feat(tui): pause the queue on ctrl+c instead of resuming the run

docs/PRODUCT.md

@@ -41,7 +41,7 @@ The evidence is in how the product fails today: the personas already produce exc
 5. **Resume capability** — Runs persist to a git-backed store and resume from the last point after interruption.
 6. **Legible loop** — A live event log, working-tree diff panel, plan tracker, and real-time cost meter show what happened, when, and why.
 7. **Operator-in-the-loop** — The agent can call `ask_operator` to pause and ask a clarifying question; the operator answers from a modal (TUI). Headless `corbits exec` unmounts `ask_operator` when stdin/stdout are not TTYs. TTY exec still reads a single line from stdin.
-8. **Mid-run steering** — Two modes while the agent is running, keyed to **whose** idle. **Parent-idle** is when the primary dispatch turn is not inside an in-flight parent tool; **session-idle** is parent-idle **and** no live fleet lanes. **Enter** soft-steers while the parent is busy — delivers at the next **parent** `tool.boundary` without stopping the current run; a long parent `run_shell` is parent-busy, so Enter is a queued steer, not a new turn. A queued steer delivers at the next parent `tool.boundary` so occupancy can pick it up. Idle-with-fleet is shipped: after a non-blocking `spawn_agent` dispatch the parent goes idle while workers keep running, mailbox mail arrives as inbound when a worker finishes or fails, and mid-hold Enter starts a new primary turn instead of queueing a steer. **Alt+Enter** queues a follow-up delivered only on session-idle (`run` goes idle; does not interrupt). Session-idle Alt+Enter is a no-op. **Ctrl+C** stops the run outright. The notice row shows distinct `steer N` / `follow-up M` badges, and held messages list in a pending column stacked on the prompt box — `↑`/`↓` select, `Enter` force-pushes one now, `Ctrl+X` drops it — instead of echoing labelled transcript rows; when steers are pending and a parent tool has been in flight a few seconds, the notice names that command. Shortcuts are listed in `/help` (`Enter` soft-steer · `Alt+Enter` follow-up · `Ctrl+C` stop).
+8. **Mid-run steering** — Two modes while the agent is running, keyed to **whose** idle. **Parent-idle** is when the primary dispatch turn is not inside an in-flight parent tool; **session-idle** is parent-idle **and** no live fleet lanes. **Enter** soft-steers while the parent is busy — delivers at the next **parent** `tool.boundary` without stopping the current run; a long parent `run_shell` is parent-busy, so Enter is a queued steer, not a new turn. A queued steer delivers at the next parent `tool.boundary` so occupancy can pick it up. Idle-with-fleet is shipped: after a non-blocking `spawn_agent` dispatch the parent goes idle while workers keep running, mailbox mail arrives as inbound when a worker finishes or fails, and mid-hold Enter starts a new primary turn instead of queueing a steer. **Alt+Enter** queues a follow-up delivered only on session-idle (`run` goes idle; does not interrupt). Session-idle Alt+Enter is a no-op. **Ctrl+C** pauses: stop the primary, hold the queue, stay idle until an explicit operator send. The notice row shows distinct `steer N` / `follow-up M` badges, and held messages list in a pending column stacked on the prompt box — `↑`/`↓` select, `Enter` force-pushes one now, `Ctrl+X` drops it — instead of echoing labelled transcript rows; when steers are pending and a parent tool has been in flight a few seconds, the notice names that command. Shortcuts are listed in `/help` (`Enter` soft-steer · `Alt+Enter` follow-up · `Ctrl+C` pause).
 9. **Orchestrator-only (TUI + exec)** — The primary session is always the orchestrator: it can act directly and delegates via `spawn_agent` (then idle; mailbox mail inbound) / `search_agents`. Nested orchestrators collect through mailbox mail the same way. Long jobs belong on workers — a parent that runs them itself stays parent-busy and holds Enter steers. Single-agent session mode, the first-run mode picker, and Settings → Session are gone (CL-5814). Legacy `sessionMode` values on disk are ignored.
 
