client3 = null;
+    let account = null;
+    let escalated = false;
+    let rolledTo = null;
+    try {
+      client3 = await deps.getBrokerClient();
+      if (client3) {
+        const r = await client3.markThrottled(throttledUntilMs);
+        account = r.account;
+        escalated = r.escalated;
+        rolledTo = r.rolledTo ?? null;
+      } else {
+        deps.log(`[throttle-tier] broker unreachable \u2014 notice only, no ledger record agent=${triggerAgent}`);
+      }
+    } catch (err) {
+      deps.log(`[throttle-tier] markThrottled failed agent=${triggerAgent}: ${err?.message ?? err}`);
+    }
+    if (escalated) {
+      deps.log(`[throttle-tier] escalated to wall account=${account ?? "?"} ` + `rolledTo=${rolledTo ?? "none (all blocked)"}`);
+      await broadcastDeduped(client3, "throttle-escalation", account, renderThrottleEscalationNotice({ account, agent: triggerAgent, rolledTo }));
+      if (rolledTo)
+        nudgeResume("throttle-escalation-resume", armedAtMs);
+      return;
+    }
+    const cooldownKey = account ?? `agent:${triggerAgent}`;
+    const verdict = evaluateThrottleNotice(noticeState, cooldownKey, now());
+    if (verdict.send) {
+      noticeState = verdict.next;
+      await broadcastDeduped(client3, "throttle-notice", account, renderThrottleNotice({
+        account,
+        agent: triggerAgent,
+        throttledUntilMs,
+        resetParsed,
+        now: new Date(now())
+      }));
+    } else {
+      deps.log(`[throttle-tier] notice suppressed (cooldown) key=${cooldownKey}`);
+    }
+    const delayMs = Math.max(throttledUntilMs - now(), 0) + THROTTLE_RETRY_NUDGE_SLACK_MS + jitterMs();
+    if (pendingNudge)
+      pendingNudge.cancel();
+    pendingNudge = schedule(() => {
+      pendingNudge = null;
+      nudgeResume("throttle-retry-resume", armedAtMs);
+    }, delayMs);
+  }
+  return {
+    fire,
+    inspect: () => ({ noticeState, nudgePending: pendingNudge != null })
+  };
+}
+
 // auto-fallback-fleet.ts
 init_card_format();
 init_auth_snapshot_format();
@@ -64738,7 +65360,7 @@ async function runFleetAutoFallback(deps) {
     };
   }
   const oldHealth = classifyHealth(oldSnap, now);
-  if (oldHealth === "healthy") {
+  if (oldHealth === "healthy" && !deps.rateLimitTrigger) {
     return {
       kind: "no-eligible-target",
       oldLabel: oldSnap.label,
@@ -64759,6 +65381,8 @@ async function runFleetAutoFallback(deps) {
         newQuota: null,
         triggerAgent: deps.triggerAgent,
         fleetSnapshots: snapshots,
+        parsedResetAt: deps.parsedResetAt ?? null,
+        cause: deps.rateLimitTrigger ? "rate-limit" : undefined,
         tz,
         now
       })
@@ -64777,6 +65401,8 @@ async function runFleetAutoFallback(deps) {
       newLabel: rolledTo,
       newQuota,
       triggerAgent: deps.triggerAgent,
+      parsedResetAt: deps.parsedResetAt ?? null,
+      cause: deps.rateLimitTrigger ? "rate-limit" : undefined,
       tz,
       now
     })
@@ -66830,13 +67456,13 @@ function buildFolderPickerCard(input) {
     validateDriveId(folder.id, "folder.id");
     rows.push([
       {
-        text: `\u2705 Allow "${truncateName(folder.name, 48)}"`,
+        text: `\u2705 Allow "${truncateName2(folder.name, 48)}"`,
         callback_data: encodeCallback(["drvpick", "grant", input.agent, folder.id])
       }
     ]);
     rows.push([
       {
-        text: `\uD83D\uDCC2 Browse "${truncateName(folder.name, 46)}"`,
+        text: `\uD83D\uDCC2 Browse "${truncateName2(folder.name, 46)}"`,
         callback_data: encodeCallback(["drvpick", "enter", input.agent, folder.id])
       }
     ]);
