switchroom 0.16.12 → 0.16.14

package/dist/cli/switchroom.js

@@ -51674,8 +51674,8 @@ import { existsSync, readFileSync } from "node:fs";
 import { dirname, join } from "node:path";
 
 // src/build-info.ts
-var VERSION = "0.16.12";
-var COMMIT_SHA = "4cab85e1";
+var VERSION = "0.16.14";
+var COMMIT_SHA = "6daa5e37";
 
 // src/cli/resolve-version.ts
 function readPackageVersion() {

package/dist/host-control/main.js

@@ -22587,7 +22587,7 @@ import { existsSync as existsSync6, readFileSync as readFileSync4 } from "node:f
 import { dirname as dirname4, join as join2 } from "node:path";
 
 // src/build-info.ts
-var VERSION = "0.16.12";
+var VERSION = "0.16.14";
 
 // src/cli/resolve-version.ts
 function readPackageVersion() {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "switchroom",
-  "version": "0.16.12",
+  "version": "0.16.14",
   "description": "Run Claude Code 24/7 on your Claude Pro/Max subscription over Telegram. Open-source alternative to OpenClaw and NanoClaw — no API keys.",
   "type": "module",
   "bin": {

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

@@ -56064,10 +56064,10 @@ function readTurnActiveMarkerAgeMs(stateDir, now) {
 }
 
 // ../src/build-info.ts
-var VERSION = "0.16.12";
-var COMMIT_SHA = "4cab85e1";
-var COMMIT_DATE = "2026-06-28T11:11:09Z";
-var LATEST_PR = 2634;
+var VERSION = "0.16.14";
+var COMMIT_SHA = "6daa5e37";
+var COMMIT_DATE = "2026-06-28T19:14:51Z";
+var LATEST_PR = 2639;
 var COMMITS_AHEAD_OF_TAG = 0;
 
 // gateway/boot-version.ts
@@ -58112,9 +58112,11 @@ var progressUpdateLastSent = new Map;
 var progressUpdateTurnCount = new Map;
 var currentTurn = null;
 var currentTurnMap = new CurrentTurnMap;
+var lastActiveTurnChatId;
 function setCurrentTurn(turn, key) {
   currentTurnMap.set(turn, key);
   currentTurn = currentTurnMap.get();
+  lastActiveTurnChatId = turn.sessionChatId;
 }
 function endCurrentTurnForKey(turn, key) {
   const ended = currentTurnMap.endTurnForKey(turn, key);
@@ -60803,10 +60805,12 @@ async function executeReply(args) {
     const _a = loadAccess();
     if (_a.allowFrom.includes(_rawChatId) || _rawChatId in _a.groups)
       return _rawChatId;
-    if (turn?.sessionChatId && (_a.allowFrom.includes(turn.sessionChatId) || (turn.sessionChatId in _a.groups))) {
-      process.stderr.write(`telegram gateway: reply: model passed chat_id "${_rawChatId}" (not allowlisted) \u2014 ` + `routing to active turn chat "${turn.sessionChatId}"
+    const fallbackChatId = turn?.sessionChatId ?? lastActiveTurnChatId;
+    if (fallbackChatId && (_a.allowFrom.includes(fallbackChatId) || (fallbackChatId in _a.groups))) {
+      const tier = turn == null ? "last-known" : "active";
+      process.stderr.write(`telegram gateway: reply: model passed chat_id "${_rawChatId}" (not allowlisted) \u2014 ` + `routing to ${tier} turn chat "${fallbackChatId}"
 `);
-      return turn.sessionChatId;
+      return fallbackChatId;
     }
     return _rawChatId;
   })();
@@ -68281,13 +68285,19 @@ bot.on("callback_query:data", async (ctx) => {
       return;
     }
     await ctx.answerCallbackQuery({ text: "Switching\u2026" }).catch(() => {});
+    let didInterimSrEdit = false;
+    if (data.startsWith(MODEL_CALLBACK_SR)) {
+      const srLabel = escapeHtmlForTg(srFriendlyLabel(data.slice(MODEL_CALLBACK_SR.length)));
+      await ctx.editMessageText(`\u23F3 Switching session to <b>${srLabel}</b>\u2026`, { parse_mode: "HTML", reply_markup: { inline_keyboard: [] } }).catch(() => {});
+      didInterimSrEdit = true;
+    }
     try {
       const prevSessionModel = activeSessionModelOverride;
       const outcome = await handleModelMenuCallback(data, modelDeps);
       if (outcome.selectedModel) {
         activeSessionModelOverride = outcome.selectedModel;
       }
-      if (outcome.toastOnly)
+      if (outcome.toastOnly && !didInterimSrEdit)
         return;
       if (outcome.selectedModel && isSrToClaudeTransition(prevSessionModel, outcome.selectedModel)) {
         const agentName3 = getMyAgentName();

