merge: upstream e798b46 (#471 OSC 7501 program status)

docs/tui-capabilities.md

@@ -18,16 +18,31 @@ observed generation window, not server hardware throughput.
 
 ## Terminal titles and notifications
 
-Terminal titles show the current state, session name and MCode, for example
-`Needs approval | Fix login | MCode`. Renaming or switching a session updates the
-title. Unnamed sessions use the project name and a short session ID. Titles are
-cleared when MCode exits or suspends and reapplied when it resumes.
-
-Configure these presentation settings in the MCode data directory's `config.yaml`:
+The interactive TUI also reports [OSC 7501 program status](https://www.superlogical.com/rex/docs/build/program-status)
+with `app=kcode`: idle, working, done, error, or blocked on permission, a question,
+or login. It reuses the current session's runtime status, including queue handoffs,
+compaction and retries, and does not depend on status line items. Cancellation and
+suspension report idle; resume reapplies the current status. Completed and failed
+results remain reported when the TUI exits. Reports contain only `state`, `kind` and
+`app`, and are sent on changes, without a heartbeat or capability query, as
+permitted by protocol revision 0.4. Supporting terminals decide how to display
+them; other terminals should ignore the sequence. The protocol treats each
+multiplexer pane as its own terminal, so Kinetick Code does not wrap reports for tmux
+passthrough; the multiplexer must implement OSC 7501 for its panes. Redirected
+output, headless exec and ACP do not emit these reports. Set
+`tui.programStatus: false` to disable them. Terminal title and notification
+settings remain independent of program status.
+
+Terminal titles show the session title, or `Kinetick Code` when the session is
+untitled. Renaming or switching a session updates the title. Agent status does
+not rewrite the title.
+
+Configure these presentation settings in the Kinetick Code data directory's `config.yaml`:
 
 ```yaml
 tui:
   terminalTitle: [status, session-name, app-name]
+  programStatus: true
   notifications:
     when: unfocused
     method: auto
@@ -38,7 +53,7 @@ Title items can be ordered or omitted; `project-name` is also available. Set
 `terminalTitle` to `null` or `[]` to disable title updates. Unknown items are ignored.
 Notification `when` accepts `unfocused`, `always` or `never`; `method` accepts
 `auto`, `osc9`, `osc777` or `bel`. Omitting `events` enables all four events; `[]`
-disables them. Apply configuration changes by restarting MCode.
+disables them. Apply configuration changes by restarting Kinetick Code.
 
 Notifications identify the session and suppress duplicates. Completion waits for
 the session's queue to finish; failed turns and requests for input can notify

packages/agent-core/test/unit/pi-turn-runner/llm-retry.test.ts

@@ -605,6 +605,36 @@ describe("withLLMRetry", () => {
     expect(error?.type === "error" ? error.error.errorMessage : undefined).toBe(message);
   });
 
+  // A request that is too large (HTTP 413, request_too_large, the local
+  // pre-send body check, or a context-window overflow) is rejected again on
+  // every resend; the BYOK retry-all budget used to send it six times.