 ## User Experience

docs/TUI.md

@@ -693,7 +693,9 @@ nearest the prompt, first selected — stay visible, and a guidance row naming
 its keys (`pending-column.ts`, painted by `syncPendingRows` in `chrome.ts`).
 The transcript only ever sees the item that actually delivers, as an ordinary
 user row — pending/delivery labels (`[will steer next]`, `[steering]`,
-`[following up]`) are gone on purpose.
+`[following up]`) are gone on purpose. After a first Ctrl+C pause the column
+is the durable surface: held items stay pending and deliver only on the next
+explicit operator send.
 
 While the column has items, `↑` at the prompt buffer's top edge selects the
 newest held item and `↑`/`↓` walk the rows; `↓` past the last row hands the
@@ -736,26 +738,26 @@ as id/status/description only. The mail and that fleet-dry continuation
 are runtime-to-agent traffic — the fleet board owns worker status — so
 neither paints a transcript row, and neither rehydrates as one.
 
-Interrupting (Ctrl+C) never discards a queued or steered message. It used to
-— the transcript literally said `interrupt — discarded N pending`, and an
-operator who queued an instruction and then lost patience destroyed the very
-thing they were trying to deliver. It now reports `interrupt — N pending
-kept`: the run stops, the queue survives, and those messages are handed over
-at the interrupt itself (`doInterrupt` drains after `port.interrupt()`), not
-left waiting on an idle event the stop may never produce (`interrupt` in
-`delivery-queue.ts` no longer clears `items`).
+Interrupting (Ctrl+C) never discards a queued or steered message. The first
+press is a pause: it stops the primary, holds the queue, and reports
+`N pending kept` (`interrupt` in `delivery-queue.ts` does not clear `items`).
+Held items stay in the pending column and deliver only on the next explicit
+operator send. Stall and expire aborts do not pause, so they still drain.
+Occupancy and ask-wake do not start a new primary while paused. A compaction
+continuation hop is deferred while paused; it stays consume-once-intact and
+delivers after resume (an explicit operator send), not dropped.
 
 **Fleet agent lanes on redirect.** Soft steer (Enter mid-run) and follow-up
 (queued drain) leave running workers alone — they never call
 `runner.ts`'s `interrupt()`, so the parent's operation signal stays live and
-spawned workers keep running. Ctrl+C is the explicit fleet teardown:
-`doInterrupt` → `port.interrupt()` → `currentAgent.close()` aborts the shared
-operation signal and routes cancellation through the fleet/session-store
-teardown path, so in-flight fleet-agent dispatch reports back as cancelled by
-the operator rather than being left to finish silently detached. `/clear` and
+spawned workers keep running. The first Ctrl+C pauses the parent and does not
+tear down the fleet. A second press while workers are live stops those workers
+(`cancelLiveWorkers`); the app stays up and the held queue is kept. A third
+press quits. With no live workers, the second press still quits. `/clear` and
 session exit still call `subAgentSessions.cancelAll` for an explicit
 session-wide cancel; that path is separate from interrupt and must stay off
-the soft-steer / follow-up gestures.
+the soft-steer / follow-up gestures. Quit still full-teardowns, including
+wiping the delivery queue.
 
 Up/Down are caret motion first inside a multi-line buffer — except while the
 pending column is engaged, when ↑ at the buffer's top edge selects a held item
@@ -813,21 +815,19 @@ The prompt repaints on every keystroke (`onFrame` in `src/tui/shell/index.ts` ca
 not on a debounce) — anything added to the prompt's paint path must stay
 cheap, because it runs at typing speed.
 
