@dotobokuri/fleet-console 1.62.0 → 1.63.0

package/dist/fleet-plugins/terminal/routes.mjs

@@ -20834,12 +20834,12 @@ var benchmarks_default = {
 };
 var models_default = {
   version: 1,
-  updatedAt: "2026-08-15T00:00:00Z",
+  updatedAt: "2026-08-17T00:00:00Z",
   providers: {
     codex: {
       name: "Codex",
       defaultModel: "gpt-5.6-sol",
-      source: "codex app-server model/list (codex-cli 0.147.0, 2026-08-11 \u2014 ultra advertised for sol/sol-wm/terra, luna stops at max) for model/effort/service-tier; contextWindow=272000 is codex debug models context_window/max_context_window (codex-cli 0.146.0, 2026-08-02)",
+      source: "codex app-server model/list (codex-cli 0.147.0, observed 2026-08-11) for model/effort/service-tier; Fleet catalog effort ladders are capped at max because harness launch capabilities are not model rungs; contextWindow=272000 is codex debug models context_window/max_context_window (codex-cli 0.146.0, observed 2026-08-02)",
       models: [
         {
           modelId: "gpt-5.6-sol",
@@ -20853,8 +20853,7 @@ var models_default = {
               "medium",
               "high",
               "xhigh",
-              "max",
-              "ultra"
+              "max"
             ]
           },
           aliases: [
@@ -20876,8 +20875,7 @@ var models_default = {
               "medium",
               "high",
               "xhigh",
-              "max",
-              "ultra"
+              "max"
             ]
           },
           aliases: [
@@ -20897,8 +20895,7 @@ var models_default = {
               "medium",
               "high",
               "xhigh",
-              "max",
-              "ultra"
+              "max"
             ]
           },
           aliases: [
@@ -20920,8 +20917,7 @@ var models_default = {
               "medium",
               "high",
               "xhigh",
-              "max",
-              "ultra"
+              "max"
             ]
           },
           aliases: [
@@ -21453,8 +21449,7 @@ var GATEWAY_REASONING_EFFORTS = [
   "medium",
   "high",
   "xhigh",
-  "max",
-  "ultra"
+  "max"
 ];
 var GatewayEffortUpstreamModelIdsSchema = external_exports.partialRecord(
   external_exports.enum(GATEWAY_REASONING_EFFORTS),
@@ -21881,8 +21876,7 @@ var ANTHROPIC_EFFORT_RUNGS = /* @__PURE__ */ new Set([
   "medium",
   "high",
   "xhigh",
-  "max",
-  "ultra"
+  "max"
 ]);
 function freezeGatewayModelEffort(effort) {
   if (!effort?.supported) return UNSUPPORTED_GATEWAY_MODEL_EFFORT;
@@ -31491,11 +31485,10 @@ var GENERAL_PURPOSE_AGENT_PROMPT = [
 ].join("\n");
 var FLEET_PLUGIN_NAME = "fleet";
 function exposedEffortLadder(modelId, ladder, exposure) {
-  const deliverable = ladder.filter((rung) => rung !== "ultra");
   const chosen = exposure?.[modelId];
-  if (chosen === void 0 || chosen.length === 0) return deliverable;
-  const narrowed = deliverable.filter((rung) => chosen.includes(rung));
-  return narrowed.length > 0 ? narrowed : deliverable;
+  if (chosen === void 0 || chosen.length === 0) return ladder;
+  const narrowed = ladder.filter((rung) => chosen.includes(rung));
+  return narrowed.length > 0 ? narrowed : ladder;
 }
 function buildGatewayCustomAgents(exposed, exposure) {
   const agents = {};
@@ -44313,6 +44306,9 @@ function shouldSyncDirectoryForTest(platform) {
 function readSocketRole(value) {
   return value === "viewer" ? "viewer" : void 0;
 }
+function readTicketChannel(value) {
+  return value === "chat" ? "chat" : void 0;
+}
 var DEFAULT_TICKET_TTL_MS = 1e4;
 function createPluginTerminalTicketRegistry(deps = {}) {
   const ttlMs = deps.ttlMs ?? DEFAULT_TICKET_TTL_MS;
@@ -45000,7 +44996,23 @@ function createPluginTerminalUpgradeHandler(deps) {
       return true;
     }