+  it.each([
+    ["minimax_api", "413 Request Entity Too Large"],
+    ["minimax_api", '413 {"type":"error","error":{"type":"request_too_large","message":"Request exceeds the maximum allowed number of bytes."}}'],
+    ["custom_provider:work", "413 Payload Too Large"],
+    ["custom_provider:work", "413 status code (no body)"],
+    ["minimax_api", "Request body too large: 70000000 bytes exceeds client limit 67108864 bytes."],
+    ["minimax_api", "400 invalid params, context window exceeds limit (2013)"],
+    ["custom_provider:work", '400 {"error":{"message":"prompt is too long: 600000 tokens > 512000 maximum"}}'],
+    ["minimax", "413 Request Entity Too Large"],
+    ["minimax", "Request body too large: 70000000 bytes exceeds client limit 67108864 bytes."],
+  ])("sends an oversized request from %s only once: %s", async (provider, message) => {
+    let attempts = 0;
+    const sleep = vi.fn(async () => {});
+    const inner = (async () => {
+      attempts += 1;
+      return errorStream(message);
+    }) as StreamFn;
+    const wrapped = withLLMRetry(inner, retryOptions({ sleep }));
+
+    const events = await collectEvents(await wrapped(fakeModel(provider), CONTEXT, {}));
+
+    expect(attempts).toBe(1);
+    expect(sleep).not.toHaveBeenCalled();
+    const error = events.find((event) => event.type === "error");
+    expect(error?.type === "error" ? error.error.errorMessage : undefined).toBe(message);
+  });
+
   it.each([
     "[Error] 400 Bad Request",
     "503 Service Unavailable: too many images queued, try again",
@@ -1181,4 +1211,40 @@ describe("deterministic request rejection classification", () => {
     expect(rejection("401 invalid api key")).toBe(false);
     expect(classifyLLMRequestRejectionMessage("invalid_request_error without a status")).toBeUndefined();
   });
+
+  it("treats HTTP 413, byte-limit and context-overflow rejections as deterministic", () => {
+    const oversized = rejection;
+    expect(oversized("413 Request Entity Too Large")).toBe(true);
+    expect(oversized("anything", 413)).toBe(true);
+    expect(oversized('{"type":"error","error":{"type":"request_too_large","message":"too big"}}')).toBe(true);
+    expect(oversized("Request body too large: 9 bytes exceeds client limit 8 bytes.")).toBe(true);
+    expect(oversized("400 invalid params, context window exceeds limit (2013)")).toBe(true);
+    expect(oversized("prompt is too long: 213462 tokens > 200000 maximum")).toBe(true);
+    expect(rejection("413 Payload Too Large")).toBe(true);
+  });
+
+  it("does not treat transient or unrelated size-like errors as deterministic", () => {
+    const oversized = rejection;
+    // The bare MiniMax 2013 code is a generic invalid-params code.
+    expect(oversized("400 invalid params (2013)")).toBe(false);
+    expect(oversized("400 Bad Request")).toBe(false);
+    expect(oversized("prompt is too long", 503)).toBe(false);
+    expect(oversized("429 rate limited: request too large for TPM")).toBe(false);
+    expect(oversized("Request body too large: timeout while uploading")).toBe(false);
+    expect(oversized("401 invalid api key")).toBe(false);
+  });
+
+  it("classifies the BYOK-attributed form of an oversized request", () => {
+    const byok = (detail: string) =>
+      `BYOK upstream error: ${JSON.stringify({
+        errorCode: LLM_ERROR_CODES.LLM_UPSTREAM_ERROR,
+        message: `BYOK provider custom_provider:fixture upstream error: ${detail}`,
+        errorSource: "byok_upstream",
+        errorDetail: detail,
+        errorProviderId: "custom_provider:fixture",
+      })}`;
+    expect(rejection(byok("413 Payload Too Large"))).toBe(true);
+    expect(rejection(byok("400 invalid params, context window exceeds limit (2013)"))).toBe(true);
+    expect(rejection(byok("503 Service Unavailable"))).toBe(false);
+  });
 });

packages/agent-modules/system-reminder/src/blocks.ts

@@ -217,7 +217,9 @@ export function buildSlimAgentContextBlock(env: AgentEnv): string {
     lines.push(`  agentName: ${env.agentName}`);
   }
   lines.push(`  SESSION ROLE: ${sessionRole}`);
-  lines.push(`  YOUR SESSION ID: ${env.sessionId}`);
+  if (!env.sessionIdInContext) {
+    lines.push(`  YOUR SESSION ID: ${env.sessionId}`);
+  }
   if (env.parentSessionId) {
     lines.push(`  PARENT SESSION: ${env.parentSessionId}`);
     lines.push(buildParentResultDeliveryLine(env, isCloud));

packages/agent-modules/system-reminder/src/types.ts

@@ -134,6 +134,13 @@ export interface AgentEnv {
   userConfiguredName?: string | undefined;
   agentRole: string;
   sessionId: string;
+  /**
+   * Host-verified fact that the model-visible context already carries
+   * `YOUR SESSION ID` (e.g. an earlier user message since the latest
+   * compaction). When true, the slim per-turn block omits the line to save
+   * tokens. Absent/false keeps the line (cloud and legacy callers).