@@ -66885,14 +67511,14 @@ function buildFolderPickerCard(input) {
   return { body, rows };
 }
 function renderBreadcrumb(trail2, leafName) {
-  const leaf = truncateName(leafName, 24);
+  const leaf = truncateName2(leafName, 24);
   if (trail2.length === 0)
     return `/${leaf}`;
-  const segments = trail2.map((t) => truncateName(t.name, 20));
+  const segments = trail2.map((t) => truncateName2(t.name, 20));
   segments.push(leaf);
   return `/${segments.join("/")}`;
 }
-function truncateName(name, max) {
+function truncateName2(name, max) {
   if (name.length <= max)
     return name;
   return `${name.slice(0, max - 1)}\u2026`;
@@ -68918,6 +69544,11 @@ function extractConfirmation(pane) {
 // ../src/agents/scaffold.ts
 import { join as join32, resolve as resolve6 } from "node:path";
 init_atomic();
+
+// ../src/agents/agent-uid.ts
+init_peercred();
+
+// ../src/agents/scaffold.ts
 init_schema();
 
 // ../src/config/users.ts
@@ -76405,27 +77036,6 @@ function getNestedObj(obj, key) {
 var DEFAULT_OPERATOR_EVENT_COOLDOWN_MS2 = 5 * 60000;
 var cooldownMap2 = new Map;
 
-// model-unavailable.ts
-init_quota_check();
-init_card_format();
-var transientUpstreamSignals2 = [
-  "not your usage limit",
-  "not your account",
-  "not your account's",
-  "temporarily limiting requests",
-  "temporarily rate",
-  "server is temporarily",
-  "would exceed your account\u2019s rate limit",
-  "would exceed your account's rate limit"
-];
-function isTransientUpstreamSignal(text4) {
-  if (typeof text4 !== "string" || text4.length === 0)
-    return false;
-  const sample = text4.length > 16384 ? text4.slice(0, 16384) : text4;
-  const lower = sample.toLowerCase();
-  return transientUpstreamSignals2.some((s) => lower.includes(s));
-}
-
 // session-tail.ts
 function sanitizeCwdToProjectName(cwd) {
   return cwd.replace(/[^a-zA-Z0-9]/g, "-");
@@ -76610,6 +77220,7 @@ function detectErrorInTranscriptLine(line) {
     const errStr = typeof obj.error === "string" ? obj.error : "";
     const text4 = extractAssistantText(obj);
     const kind2 = status === 429 ? isTransientUpstreamSignal(`${text4}
+${errStr}`) || isLitellmProxyLocal4292(`${text4}
 ${errStr}`) ? "rate-limited" : "quota-exhausted" : classifyClaudeError({ type: errStr, status, message: text4 });
     return {
       kind: kind2,
@@ -80470,10 +81081,10 @@ function readTurnActiveMarkerAgeMs(stateDir, now) {
 }
 
 // ../src/build-info.ts
-var VERSION = "0.18.13";
-var COMMIT_SHA = "58d180bf";
-var COMMIT_DATE = "2026-07-12T00:08:55Z";
-var LATEST_PR = 3156;
+var VERSION = "0.18.15";
+var COMMIT_SHA = "2fa611a1";
+var COMMIT_DATE = "2026-07-12T04:44:56Z";
+var LATEST_PR = 3171;
 var COMMITS_AHEAD_OF_TAG = 0;
 
 // gateway/boot-version.ts
@@ -85140,22 +85751,71 @@ var inboundCoalescer = createInboundCoalescer({
   merge: (entries) => {
     const last = entries[entries.length - 1];
     const { primary, extras } = splitCoalescedAttachments(entries, (e) => e.downloadImage != null || e.attachment != null, coalesceMaxAttachments());
+    const forwardOrigins = dedupeForwardOrigins(entries.map((e) => e.forwardOrigin));
     return {
       text: entries.map((e) => e.text).filter((t) => t.length > 0).join(`
 `),
       ctx: last.ctx,
       downloadImage: primary?.downloadImage,
       attachment: primary?.attachment,
-      extraAttachments: extras.length > 0 ? extras.map((e) => ({ downloadImage: e.downloadImage, attachment: e.attachment })) : undefined
+      extraAttachments: extras.length > 0 ? extras.map((e) => ({ downloadImage: e.downloadImage, attachment: e.attachment })) : undefined,
+      forwardOrigins: forwardOrigins.length > 0 ? forwardOrigins : undefined