     const context = deps.tickets.consume(url2.searchParams.get("ticket"));
-    if (!context || !deps.sessions.canAttach(context.sessionId)) {
+    if (!context) {
+      socket.destroy();
+      return true;
+    }
+    if (context.channel === "chat") {
+      const attachChat = deps.attachChat;
+      if (!attachChat) {
+        socket.destroy();
+        return true;
+      }
+      const server2 = getWebSocketServer();
+      server2.handleUpgrade(req, socket, head, (ws) => {
+        attachChat(ws, context);
+      });
+      return true;
+    }
+    if (!deps.sessions.canAttach(context.sessionId)) {
       socket.destroy();
       return true;
     }
@@ -45039,6 +45051,7 @@ function readUpgradeUrl(req) {
 var SHELL_OPERATION_TYPE = "shell";
 function createTerminalRuntime(ctx) {
   const tickets = createPluginTerminalTicketRegistry();
+  let chatAttach = null;
   const terminalExitListeners = /* @__PURE__ */ new Set();
   const terminalTitleListeners = /* @__PURE__ */ new Map();
   const terminalLaunchResolvers = /* @__PURE__ */ new Map();
@@ -45061,7 +45074,14 @@ function createTerminalRuntime(ctx) {
   const upgrade = createPluginTerminalUpgradeHandler({
     tickets,
     sessions,
-    isAuthorized: ctx.host.security.isTerminalAuthorized
+    isAuthorized: ctx.host.security.isTerminalAuthorized,
+    attachChat: (socket, context) => {
+      if (!chatAttach) {
+        socket.close(1013, "chat_unavailable");
+        return;
+      }
+      chatAttach(socket, context);
+    }
   });
   return {
     handleUpgrade: upgrade.handleUpgrade,
@@ -45097,7 +45117,14 @@ function createTerminalRuntime(ctx) {
         if (terminalLaunchResolvers.get(operationType) === resolver) terminalLaunchResolvers.delete(operationType);
       };
     },
+    bindChatAttach: (attach) => {
+      chatAttach = attach;
+      return () => {
+        if (chatAttach === attach) chatAttach = null;
+      };
+    },
     stop: async () => {
+      chatAttach = null;
       upgrade.close();
       await sessions.stop();
       terminalExitListeners.clear();
@@ -45345,6 +45372,7 @@ var EFFORT_LABELS = {
   ultra: "ULTRACODE"
 };
 var APEX_EFFORTS = ["max", "ultra"];
+var ULTRACODE_LAUNCH_EFFORT = "ultra";
 var NATIVE_MODEL_LABELS = {
   // Claude Code's 1M coordinates stay under their plain menu labels.
   "fable[1m]": "Fable",
@@ -45401,27 +45429,27 @@ function buildClaudeGatewayLaunchVariants(selection) {
   }
   return groups;
 }
-var EFFORT_AXIS = [...NATIVE_CLAUDE_EFFORTS, "ultra"];
+var EFFORT_AXIS = [...NATIVE_CLAUDE_EFFORTS, ULTRACODE_LAUNCH_EFFORT];
 function providerGroupId(provider) {
   return `gateway:${provider}`;
 }
 function toGatewayRow(model, selection) {
-  const efforts = (selection.effortExposure[model.id] ?? exposableEffortLadder(model)).filter((effort) => EFFORT_AXIS.includes(effort));
-  const gatedEfforts = APEX_EFFORTS.filter((effort) => efforts.includes(effort));
-  const effortAxis = EFFORT_AXIS.filter((effort) => !APEX_EFFORTS.includes(effort) || efforts.includes(effort));
+  const ordinary = selection.effortExposure[model.id] ?? exposableEffortLadder(model);
+  const hasMax = ordinary.includes("max");
+  const effortAxis = ordinary.length === 0 ? [ULTRACODE_LAUNCH_EFFORT] : [...NATIVE_CLAUDE_EFFORTS.filter((effort) => effort !== "max" || hasMax), ULTRACODE_LAUNCH_EFFORT];
+  const gatedEfforts = hasMax ? ["max", ULTRACODE_LAUNCH_EFFORT] : [ULTRACODE_LAUNCH_EFFORT];
+  const chips = [...ordinary, ULTRACODE_LAUNCH_EFFORT];
   return {
     id: model.id,
     label: bareModelName(model),
     launch: { model: model.id },
-    ...efforts.length > 0 ? {
-      effortAxis,
-      ...gatedEfforts.length > 0 ? { gatedEfforts } : {},
-      chips: efforts.map((effort) => ({
-        id: effort,