+   */
+  sessionIdInContext?: boolean | undefined;
   /** Session type — 0=Branch, 1=Root (formerly Main). */
   sessionType?: number | undefined;
   /** Parent session ID — set when this is a child session. */

packages/config/src/tui-config.ts

@@ -46,6 +46,8 @@ export interface TuiConfig {
   };
   /** Show contextual Tips in the idle composer header. Defaults to true. */
   showTips?: boolean;
+  /** Report OSC 7501 program status to the terminal. Defaults to true. */
+  programStatus?: boolean;
   /**
    * Status line items, in display order.
    *
@@ -115,6 +117,7 @@ export function parseTuiConfig(raw: Record<string, unknown>): TuiConfig {
           }
         : {}),
     ...(typeof tui.showTips === 'boolean' ? { showTips: tui.showTips } : {}),
+    ...(typeof tui.programStatus === 'boolean' ? { programStatus: tui.programStatus } : {}),
     ...(notifications ? { notifications } : {}),
     ...(statusLine ? { statusLine } : {}),
     ...(customStatusLine ? { customStatusLine } : {}),

packages/local-runtime-v2/src/compat/v1/agent-host.test.ts

@@ -132,6 +132,22 @@ describe("createV1AgentHostProductCapabilities", () => {
       turnId: "turn-1",
       deferTelemetry: true,
     });
+    await product.inputPreparation.reminders.buildSystem({
+      session: { sessionId: "session-1" },
+      agent: { agentName: "mavis", resourceAgentName: "main" },
+      agentConfig: {},
+      promptText: "again",
+      turnId: "turn-2",
+      sessionIdInContext: true,
+    } as never);
+    expect(hosted.reminders.buildSystem).toHaveBeenLastCalledWith({
+      session: { sessionId: "session-1" },
+      resourceAgentName: "main",
+      promptText: "again",
+      turnId: "turn-2",
+      deferTelemetry: true,
+      sessionIdInContext: true,
+    });
     expect(product.executor.fileChanges.begin).toBe(hosted.fileChanges.begin);
     expect(product.runner.contextUsage).toBe(hosted.contextUsage);
     expect(product.runner.logger).toBe(hosted.runnerLogger);

packages/local-runtime-v2/src/compat/v1/agent-host.ts

@@ -195,6 +195,7 @@ export function createV1AgentHostProductCapabilities(
             promptText: input.promptText,
             turnId: input.turnId,
             deferTelemetry: true,
+            ...(input.sessionIdInContext ? { sessionIdInContext: true } : {}),
             ...(model ? { model } : {}),
           });
         },

packages/local-runtime-v2/src/service/model-system/README.md

@@ -63,7 +63,7 @@ Official OpenCode Go HTTPS endpoints under `https://opencode.ai/zen/go` receive
 
 登录完成后，`CodexOAuthManager` 从当前 profile 的 `codex-auth.json` 获取凭据，复用
 `AuthStorage` 的刷新机制和宿主 `fetch`，请求
-`https://chatgpt.com/backend-api/codex/models?client_version=0.153.0`。
+`https://chatgpt.com/backend-api/codex/models?client_version=0.162.0`。
 `connectivity/codex-model-discovery.ts` 负责这个独立的 Codex 协议。兼容版本来自已验证的
 Codex 客户端协议，与 MiniMax Code 产品版本独立；模型 ID 和思考等级由响应提供。OAuth 解析复用可配置思考协议，将远端等级原样映射到请求；
 `ultra` 等新增等级不依赖 Pi 的静态等级枚举。

packages/local-runtime-v2/src/service/model-system/codex-oauth.test.ts

@@ -795,7 +795,7 @@ describe("Codex OAuth profile credentials", () => {
       fetch: fetchImpl,
     });
     expect(fetchImpl).toHaveBeenCalledWith(
-      expect.stringContaining("/codex/models?"),
+      "https://chatgpt.com/backend-api/codex/models?client_version=0.162.0",
       expect.objectContaining({
         headers: expect.objectContaining({
           Authorization: "Bearer fresh",

packages/local-runtime-v2/src/service/model-system/connectivity/codex-model-discovery.ts

@@ -2,7 +2,7 @@ import type { LocalCustomProviderConfig, LocalModelConfig } from '../contracts.j
 
 const CODEX_BASE_URL = 'https://chatgpt.com/backend-api';
 // Codex model discovery negotiates against the Codex client version, independent of our app version.