     };
   },
   onFlush: (key, merged) => {
     bufferedAttachmentKeys.delete(key);
-    handleInbound(merged.ctx, merged.text, merged.downloadImage, merged.attachment, merged.extraAttachments);
+    handleInbound(merged.ctx, merged.text, merged.downloadImage, merged.attachment, merged.extraAttachments, merged.forwardOrigins);
   }
 });
 function emitGatewayOperatorEvent(event) {
   const { agent, kind } = event;
+  let throttleEscalation = null;
+  let escalationFired = false;
+  const rateLimit429Classification = kind === "rate-limited" ? classify429Detail(event.detail) : null;
+  if (rateLimit429Classification != null && rateLimit429Classification !== "account-scoped") {
+    emitRuntimeMetric(build429ClassifiedMetric({
+      agent,
+      detail: event.detail,
+      classification: rateLimit429Classification,
+      action: "calm",
+      now: Date.now()
+    }));
+    if (rateLimit429Classification === "litellm-local") {
+      process.stderr.write(`telegram gateway: 429 classified litellm-proxy-local agent=${agent} \u2014 ` + `calm path, no account attribution, no failover
+`);
+    }
+  }
+  if (rateLimit429Classification === "account-scoped") {
+    const throttleDecision = decideThrottleTier({
+      detail: event.detail,
+      now: Date.now(),
+      thresholdMs: throttleRetryInPlaceMaxMs()
+    });
+    emitRuntimeMetric(build429ClassifiedMetric({
+      agent,
+      detail: event.detail,
+      classification: "account-scoped",
+      action: throttleDecision.action === "failover" ? "failover" : "throttle",
+      now: Date.now()
+    }));
+    if (throttleDecision.action === "throttle") {
+      process.stderr.write(`telegram gateway: throttle-tier staying-put agent=${agent} until=${new Date(throttleDecision.throttledUntilMs).toISOString()} parsedReset=${throttleDecision.resetParsed}
+`);
+      try {
+        recordOperatorEvent(event);
+      } catch {}
+      throttleTierRunner.fire(agent, throttleDecision.throttledUntilMs, throttleDecision.resetParsed);
+      return;
+    }
+    if (throttleDecision.action === "failover") {
+      const resetAt = new Date(throttleDecision.resetAtMs);
+      throttleEscalation = { kind: "rate_limited", resetAt, raw: event.detail };
+      if (wouldFireFleetAutoFallback()) {
+        escalationFired = true;
+        fireFleetAutoFallback(agent, resolveExhaustUntil(resetAt.getTime()), resetAt, "rate-limit");
+      }
+    }
+  }
   if (!shouldEmitOperatorEvent(agent, kind)) {
     process.stderr.write(`telegram gateway: operator-event suppressed (cooldown) agent=${agent} kind=${kind}
 `);
@@ -85167,13 +85827,13 @@ function emitGatewayOperatorEvent(event) {
     process.stderr.write(`telegram gateway: recordOperatorEvent failed agent=${agent} kind=${kind}: ${err.message}
 `);
   }
-  const modelUnavailable = resolveModelUnavailableFromOperatorEvent(event);
+  const modelUnavailable = throttleEscalation ?? resolveModelUnavailableFromOperatorEvent(event);
   let renderedText;
   let renderedKeyboard;
   let cardPromisedFallback = false;
   if (modelUnavailable) {
-    const isAutoKind = modelUnavailable.kind === "quota_exhausted";
-    const willActuallyFire = isAutoKind && wouldFireFleetAutoFallback();
+    const isAutoKind = modelUnavailable.kind === "quota_exhausted" || modelUnavailable.kind === "rate_limited";
+    const willActuallyFire = throttleEscalation != null ? escalationFired : isAutoKind && wouldFireFleetAutoFallback();
     process.stderr.write(`telegram gateway: operator-event suppressing-raw-stderr-for-model-unavailable agent=${agent} kind=${kind} detected=${modelUnavailable.kind} autoKind=${isAutoKind} willFire=${willActuallyFire}