-        label: EFFORT_LABELS[effort] ?? effort.toUpperCase(),
-        launch: { model: model.id, effort }
-      }))
-    } : {}
+    effortAxis,
+    gatedEfforts,
+    chips: chips.map((effort) => ({
+      id: effort,
+      label: EFFORT_LABELS[effort] ?? effort.toUpperCase(),
+      launch: { model: model.id, effort }
+    }))
   };
 }
 function resolveDisabledReason(cli) {
@@ -45684,6 +45712,32 @@ async function createIsolatedClaudeConfigDir(tempRoot = tmpdir()) {
     }
   };
 }
+function readVendorContextUsage(response) {
+  if (typeof response !== "object" || response === null) return null;
+  const usage4 = response;
+  if (typeof usage4.totalTokens !== "number" || typeof usage4.maxTokens !== "number") return null;
+  const rows = Array.isArray(usage4.categories) ? usage4.categories : [];
+  const memory = Array.isArray(usage4.memoryFiles) ? usage4.memoryFiles : [];
+  const mcp = Array.isArray(usage4.mcpTools) ? usage4.mcpTools : [];
+  return {
+    total: usage4.totalTokens,
+    max: usage4.maxTokens,
+    model: typeof usage4.model === "string" ? usage4.model : "",
+    compactAt: usage4.isAutoCompactEnabled === true && typeof usage4.autoCompactThreshold === "number" ? usage4.autoCompactThreshold : null,
+    categories: rows.flatMap((row) => {
+      const entry = row;
+      return typeof entry?.name === "string" && typeof entry.tokens === "number" ? [{ name: entry.name, tokens: entry.tokens, deferred: entry.isDeferred === true }] : [];
+    }),
+    memoryFiles: memory.flatMap((row) => {
+      const entry = row;
+      return typeof entry?.path === "string" && typeof entry.tokens === "number" ? [{ path: entry.path, tokens: entry.tokens }] : [];
+    }),
+    mcpTools: mcp.flatMap((row) => {
+      const entry = row;
+      return typeof entry?.name === "string" && typeof entry.tokens === "number" ? [{ name: entry.name, server: typeof entry.serverName === "string" ? entry.serverName : "", tokens: entry.tokens }] : [];
+    })
+  };
+}
 function runVendorQuery(input) {
   const run = query({
     prompt: input.prompt,
@@ -45691,6 +45745,14 @@ function runVendorQuery(input) {
   });
   let closed = false;
   return {
+    async getContextUsage() {
+      if (closed || typeof run.getContextUsage !== "function") return null;
+      try {
+        return readVendorContextUsage(await run.getContextUsage());
+      } catch {
+        return null;
+      }
+    },
     [Symbol.asyncIterator]() {
       const iterator = run[Symbol.asyncIterator]();
       return {
@@ -45814,7 +45876,8 @@ async function createClaudeGatewaySdk(options) {
           } finally {
             release();
           }
-        }
+        },
+        getContextUsage: () => run.getContextUsage()
       };
       active = tracked;
       return tracked;
@@ -46401,7 +46464,8 @@ var GatewayLaunchOptionError = class extends Error {
   }
 };
 function isGatewayLaunchEffortAllowed(selection, model, effort) {
-  if (![...NATIVE_CLAUDE_EFFORTS, "ultra"].includes(effort)) return false;
+  if (effort === "ultra") return true;
+  if (!GATEWAY_REASONING_EFFORTS.includes(effort)) return false;
   const efforts = selection.effortExposure[model.id] ?? exposableEffortLadder(model);
   return efforts.includes(effort);
 }
@@ -47951,6 +48015,8 @@ function capText(text2) {
 // ../fleet-plugins/terminal/server/agent-api/chat-session.ts
 var JOURNAL_CAP = 2e3;
 var TOOL_NAME_CAP = 500;
+var CONTEXT_RESERVED_NAME = "Autocompact buffer";
+var CONTEXT_UNSPENT = /* @__PURE__ */ new Set([CONTEXT_RESERVED_NAME, "Free space"]);
 var JOB_KIND_CAP = 200;
 var JOB_TAIL_READ_BYTES = 256 * 1024;
 var JOB_TRANSCRIPT_READ_BYTES = 4 * 1024 * 1024;
@@ -48041,6 +48107,11 @@ var AgentChatSession = class {
    * dispose(). 어느 경로로도 비워지지 않으면 SDK는 무기한 멈춘다(권한 요청에 park deadline이 없다).
