fix(chat): make automatic naming actually name things

apps/ade-cli/README.md

@@ -480,7 +480,7 @@ ade chat read session-id --limit 20 --max-chars 8000 --text
 ade chat read session-id --page --cursor 4096 --limit 20 --max-chars 8000 --text
 ade chat message session-id --kind auto --text "status/context"
 ade chat steer session-id --text "active-turn context"
-ade chat note "testing desktop auth fallback"               # update Work status (3–6 words, max 72 characters); add --session <id> to target explicitly
+ade chat note "testing desktop auth fallback"               # update Work status (aim for 6 words or fewer; truncated past 72 characters); add --session <id> to target explicitly
 ade chat ask "Which account should I use?"                 # escalate a blocking question; add --session <id> to target explicitly
 ade session show session-id --text                          # settle/snooze state, and why a snoozed row came back
 ade session snooze session-id --for 1h                      # 30m|1h|4h|1d|1.5h; a bare number means minutes; relative durations cap at 30d

apps/ade-cli/src/cli.ts

@@ -29,6 +29,10 @@ import {
   type DoctorRow,
 } from "./commands/doctor";
 export { readInstalledDesktopVersion } from "./commands/doctor";
+import {
+  MAX_STATUS_NOTE_CHARACTERS,
+  STATUS_NOTE_GUIDELINE_WORDS,
+} from "../../desktop/src/shared/sessionStatusNote";
 import { buildDeeplink, type DeeplinkEnvelope } from "../../desktop/src/shared/deeplinks";
 import { buildPairingQrPayload } from "../../desktop/src/shared/pairingQr";
 import { buildWebClientPairUrl } from "../../desktop/src/shared/webClientUrl";
@@ -1743,7 +1747,7 @@ const HELP_BY_COMMAND: Record<string, string> = {
     $ ade chat create --from-linear-issue ENG-431 --parent <session> --type subagent
                                                     Start a child chat with an attached issue + kickoff (alias: --linear-issue-json)
     $ ade chat send <session> --text "next step"    Send a message; steers automatically if the turn is active
-    $ ade chat note "testing desktop auth fallback" # Update the Work status line (3–6 words, max 72 characters)
+    $ ade chat note "testing desktop auth fallback" # Update the Work status line (aim for ${STATUS_NOTE_GUIDELINE_WORDS} words or fewer; truncated past ${MAX_STATUS_NOTE_CHARACTERS} characters)
     $ ade chat ask "Which account should I use?"    Escalate a blocking question to the user
                                                     'note' and 'ask' default to the caller and accept --session <id>.
                                                     'chat settle' / 'chat unsettle' were removed: only the user (or a

apps/ade-cli/src/services/sync/rosterBuilder.test.ts

@@ -263,7 +263,8 @@ describe("buildRosterSnapshot", () => {
 
     expect(byId.get("chat-run")).toMatchObject({
       settledAt: "2026-01-02T00:01:00Z",
-      statusNote: "Indexing complete and waiting for final…",
+      // Eight words survive: the note only truncates past 72 characters.
+      statusNote: "Indexing complete and waiting for final review now",
       exitCode: null,
     });
     expect(byId.get("chat-await")).toMatchObject({

apps/ade-cli/src/tuiClient/__tests__/RightPane.test.tsx

@@ -1379,3 +1379,24 @@ describe("RightPane feedback form", () => {
     expect(draft.summary).toBe("add dark mode");
   });
 });
+
+describe("RightPane list rows", () => {
+  it("clips a long chat row to the pane instead of wrapping it onto extra lines", () => {
+    // A status note is bounded at 72 characters, not at six words, so a chat
+    // row can easily exceed the 38-column pane. It must clip, not wrap.
+    const longRow = "● claude-auth-login · claude · done: reworked automatic naming so the note keeps its meaning";
+    const content = {
+      kind: "list" as const,
+      title: "Chats",
+      rows: [longRow, "● short row"],
+      action: { kind: "chat-list" as const, ids: ["a", "b"] },
+    };
+    const result = render(<RightPane content={content} focused width={38} />);
+    const lines = stripAnsi(result.lastFrame() ?? "").split("\n");
+    const rowLines = lines.filter((line) => line.includes("claude-auth-login"));
+    expect(rowLines).toHaveLength(1);
+    expect(rowLines[0]?.length).toBeLessThanOrEqual(38);
+    expect(rowLines[0]).toContain("…");
+    expect(lines.some((line) => line.includes("short row"))).toBe(true);
+  });
+});