 `);
     renderedText = formatModelUnavailableCard(modelUnavailable, agent, {
@@ -85181,9 +85841,9 @@ function emitGatewayOperatorEvent(event) {
     });
     renderedKeyboard = undefined;
     cardPromisedFallback = willActuallyFire;
-    if (willActuallyFire) {
+    if (willActuallyFire && throttleEscalation == null) {
       const untilMs = resolveExhaustUntil(modelUnavailable.resetAt?.getTime());
-      fireFleetAutoFallback(agent, untilMs);
+      fireFleetAutoFallback(agent, untilMs, modelUnavailable.resetAt);
     }
   } else {
     try {
@@ -86899,7 +87559,7 @@ var ipcServer = createIpcServer({
     const untilMs = resolveExhaustUntil(msg.resetAt);
     process.stderr.write(`telegram gateway: quota_wall_detected agent=${msg.agentName} until=${new Date(untilMs).toISOString()}` + (msg.resetAt == null ? " (reset unparsed \u2192 +7d default)" : "") + ` \u2014 triggering fleet auto-fallback
 `);
-    fireFleetAutoFallback(msg.agentName, untilMs);
+    fireFleetAutoFallback(msg.agentName, untilMs, msg.resetAt != null ? new Date(msg.resetAt) : undefined);
   },
   onQueryPendingPermission(client3, msg) {
     const self = process.env.SWITCHROOM_AGENT_NAME;
@@ -90392,7 +91052,13 @@ async function handleInboundCoalesced(ctx, text5, downloadImage, attachment) {
   maybeEarlyAckReaction(ctx, from);
   maybePokeFloorForMidTurnInbound(ctx, from);
   const key = inboundCoalesceKey(String(ctx.chat.id), ctx.message?.message_thread_id, String(from.id));
-  const result = inboundCoalescer.enqueue(key, { text: text5, ctx, downloadImage, attachment });
+  const result = inboundCoalescer.enqueue(key, {
+    text: text5,
+    ctx,
+    downloadImage,
+    attachment,
+    forwardOrigin: parseForwardOrigin(ctx.message?.forward_origin)
+  });
   if (result.bypass)
     return handleInbound(ctx, text5, downloadImage, attachment);
   if (hasAttachment)
@@ -90430,7 +91096,7 @@ function maybePokeFloorForMidTurnInbound(ctx, from) {
     return;
   pokeFloorNow(key, Date.now());
 }
-async function handleInbound(ctx, text5, downloadImage, attachment, extraAttachments) {
+async function handleInbound(ctx, text5, downloadImage, attachment, extraAttachments, coalescedForwardOrigins) {
   markIdleActivity();
   const isTopicMessage = ctx.message?.is_topic_message ?? false;
   const messageThreadId = ctx.message?.message_thread_id;
@@ -91056,6 +91722,9 @@ ${preBlock(write.output)}`;
   const replyToTextRaw = replyToMsg ? replyToMsg.text ?? replyToMsg.caption ?? undefined : undefined;
   const replyToText = replyToTextRaw != null ? replyToTextRaw.length > REPLY_TO_TEXT_MAX ? replyToTextRaw.slice(0, REPLY_TO_TEXT_MAX - 1) + "\u2026" : replyToTextRaw : undefined;
   const replyToTextEscaped = formatReplyToText(replyToTextRaw, REPLY_TO_TEXT_MAX);
+  const forwardOrigins = coalescedForwardOrigins ?? dedupeForwardOrigins([parseForwardOrigin(ctx.message?.forward_origin)]);
+  const forwardOriginMeta = buildForwardOriginMeta(forwardOrigins);
+  const primaryForwardOrigin = forwardOrigins[0];
   if (HISTORY_ENABLED) {
     try {
       recordInbound({
@@ -91068,7 +91737,12 @@ ${preBlock(write.output)}`;
         text: effectiveText,
         attachment_kind: attachment?.kind,
         reply_to_message_id: replyToMessageId ?? null,
-        reply_to_text: replyToText ?? null
+        reply_to_text: replyToText ?? null,
+        forwarded_from: primaryForwardOrigin?.name ?? null,
+        forwarded_from_type: primaryForwardOrigin?.type ?? null,
+        forwarded_from_id: primaryForwardOrigin?.id != null ? String(primaryForwardOrigin.id) : null,