    */
   pendingAsks = /* @__PURE__ */ new Map();
+  /**
+   * 문맥 스냅숏의 세대. 턴마다 하나씩 오르며, 늦게 도착한 응답이 자기 턴의 것인지 가른다 —
+   * 답은 턴이 끝난 뒤에 오므로(실측), 그 사이 새 턴이 시작했다면 그 값은 남의 턴 자리에 앉는다.
+   */
+  contextGeneration = 0;
   constructor(operationId, seed, createSdk) {
     this.operationId = operationId;
     this.seed = seed;
@@ -48352,6 +48423,58 @@ var AgentChatSession = class {
       this.seed.reportAwaiting(true);
     });
   }
+  /**
+   * 턴이 시작하는 순간 자식에게 문맥 내역을 묻고, 답이 오면 그대로 저널에 싣는다.
+   *
+   * **이 한 번이 유일한 기회다**(실측). 자식은 턴이 시작되면 control 채널을 닫아, +0ms의 요청만
+   * 답을 받고 그 뒤의 요청은 턴이 10초를 돌든 전부 "Query closed before response received"로
+   * 끝난다. 주기적으로 물어 최신값을 좇는 설계는 이 자식에서 성립하지 않는다.
+   *
+   * 그래서 값은 **이 턴이 시작될 때까지의** 문맥이다 — 방금 끝난 턴이 더한 몫은 다음 턴이
+   * 시작될 때 비로소 드러난다. 화면은 그 지연을 감추지 않는다.
+   *
+   * 응답을 기다리지 않는 이유는 스트림 때문이다: 여기서 await하면 첫 메시지 소비가 그만큼 늦다.
+   * 늦게 도착한 답도 같은 턴의 시작 시점 값이므로, 그때 실어도 뜻이 달라지지 않는다.
+   */
+  snapshotContextAtTurnStart(run) {
+    const generation = ++this.contextGeneration;
+    try {
+      void run.getContextUsage().then((usage4) => {
+        if (usage4 && generation === this.contextGeneration) this.pushContext(usage4);
+      }, () => void 0);
+    } catch {
+    }
+  }
+  /**
+   * 턴 시작 시점의 문맥을 저널에 심는다.
+   *
+   * 총량을 vendor의 `total`이 아니라 **카테고리 합으로 다시 세는** 이유는 실측이다: 게이트웨이
+   * 세션에서 그 값(24,948)은 카테고리 합(129,670)과 크게 어긋났다 — 전자는 게이트웨이가 실제로
+   * 읽은 토큰이고 후자는 CLI가 센 로컬 배분이다. 화면은 내역과 총량을 나란히 세우므로 둘이
+   * 다른 셈을 쓰면 사용자가 보는 자리에서 모순이 드러난다. 내역이 곧 총량이어야 한다.
+   *
+   * 예약분(자동 압축 여유)과 남은 자리는 **쓴 것이 아니다**. 같은 목록에 두면 합이 언제나 창
+   * 전체가 되어 미터가 항상 100%를 가리킨다.