apps/ade-cli/src/tuiClient/components/RightPane.tsx

@@ -2385,14 +2385,21 @@ function RightPaneComponent({
             // list) can't scroll past the content into a blank pane.
             const listStart = Math.max(0, Math.min(scrollOffsetRows, Math.max(0, content.rows.length - DETAILS_BODY_MAX_LINES)));
             const visibleRows = content.rows.slice(listStart, listStart + DETAILS_BODY_MAX_LINES);
+            // A row is free text (a chat row carries an agent-authored status
+            // note up to its full 72-character budget), so it has to be clipped
+            // to the pane: an unclipped row wraps onto extra lines and pushes
+            // the "N more" footer and every row below it out of the window.
+            const rowWidth = Math.max(8, paneWidth - 4);
             return content.rows.length ? visibleRows.map((row, visibleIndex) => {
               const index = listStart + visibleIndex;
+              const prefix = content.action ? `${index === selectedIndex ? theme.rail : " "} ` : "";
               return (
                 <Text
                   key={`${content.action?.ids[index] ?? row}:${index}`}
+                  wrap="truncate-end"
                   color={content.action && (index === selectedIndex || hoveredId === `right:list:${index}`) ? theme.color.violet : undefined}
                 >
-                  {content.action ? `${index === selectedIndex ? theme.rail : " "} ${row}` : row}
+                  {`${prefix}${endTruncate(row, Math.max(4, rowWidth - prefix.length))}`}
                 </Text>
               );
             }) : <Text color={theme.color.t4} dimColor>{content.emptyText ?? "No data."}</Text>;

apps/desktop/resources/agent-skills/ade-cli-control-plane/SKILL.md

@@ -229,9 +229,9 @@ row. If you are blocked, `ade chat ask "<question>"` raises the row's hand.
 Treat the status line and hand-raise as separate signals:
 
 - **`ade chat note` explains the current state.** Write one concrete,
-  present-tense summary of **3–6 words and at most 72 characters**. ADE
-  truncates longer notes, so put the decisive state first and never write a
-  full sentence.
+  present-tense summary aiming for **6 words or fewer** — a guideline, not a
+  hard limit. ADE truncates past **72 characters**, so put the decisive state
+  first and never write a full sentence, but a long note still beats no note.
   Good: `CI green; awaiting Codex review`
   Bad: `Working`, `Still looking`, `Blocked`, or `Done`.
 - **`ade chat ask` means work cannot continue without a user answer.** Ask the

apps/desktop/src/main/services/chat/agentChatService.test.ts

@@ -1597,6 +1597,45 @@ function createService(overrides: Record<string, unknown> = {}) {
   return { service, logger, laneService, sessionService, projectConfigService, aiIntegrationService };
 }
 