-const CODEX_CATALOG_CLIENT_VERSION = '0.153.0';
+const CODEX_CATALOG_CLIENT_VERSION = '0.162.0';
 
 export interface CodexModelCredentials {
   access: string;

packages/local-runtime-v2/src/service/turn-system/agent-host/assembly/local-turn-execution-preparation.ts

@@ -45,6 +45,7 @@ export class NativeLocalTurnExecutionPreparationSource<
       ...(input.execution.desktopCapabilities
         ? { desktopCapabilities: input.execution.desktopCapabilities }
         : {}),
+      history: (input.execution.runnerHistory ?? input.execution.history)?.messages ?? [],
     });
     const browserAssets = [...(prepared.browserAssets ?? [])];
     const context = createToolContext(

packages/local-runtime-v2/src/service/turn-system/agent-host/assembly/local-turn-input-preparation.ts

@@ -21,6 +21,7 @@ import {
 } from '../canonical-user-input.js';
 import { createBackgroundTaskHostMetadata } from '../history/background/host-metadata.js';
 import { assertAgentHostCapabilityAvailable } from '../empty-dependencies.js';
+import { messagesCarrySessionId } from '../compaction/session-identity.js';
 import {
   prepareLocalInlineMedia,
   type LocalInlineMediaCandidate,
@@ -86,6 +87,12 @@ export interface LocalTurnReminderFacts {
     readonly promptText: string;
     readonly turnId: string;
     readonly desktopCapabilities?: AgentHostTurnCapabilityView;
+    /**
+     * True when a user message in the model-visible history since the latest
+     * compaction already contains this session's ID, so per-turn reminders may
+     * omit it. False on the first turn and after a compaction dropped it.
+     */
+    readonly sessionIdInContext?: boolean;
   }) => Promise<{
     readonly content: string;
     readonly diagnostic?: unknown;
@@ -135,6 +142,8 @@ export interface LocalTurnInputPreparationRequest {
   readonly immediateSendBatch?: AgentHostExecutionRequest['immediateSendBatch'];
   readonly provenance: AgentHostTurnProvenance;
   readonly desktopCapabilities?: AgentHostTurnCapabilityView;
+  /** Provider-facing history before this turn's user message. */
+  readonly history?: readonly unknown[];
 }
 
 export interface PreparedLocalTurnInput {
@@ -203,6 +212,9 @@ export class LocalTurnInputPreparer {
       promptText,
       turnId: input.lease.turnId,
       ...(input.desktopCapabilities ? { desktopCapabilities: input.desktopCapabilities } : {}),
+      ...(input.history && messagesCarrySessionId(input.history, input.session.sessionId)
+        ? { sessionIdInContext: true }
+        : {}),
     });
     const system = validateSystemReminder(systemValue);
     const systemBlock = optionalReminderBlock('system', system.content);