+   */
+  pushContext(usage4) {
+    const spent = usage4.categories.filter((category) => !category.deferred && category.tokens > 0 && !CONTEXT_UNSPENT.has(category.name));
+    const reserved = usage4.categories.find((category) => category.name === CONTEXT_RESERVED_NAME);
+    this.push({
+      kind: "context",
+      total: spent.reduce((sum, category) => sum + category.tokens, 0),
+      max: usage4.max,
+      ...reserved && reserved.tokens > 0 ? { reserved: reserved.tokens } : {},
+      ...usage4.compactAt === null ? {} : { compactAt: usage4.compactAt },
+      slices: spent.map((category) => ({ name: category.name, tokens: category.tokens })),
+      ...usage4.memoryFiles.length > 0 ? { memoryFiles: usage4.memoryFiles.map((file2) => ({ name: file2.path, tokens: file2.tokens })) } : {},
+      ...usage4.mcpTools.length > 0 ? {
+        mcpTools: usage4.mcpTools.map((tool) => ({
+          name: tool.server ? `${tool.server} \xB7 ${tool.name}` : tool.name,
+          tokens: tool.tokens
+        }))
+      } : {}
+    });
+  }
   push(event) {
     const entry = { seq: ++this.seq, event };
     if (event.kind === "job-progress") {
@@ -48466,6 +48589,7 @@ var AgentChatSession = class {
         return;
       }
       this.activeRun = run;
+      this.snapshotContextAtTurnStart(run);
       try {
         for await (const message of run) {
           if (typeof message.session_id === "string" && message.session_id.length > 0) {
@@ -48617,6 +48741,105 @@ var AgentChatRegistry = class {
   }
 };
 
+// ../fleet-plugins/terminal/server/agent-api/chat-ws.ts
+var MAX_CHAT_ANSWER_MESSAGE_CHARS = 2e3;
+var CHAT_UNAVAILABLE_CLOSE_CODE = 1013;
+function attachAgentChatSocket(socket, start) {
+  let closed = false;
+  let unsubscribe = null;
+  let session = null;
+  socket.once("close", () => {
+    closed = true;
+    unsubscribe?.();
+    unsubscribe = null;
+    session = null;
+  });
+  socket.on("message", (data, isBinary) => {
+    if (closed || isBinary || !session) return;
+    const command = readChatSocketCommand(decodeSocketText(data));
+    if (!command) {
+      sendJson(socket, { type: "nack", error: "invalid_command", ...commandId(data) ? { id: commandId(data) } : {} });
+      return;
+    }
+    if (command.type === "stop") {
+      if (!session.stopTurn()) {
+        sendJson(socket, { type: "nack", id: command.id, error: "chat_idle" });
+        return;
+      }
+      sendJson(socket, { type: "ok", id: command.id });
+      return;
+    }
+    const result = session.answer(command.askId, {
+      ...command.answers ? { answers: command.answers } : {},
+      ...command.approve === true ? { approve: true } : {},
+      ...command.message !== void 0 ? { message: command.message.slice(0, MAX_CHAT_ANSWER_MESSAGE_CHARS) } : {}
+    });
+    if (!result.ok) {
+      sendJson(socket, { type: "nack", id: command.id, error: result.error });
+      return;
+    }
+    sendJson(socket, { type: "ok", id: command.id });
+  });
+  void start().then((ready) => {
+    if (closed) return;
+    if ("error" in ready) {
+      sendJson(socket, { seq: 0, event: { kind: "error", code: ready.error } });
+      socket.close(CHAT_UNAVAILABLE_CLOSE_CODE, ready.error);
+      return;
+    }
+    session = ready;
+    unsubscribe = ready.subscribe((entry) => {
+      if (!closed) sendJson(socket, entry);
+    });
+  }).catch(() => {
+    if (closed) return;
+    sendJson(socket, { seq: 0, event: { kind: "error", code: "chat_unavailable" } });
+    socket.close(CHAT_UNAVAILABLE_CLOSE_CODE, "chat_unavailable");
+  });
+}
+function sendJson(socket, value) {
+  if (socket.readyState !== 1) return;
+  socket.send(Buffer.from(JSON.stringify(value), "utf8"), { binary: false });
+}
+function decodeSocketText(data) {
+  if (Buffer.isBuffer(data)) return data.toString("utf8");
+  if (Array.isArray(data)) return Buffer.concat(data).toString("utf8");
+  return Buffer.from(data).toString("utf8");
+}
+function commandId(data) {
+  try {
+    const parsed = JSON.parse(decodeSocketText(data));
+    if (parsed && typeof parsed === "object" && typeof parsed.id === "string") {