+async function waitFor(condition: () => boolean, timeoutMs = 5_000): Promise<void> {
+  const deadline = Date.now() + timeoutMs;
+  while (!condition()) {
+    if (Date.now() > deadline) return;
+    await new Promise((resolve) => setTimeout(resolve, 10));
+  }
+}
+
+function installAutoTitleAuth(): void {
+  // Auto-titling is skipped outright when no model is reachable.
+  vi.mocked(detectAllAuth).mockResolvedValue([
+    { type: "cli-subscription" as any, cli: "codex", authenticated: true, path: "/usr/bin/codex", verified: true },
+    { type: "cli-subscription" as any, cli: "claude", authenticated: true, path: "/usr/bin/claude", verified: true },
+  ] as never);
+}
+
+function installAutoTitleClaudeStream(): void {
+  let streamCall = 0;
+  vi.mocked(claudeSdkCreateSessionCompat).mockReturnValue({
+    send: vi.fn().mockResolvedValue(undefined),
+    stream: vi.fn(() => (async function* () {
+      streamCall += 1;
+      if (streamCall === 1) {
+        yield { type: "system", subtype: "init", session_id: "sdk-session-1", slash_commands: [] };
+        yield { type: "result", usage: { input_tokens: 1, output_tokens: 1 } };
+        return;
+      }
+      yield {
+        type: "assistant",
+        message: { content: [{ type: "text", text: "Done" }], usage: { input_tokens: 1, output_tokens: 1 } },
+      };
+      yield { type: "result", usage: { input_tokens: 1, output_tokens: 1 } };
+    })()),
+    close: vi.fn(),
+    sessionId: "sdk-session-1",
+    setPermissionMode: vi.fn().mockResolvedValue(undefined),
+  } as any);
+}
+
 const HANDOFF_TEST_SHA = "1234567890abcdef1234567890abcdef12345678";
 const HANDOFF_BEHIND_SHA = "0123456789abcdef0123456789abcdef01234567";
 const HANDOFF_DIVERGED_SHA = "fedcba9876543210fedcba9876543210fedcba98";
@@ -13426,6 +13465,76 @@ describe("createAgentChatService", () => {
 
       expect(aiIntegrationService.summarizeTerminal).not.toHaveBeenCalled();
     });
+
+    it("does not clobber a manual rename that lands while auto-titling is in flight", async () => {
+      const events: AgentChatEventEnvelope[] = [];
+      installAutoTitleClaudeStream();
+      installAutoTitleAuth();
+
+      let renameDuringNaming: Promise<unknown> | null = null;
+      const { service, sessionService, aiIntegrationService } = createService({
+        onEvent: (event: AgentChatEventEnvelope) => events.push(event),
+      });
+      const session = await service.createSession({ laneId: "lane-1", provider: "claude", model: "sonnet" });
+
+      // The naming request only resolves after the user has renamed the chat,
+      // which is exactly the race that used to overwrite their title and clear
+      // the manuallyNamed flag.
+      aiIntegrationService.summarizeTerminal.mockImplementation(async () => {
+        if (renameDuringNaming) return { text: "Model Picked That" } as never;
+        renameDuringNaming = service.updateSession({
+          sessionId: session.id,
+          title: "User Picked This",
+          manuallyNamed: true,
+        });
+        await renameDuringNaming;
+        return { text: "Model Picked That" } as never;
+      });
+
+      await service.sendMessage({ sessionId: session.id, text: "Build me a new feature" });
+      await waitForEvent(events, (event): event is AgentChatEventEnvelope => event.event.type === "done");
+      await waitFor(() => Boolean(renameDuringNaming));
+      await renameDuringNaming;
+      // Wait for the clobber itself rather than a fixed delay: pre-fix code
+      // writes the model title as soon as the naming call resolves, so this
+      // returns immediately when the regression is present and costs a bounded
+      // wait when it is not.
+      await waitFor(() => sessionService.get(session.id)?.title === "Model Picked That", 1_000);
+
+      expect(renameDuringNaming, "auto-title never ran, so the race was not exercised").not.toBeNull();
+      expect(aiIntegrationService.summarizeTerminal).toHaveBeenCalled();
+      expect(sessionService.get(session.id)?.title).toBe("User Picked This");
+      expect(sessionService.get(session.id)?.manuallyNamed).toBe(true);
+      expect(sessionService.updateMeta).not.toHaveBeenCalledWith(
+        expect.objectContaining({ title: "Model Picked That" }),
+      );
+    });
+
+    it("falls back to a deterministic title when every naming model fails", async () => {
+      const events: AgentChatEventEnvelope[] = [];
+      installAutoTitleClaudeStream();
+      installAutoTitleAuth();
+
+      const { service, sessionService, aiIntegrationService } = createService({
+        onEvent: (event: AgentChatEventEnvelope) => events.push(event),
+      });
+      // A provider-level failure condemns each provider in turn, so the chain
+      // runs out of models — the chat must still never sit on "Claude Chat".
+      aiIntegrationService.summarizeTerminal.mockRejectedValue(
+        new Error("The model is not supported when using Codex with a ChatGPT account."),
+      );
+
+      const session = await service.createSession({ laneId: "lane-1", provider: "claude", model: "sonnet" });
+      await service.sendMessage({ sessionId: session.id, text: "Rewrite the lane naming fallback chain" });
+      await waitForEvent(events, (event): event is AgentChatEventEnvelope => event.event.type === "done");
+      await waitFor(() => (sessionService.get(session.id)?.title ?? "") !== "Claude Chat");
+
+      const title = sessionService.get(session.id)?.title ?? "";
+      expect(aiIntegrationService.summarizeTerminal).toHaveBeenCalled();
+      expect(title).not.toBe("Claude Chat");
+      expect(title.split(/\s+/).filter(Boolean).length).toBeGreaterThanOrEqual(2);
+      expect(title.toLowerCase()).toContain("lane");
+    });
   });
 