package/telegram-plugin/gateway/gateway.ts

@@ -278,6 +278,8 @@ import {
   isSrToClaudeTransition,
   MODEL_CALLBACK_PREFIX,
   MODEL_CALLBACK_HEADER,
+  MODEL_CALLBACK_SR,
+  srFriendlyLabel,
   type ModelMenuDeps,
   type ModelCommandDeps,
   type ModelMenuReply,
@@ -2216,6 +2218,13 @@ type CurrentTurn = {
 // is never written and this is exactly the old singleton.
 let currentTurn: CurrentTurn | null = null
 const currentTurnMap = new CurrentTurnMap<CurrentTurn>()
+// Captures the most-recently-started turn's sessionChatId. Unlike currentTurn,
+// this is NOT cleared by the silence poke (firePoke/clearTurnStarted). It lets
+// the Bug B fallback in executeReply route to the correct chat even when the
+// model calls reply AFTER the silence poke has nulled currentTurn (Bug D).
+// Safe in single-tenant gateways: the fallback only fires when the raw chat_id
+// fails the allowlist, and this value was itself validated at inbound receipt.
+let lastActiveTurnChatId: string | undefined
 
 /**
  * Set the live turn for `key`. Flag-branches in ONE place (inside the map):
@@ -2227,6 +2236,7 @@ const currentTurnMap = new CurrentTurnMap<CurrentTurn>()
 function setCurrentTurn(turn: CurrentTurn, key: string): void {
   currentTurnMap.set(turn, key)
   currentTurn = currentTurnMap.get() // mirror most-recent-set (== `turn`)
+  lastActiveTurnChatId = turn.sessionChatId
 }
 
 /**
@@ -7977,17 +7987,23 @@ async function executeReply(args: Record<string, unknown>): Promise<{ content: A
   // Non-Claude models (e.g. Gemini via LiteLLM sr-* routing) sometimes pass a
   // chat_id that is not in the allowlist — either an int/string mismatch, or
   // the model echoing the wrong identifier from context. When the raw value
-  // fails the allowlist check and the active turn has a validated sessionChatId,
-  // fall back to the turn's origin chat so the reply still lands correctly.
+  // fails the allowlist check, fall back to the active (or last-known) turn's
+  // validated sessionChatId so the reply still lands correctly.
+  // Tier 1: live turn (currentTurn was non-null at reply entry).
+  // Tier 2: last-known turn (survives silence poke — Bug D fix; currentTurn is
+  //   null because clearTurnStarted fired ≥5 min of model silence, but the model
+  //   finally called reply after the poke).
   const chat_id = (() => {
     const _a = loadAccess()
     if (_a.allowFrom.includes(_rawChatId) || _rawChatId in _a.groups) return _rawChatId
-    if (turn?.sessionChatId && (_a.allowFrom.includes(turn.sessionChatId) || turn.sessionChatId in _a.groups)) {
+    const fallbackChatId = turn?.sessionChatId ?? lastActiveTurnChatId
+    if (fallbackChatId && (_a.allowFrom.includes(fallbackChatId) || fallbackChatId in _a.groups)) {
+      const tier = turn == null ? 'last-known' : 'active'
       process.stderr.write(
         `telegram gateway: reply: model passed chat_id "${_rawChatId}" (not allowlisted) — ` +
-          `routing to active turn chat "${turn.sessionChatId}"\n`,
+          `routing to ${tier} turn chat "${fallbackChatId}"\n`,
       )
-      return turn.sessionChatId
+      return fallbackChatId
     }
     return _rawChatId // let assertAllowedChat below throw the human-readable error
   })()
@@ -20835,6 +20851,23 @@ bot.on('callback_query:data', async ctx => {
       return
     }
     await ctx.answerCallbackQuery({ text: 'Switching…' }).catch(() => {})
+    // sr-* inject waits for claude to respond (can take 10-30s). Edit the
+    // menu immediately to show a "working on it" state so the operator isn't
+    // left looking at a stale menu with no feedback. The final edit (✅/❌)
+    // replaces this once the inject returns.
+    // NOTE: this await yields the event loop, so a new inbound turn could
+    // start between here and handleModelMenuCallback's inner isBusy() check.
+    // We track whether we applied the interim edit so we can skip the
+    // toastOnly short-circuit if we did — a toastOnly return after the interim
+    // edit would leave the menu stuck button-less.
+    let didInterimSrEdit = false
+    if (data.startsWith(MODEL_CALLBACK_SR)) {
+      const srLabel = escapeHtmlForTg(srFriendlyLabel(data.slice(MODEL_CALLBACK_SR.length)))
+      await ctx
+        .editMessageText(`⏳ Switching session to <b>${srLabel}</b>…`, { parse_mode: 'HTML', reply_markup: { inline_keyboard: [] } })
+        .catch(() => {})
+      didInterimSrEdit = true
+    }
     try {
       const prevSessionModel = activeSessionModelOverride
       const outcome = await handleModelMenuCallback(data, modelDeps)
@@ -20844,9 +20877,11 @@ bot.on('callback_query:data', async ctx => {
       if (outcome.selectedModel) {
         activeSessionModelOverride = outcome.selectedModel
       }
-      // toastOnly: a no-op outcome that should not disturb the menu (defence
-      // in depth — the isBusy() short-circuit above is the live path).
-      if (outcome.toastOnly) return
+      // toastOnly: leave the menu untouched — but only if we haven't already
+      // cleared its buttons with the interim sr-* edit. If we have, fall
+      // through to the final edit so the message is recovered (busyReply or
+      // the full menu) rather than left permanently button-less.
+      if (outcome.toastOnly && !didInterimSrEdit) return
 