+        forwarded_date: forwardOriginDateIso(primaryForwardOrigin),
+        forwarded_message_id: primaryForwardOrigin?.messageId ?? null
       });
     } catch (err) {
       process.stderr.write(`telegram gateway: history recordInbound failed: ${err}
@@ -91134,6 +91808,7 @@ ${preBlock(write.output)}`;
       ...imagePath ? { image_path: imagePath } : {},
       ...replyToMessageId != null ? { reply_to_message_id: String(replyToMessageId) } : {},
       ...replyToTextEscaped != null && replyToTextEscaped.length > 0 ? { reply_to_text: replyToTextEscaped } : {},
+      ...forwardOriginMeta,
       ...isQueuedMidTurn || isQueuedPrefix ? { queued: "true" } : {},
       ...isSteering ? { steering: "true" } : {},
       ...priorTurnInProgress ? { prior_turn_in_progress: "true" } : {},
@@ -93238,12 +93913,33 @@ function newestActiveTurnStartedAtMs() {
 function wouldFireFleetAutoFallback() {
   return fleetFallbackGate.wouldFire();
 }
-async function fireFleetAutoFallback(triggerAgent, untilMs) {
-  return fleetFallbackGate.fire(() => doFireFleetAutoFallback(triggerAgent, untilMs), (err) => {
+async function fireFleetAutoFallback(triggerAgent, untilMs, parsedResetAt, trigger) {
+  return fleetFallbackGate.fire(() => doFireFleetAutoFallback(triggerAgent, untilMs, parsedResetAt, trigger), (err) => {
     process.stderr.write(`telegram gateway: [fleet-fallback] error agent=${triggerAgent}: ${err?.message ?? err}
 `);
   });
 }
+var throttleTierRunner = createThrottleTierRunner({
+  agentName: process.env.SWITCHROOM_AGENT_NAME ?? "",
+  getBrokerClient: () => getAuthBrokerClient2(process.env.SWITCHROOM_AGENT_NAME ?? ""),
+  listNoticeChats: () => loadAccess().allowFrom,
+  sendNotice: (chat_id, markdown) => {
+    swallowingApiCall(() => bot.api.sendRichMessage(chat_id, richMessage2(markdown), { disable_notification: true }), { chat_id: String(chat_id), verb: "throttle-tier:notify" });
+  },
+  resumeDecide: (ts) => fleetFallbackResumeGate.decide(ts),
+  newestActiveTurnStartedAtMs,
+  turnInFlight: () => turnInFlightForGate(),
+  deferRestartToTurnComplete: (agentName3, reason) => {
+    process.stderr.write(`telegram gateway: [throttle-tier] restart deferred to turn-complete agent=${agentName3} reason=${reason}
+`);
+    pendingRestarts.set(agentName3, Date.now());
+  },
+  restartNow: (agentName3, reason) => {
+    triggerSelfRestart(agentName3, reason);
+  },
+  log: (m) => process.stderr.write(`telegram gateway: ${m}
+`)
+});
 var fallbackFailureNoticeState = { lastSentAtMs: 0 };
 var fallbackAllBlockedNoticeState = { lastSentAtMs: 0 };
 function broadcastFleetFallbackFailure(triggerAgent, reason) {
@@ -93264,7 +93960,7 @@ function broadcastFleetFallbackFailure(triggerAgent, reason) {
     swallowingApiCall(() => bot.api.sendRichMessage(chat_id, richMessage2(html), {}), { chat_id, verb: "fleet-fallback:failure-notify" });
   }
 }
-async function doFireFleetAutoFallback(triggerAgent, untilMs) {
+async function doFireFleetAutoFallback(triggerAgent, untilMs, parsedResetAt, trigger) {
   try {
     const client3 = await getAuthBrokerClient2(triggerAgent);
     if (!client3) {
@@ -93288,7 +93984,9 @@ async function doFireFleetAutoFallback(triggerAgent, untilMs) {
         return { rolledTo: r.rolledTo ?? null, rolled: r.rolled };
       },
       triggerAgent,
-      tz
+      tz,
+      parsedResetAt,
+      rateLimitTrigger: trigger === "rate-limit"
     });
     process.stderr.write(`telegram gateway: [fleet-fallback] outcome=${outcome.kind} agent=${triggerAgent}` + (outcome.kind === "switched" ? ` old=${outcome.oldLabel} new=${outcome.newLabel}` : "") + `
 `);