+      return parsed.id;
+    }
+  } catch {
+  }
+  return void 0;
+}
+function readChatSocketCommand(raw) {
+  let parsed;
+  try {
+    parsed = JSON.parse(raw);
+  } catch {
+    return null;
+  }
+  if (!parsed || typeof parsed !== "object") return null;
+  const value = parsed;
+  if (typeof value.id !== "string" || value.id.length === 0) return null;
+  if (value.type === "stop") return { type: "stop", id: value.id };
+  if (value.type !== "answer" || typeof value.askId !== "string" || value.askId.length === 0) return null;
+  const answers = Array.isArray(value.answers) ? value.answers.filter((entry) => typeof entry === "string") : void 0;
+  if (Array.isArray(value.answers) && answers?.length !== value.answers.length) return null;
+  return {
+    type: "answer",
+    id: value.id,
+    askId: value.askId,
+    ...answers ? { answers } : {},
+    ...value.approve === true ? { approve: true } : {},
+    ...typeof value.message === "string" ? { message: value.message } : {}
+  };
+}
+
 // ../fleet-plugins/terminal/server/agent-api/analysis-types.ts
 var ANALYSIS_ERROR_CODES = {
   captureMissing: "analysis_capture_missing",
@@ -48909,7 +49132,7 @@ var TERMINAL_PLUGIN_ID = "terminal";
 var CHAT_MODE_PAYLOAD_KEY = "chatMode";
 var CHAT_BORN_PAYLOAD_KEY = "chatBorn";
 var CHAT_SUPPORTED_CLI_ID = "claude-gateway";
-var MAX_CHAT_ANSWER_MESSAGE_CHARS = 2e3;
+var MAX_CHAT_ANSWER_MESSAGE_CHARS2 = 2e3;
 function resolveClaudeConfigDir() {
   const configured = process4.env.CLAUDE_CONFIG_DIR;
   return configured && configured.length > 0 ? configured : path19__default.join(os3__default.homedir(), ".claude");
@@ -48934,7 +49157,7 @@ async function registerAgentRoutes(ctx, terminalRuntime, deps) {
     { method: "POST", path: "/sessions/:sessionId/message", summary: "Deliver a prompt to an Agent session.", category: "Terminal Plugin", gate: "origin-write", transport: "http" },
     { method: "POST", path: "/sessions/:sessionId/chat", summary: "Switch an Agent session to chat mode.", category: "Terminal Plugin", gate: "origin-write", transport: "http" },
     { method: "DELETE", path: "/sessions/:sessionId/chat", summary: "Leave chat mode for an Agent session.", category: "Terminal Plugin", gate: "origin-write", transport: "http" },
-    { method: "GET", path: "/sessions/:sessionId/chat-stream", summary: "Stream Agent chat events.", category: "Terminal Plugin", gate: "origin-write", transport: "sse" },
+    { method: "GET", path: "/sessions/:sessionId/chat-stream", summary: "Removed \u2014 Chat Mode observation uses the ticketed WebSocket.", category: "Terminal Plugin", gate: "origin-write", transport: "http" },
     { method: "POST", path: "/sessions/:sessionId/chat-answer", summary: "Answer a pending Agent chat question.", category: "Terminal Plugin", gate: "origin-write", transport: "http" },
     { method: "POST", path: "/sessions/:sessionId/chat-stop", summary: "Stop the in-flight Agent chat turn.", category: "Terminal Plugin", gate: "origin-write", transport: "http" },
     { method: "GET", path: "/sessions/:sessionId/chat-job", summary: "Read one Agent chat background job's detail.", category: "Terminal Plugin", gate: "origin-write", transport: "http" },
@@ -49001,6 +49224,27 @@ async function createAgentApi(ctx, terminalRuntime, deps) {
   const oscActivityTrackers = /* @__PURE__ */ new Map();
   const testChatSdkFactory = globalThis.__fleetAgentChatSdkFactory;
   const chatRegistry = testChatSdkFactory ? new AgentChatRegistry(testChatSdkFactory) : new AgentChatRegistry();
+  const unbindChatAttach = terminalRuntime.bindChatAttach((socket, context) => {
+    const sessionId = context.sessionId;
+    attachAgentChatSocket(socket, async () => {
+      const node = ctx.host.operations.get(sessionId);
+      if (!node || node.pluginId !== ctx.pluginId || node.type !== AGENT_OPERATION_TYPE) {
+        return { error: "session_not_found" };
+      }
+      if (node.payload[CHAT_MODE_PAYLOAD_KEY] !== true) return { error: "chat_not_active" };
+      const seed = await resolveChatSeed(node);
+      if (!seed.ok) return { error: seed.error };