       // sr-* → Claude transition via the model menu: trigger a graceful restart.
       // Switching FROM an sr-* (LiteLLM/OpenRouter) model BACK to a Claude model

package/telegram-plugin/gateway/model-command.ts

@@ -345,7 +345,7 @@ export function expandSrAlias(arg: string): string {
   return SR_MODEL_ALIASES[arg.toLowerCase()] ?? arg
 }
 
-function srFriendlyLabel(srName: string): string {
+export function srFriendlyLabel(srName: string): string {
   return SR_MODEL_LABELS[srName] ?? srName.replace(/^sr-/, '').replace(/-/g, ' ')
 }
 

package/telegram-plugin/uat/scenarios/jtbd-model-litellm-sr-dm.test.ts

@@ -59,10 +59,13 @@ describe("uat: /model sr-* LiteLLM routing — section headers + session switch
         const openrouterHeader = flat.find(
           (b) => b.text.includes("OpenRouter") && b.callbackData === "mdl:h",
         );
-        // Prefer a fast non-reasoning model for the E2E test; reasoning models
-        // (deepseek-r1, o1, o3) take 2-5 min per response and hit the silence poke.
+        // Prefer deepseek-v3 (non-thinking, consistently fast) for the E2E test.
+        // gemini-2.5-flash may run in thinking mode via OpenRouter (5+ min latency),
+        // reasoning models (deepseek-r1, o1, o3) also take 2-5 min and hit the
+        // silence poke — both are OK now (Bug D fallback), but slow tests are painful.
         const srButton =
-          flat.find((b) => b.callbackData?.startsWith("mdl:sr:") && /flash/.test(b.callbackData)) ??
+          flat.find((b) => b.callbackData?.startsWith("mdl:sr:") && /deepseek-v3/.test(b.callbackData)) ??
+          flat.find((b) => b.callbackData?.startsWith("mdl:sr:") && /flash/.test(b.callbackData) && !/thinking/.test(b.callbackData)) ??
           flat.find((b) => b.callbackData?.startsWith("mdl:sr:") && !/r1|o1|o3|thinking/.test(b.callbackData)) ??
           flat.find((b) => b.callbackData?.startsWith("mdl:sr:"));
 
@@ -93,10 +96,12 @@ describe("uat: /model sr-* LiteLLM routing — section headers + session switch
         expect((kbAfter ?? []).flat().length, "menu keeps buttons after sr-* tap").toBeGreaterThan(0);
 
         // ── 3. Send a quick message to generate a LiteLLM-routed turn ──
-        // Use 120s — fast models (gemini-flash, deepseek-v3) respond in <10s,
-        // but the request still has to go through the model switch inject + proxy.
+        // 480s budget: fast models (deepseek-v3) reply in <10s; Gemini 2.5 Flash
+        // may run thinking mode via OpenRouter and take 5+ min. The silence poke
+        // fires at 300s, which is OK — the Bug D fallback (lastActiveTurnChatId)
+        // routes the reply even when currentTurn was nulled by the poke.
         await sc.sendDM("Just reply with the word OK.");
-        await sc.expectMessage(/ok/i, { from: "bot", timeout: 120_000 });
+        await sc.expectMessage(/ok/i, { from: "bot", timeout: 480_000 });
 
         // ── 4. LiteLLM spend attribution ────────────────────────────────
         if (spendBefore >= 0) {
@@ -126,7 +131,7 @@ describe("uat: /model sr-* LiteLLM routing — section headers + session switch
         await sc.tearDown();
       }
     },
-    210_000,
+    660_000,
   );
 
   it(
@@ -135,11 +140,11 @@ describe("uat: /model sr-* LiteLLM routing — section headers + session switch
       const sc = await spinUp({ agent: AGENT });
       try {
         await sc.sendDM("/model");
-        // 60s — test 2 runs after test 1's restore restart, which takes ~15s.
-        // If the restart is still finishing when /model lands, it may be queued.
+        // 90s — test 2 runs after test 1's restore restart. The model-switch
+        // restart can take 30–60s to fully boot; 90s gives comfortable margin.
         const menu = await sc.expectMessage(/Default \(new sessions\):/i, {
           from: "bot",
-          timeout: 60_000,
+          timeout: 90_000,
         });
         const kb = await sc.driver.getKeyboard(sc.botUserId, menu.messageId);
         const flat = (kb ?? []).flat();