   // --------------------------------------------------------------------------
@@ -37729,10 +37838,36 @@ describe("suggestLaneNameFromPrompt", () => {
     });
 
     expect(result.laneTitle).toBe("Claude OAuth Login");
-    expect(result.branchFragment).toBe("claude-auth-login-button");
+    expect(result.branchFragment).toBe("claude-auth-login-button-hangs");
+  });
+
+  it("clamps an over-long AI identity instead of discarding it for a slug", async () => {
+    vi.mocked(detectAllAuth).mockResolvedValue([
+      { type: "cli-subscription" as any, cli: "codex", authenticated: true, path: "/usr/bin/codex", verified: true },
+    ]);
+    const { service, aiIntegrationService } = createSuggestService();
+    // Six words is guidance for the model, not a gate: a seven-word answer is
+    // trimmed, never thrown away in favour of the deterministic slug.
+    vi.mocked(aiIntegrationService.summarizeTerminal).mockResolvedValueOnce({
+      text: JSON.stringify({
+        laneTitle: "Rework Session Naming Fallback Chain For Chats",
+        branchFragment: "rework-session-naming-fallback-chain-for-chats",
+      }),
+    } as any);
+
+    const result = await service.generateAutoLaneIdentity({
+      prompt: "Rework the session naming fallback chain",
+      modelId: "openai/gpt-5.4",
+      laneId: "lane-1",
+      temporaryBranch: "ade/1a2b3c4d",
+    });
+
+    expect(result.source).toBe("ai");
+    expect(result.laneTitle).toBe("Rework Session Naming Fallback Chain For");
+    expect(result.branchFragment).toBe("rework-session-naming-fallback-chain-for");
   });
 
-  it("treats fully invalid structured fields as deterministic fallback", async () => {
+  it("retries the next model when structured fields are unusable, then falls back deterministically", async () => {
     vi.mocked(detectAllAuth).mockResolvedValue([
       { type: "cli-subscription" as any, cli: "codex", authenticated: true, path: "/usr/bin/codex", verified: true },
     ]);
@@ -37749,11 +37884,13 @@ describe("suggestLaneNameFromPrompt", () => {
     });
 
     expect(result).toMatchObject({
-      laneTitle: "Claude Auth Login Button",
-      branchFragment: "claude-auth-login-button",
+      laneTitle: "Claude Auth Login Button Hangs",
+      branchFragment: "claude-auth-login-button-hangs",
       source: "deterministic",
     });
-    expect(aiIntegrationService.summarizeTerminal).toHaveBeenCalledTimes(1);
+    // An unusable answer no longer ends the chain: the remaining candidates
+    // still get a turn before naming settles for the deterministic slug.
+    expect(aiIntegrationService.summarizeTerminal).toHaveBeenCalledTimes(3);
   });
 
   it("uses the configured naming model before the launched model", async () => {