+      if (ctx.host.operations.get(sessionId)?.payload[CHAT_MODE_PAYLOAD_KEY] !== true) {
+        return { error: "chat_not_active" };
+      }
+      try {
+        return await chatRegistry.ensure(sessionId, () => seed.seed);
+      } catch {
+        return { error: "chat_unavailable" };
+      }
+    });
+  });
+  ctx.host.lifecycle.registerCleanup(unbindChatAttach);
   const launchResolver = createAgentTerminalLaunchResolver({
     agentRuntime: runtime,
     ...deps.aiGateway ? { aiGateway: deps.aiGateway } : {},
@@ -49136,6 +49380,23 @@ async function createAgentApi(ctx, terminalRuntime, deps) {
         ctx.host.http.writeJson(res, 404, { error: "terminal_session_not_found" });
         return true;
       }
+      const channel = readTicketChannel(body?.channel);
+      if (channel === "chat") {
+        if (operation.payload[CHAT_MODE_PAYLOAD_KEY] !== true) {
+          ctx.host.http.writeJson(res, 409, { error: "chat_not_active" });
+          return true;
+        }
+        ctx.host.http.writeJson(res, 200, terminalRuntime.issueTicket({
+          cwd: readPayloadString(operation.payload, "cwd") ?? ctx.host.paths.resolveTheaterPath(operation.theaterId) ?? "",
+          sessionId: operation.id,
+          operationId: operation.id,
+          operationType: operation.type,
+          pluginId: operation.pluginId,
+          theaterId: operation.theaterId,
+          channel: "chat"
+        }));
+        return true;
+      }
       if (observability.getTerminalSessionInfo(operation.id)?.status === "dormant") {
         ctx.host.http.writeJson(res, 409, { error: "operation_dormant" });
         return true;
@@ -49750,7 +50011,7 @@ async function createAgentApi(ctx, terminalRuntime, deps) {
     const result = chat.answer(askId, {
       ...answers ? { answers } : {},
       ...body?.approve === true ? { approve: true } : {},
-      ...typeof body?.message === "string" ? { message: body.message.slice(0, MAX_CHAT_ANSWER_MESSAGE_CHARS) } : {}
+      ...typeof body?.message === "string" ? { message: body.message.slice(0, MAX_CHAT_ANSWER_MESSAGE_CHARS2) } : {}
     });
     if (!result.ok) {
       const status = result.error === "ask_not_found" ? 404 : result.error === "plan_truncated" ? 409 : 400;
@@ -49818,56 +50079,7 @@ async function createAgentApi(ctx, terminalRuntime, deps) {
       ctx.host.http.writeJson(res, 409, { error: "chat_not_active" });
       return true;
     }
-    let closed = false;
-    let unsubscribe = null;
-    const write = (data) => {
-      if (!closed && !res.writableEnded && !res.destroyed) res.write(data);
-    };
-    res.writeHead(200, {
-      "Content-Type": "text/event-stream",
-      "Cache-Control": "no-store",
-      Connection: "keep-alive",
-      "Cross-Origin-Opener-Policy": "same-origin",
-      "Cross-Origin-Resource-Policy": "same-origin",
-      "Referrer-Policy": "no-referrer",
-      "X-Content-Type-Options": "nosniff"
-    });
-    write(": open\n\n");
-    const keepalive = setInterval(() => write(": keepalive\n\n"), 3e4);
-    req.on("close", () => {
-      closed = true;
-      clearInterval(keepalive);
-      unsubscribe?.();
-    });
-    void (async () => {
-      const seed = await resolveChatSeed(node);
-      if (!seed.ok) {
-        write(`data: ${JSON.stringify({ seq: 0, event: { kind: "error", code: seed.error } })}
-
-`);
-        res.end();
-        return;
-      }
-      if (ctx.host.operations.get(sessionId)?.payload[CHAT_MODE_PAYLOAD_KEY] !== true) {
-        write(`data: ${JSON.stringify({ seq: 0, event: { kind: "error", code: "chat_not_active" } })}
-
-`);
-        res.end();
-        return;
-      }
-      try {
-        const chat = await chatRegistry.ensure(sessionId, () => seed.seed);
-        if (closed) return;
-        unsubscribe = chat.subscribe((entry) => write(`data: ${JSON.stringify(entry)}
-
-`));
-      } catch {
-        write(`data: ${JSON.stringify({ seq: 0, event: { kind: "error", code: "chat_unavailable" } })}
-
-`);
-        res.end();
-      }
-    })();
+    ctx.host.http.writeJson(res, 410, { error: "chat_stream_moved" });
     return true;
   }
   async function resolveChatSeed(node) {