-Ctrl+C interrupts a busy run, or clears idle prompt text and pending
+Ctrl+C pauses a busy run (stop the primary, hold the queue, stay idle until
+an explicit operator send), or clears idle prompt text and pending
 attachments. Clearing prompt text arms a 2-second quit window
-(`CTRL_C_EXIT_WINDOW_MS`); clearing attachments alone does not. A second
-Ctrl+C while the window is open quits — this
-replaced an Ink-era yes/no exit-confirm modal with the same intent (an
-explicit second confirmation) without adding a modal (`handleCtrlC`,
-`src/tui/shell/prompt.ts`). See "Soft steer vs. follow-up" above for the two
-mid-run gestures and what interrupting does to fleet-agent lanes. The interrupt
-keeps whatever is sitting in the queue rather than discarding it — the
-operator typed those messages meaning them delivered, not meaning "cancel
-this run and also throw away what I typed"; the transcript row says so
-(`"N pending kept"`). Kept items are handed over at the
-interrupt itself (`doInterrupt` in `runtime-bridge.ts` drains after
-`port.interrupt()`), serialized behind the agent rebuild the stop starts —
-a stop does not reliably produce an idle event to drain against later.
+(`CTRL_C_EXIT_WINDOW_MS`); clearing attachments alone does not. With live
+sub-agents, a second Ctrl+C in the window stops the workers and keeps the
+app up (queue kept); a third quits. With no live workers, the second press
+still quits. The extra press is the confirmation, with no modal
+(`handleCtrlC`, `src/tui/shell/keys.ts`). See "Soft steer vs. follow-up"
+above for the two mid-run gestures and what interrupting does to
+fleet-agent lanes. The pause keeps whatever is sitting in the queue rather
+than discarding it — the transcript row says so (`"N pending kept"`). Held
+items deliver only on the next explicit send; occupancy and ask-wake do
+not start a new primary while paused.
 
 ## Overflows, scrolling, and key macros
 

src/subagent/fleet-dry-drive.test.ts

@@ -680,6 +680,31 @@ describe("driveOpenTasksAfterFleetDry", () => {
     ).toBe(false);
   });
 
+  test("does not send if begin does not claim the parent turn", async () => {
+    const records = recordsOf({
+      w1: { status: "done", report: "ok" },
+    });
+    const sends: string[] = [];
+    expect(
+      await driveOpenTasksAfterFleetDry({
+        previousRunning: 1,
+        running: 0,
+        openTasks: [openTask],
+        parentProcessing: false,
+        isParentProcessing: () => false,
+        mailbox: peekMailbox(records),
+        lanes: [],
+        beginSystemContinuation: () => undefined,
+        send: (prompt) => {
+          sends.push(prompt);
+          return ACCEPTED_DELIVERY;
+        },
+      }),
+    ).toBe(false);
+    expect(sends).toEqual([]);
+    expect(records.get("w1")?.collected).not.toBe(true);
+  });
+
   test("deferred dry edge after parentProcessing still collects and sends", async () => {
     const records = recordsOf({
       w1: { status: "done", report: "ok" },

src/subagent/fleet-dry-drive.ts

@@ -543,6 +543,8 @@ export function driveOpenTasksAfterFleetDry(args: {
   running?: number | undefined;
   openTasks: readonly Task[];
   parentProcessing: boolean;
+  /** Live check after awaited collect; occupancy must not start a turn already claimed. */
+  isParentProcessing?: () => boolean;
   deferredDryEdge?: boolean;
   mailbox: FleetDryMailbox | undefined;
   lanes: readonly FleetDryLane[];
@@ -601,6 +603,13 @@ async function driveOpenTasksAfterFleetDrySpill(
     const reports = dedupeByAgentId([...digested, ...stubs]);
     prompt = buildFleetDryContinuationPrompt(tasks, reports);
     args.beginSystemContinuation(prompt);
+    // begin is a no-op while paused. Do not send; release the delivering
+    // claim so a later explicit send can still deliver. The dry latch
+    // unclaims when this drive returns false.
+    if (args.isParentProcessing?.() === false) {
+      releaseOccupancyDelivering(args.mailbox, takeIds);
+      return false;
+    }
   } catch {
     return fail();
   }

src/subagent/mailbox-mail-drive.test.ts

@@ -578,6 +578,28 @@ describe("driveMailboxMail", () => {
     expect(records.get("w1")?.collected).not.toBe(true);
   });
 
+  test("does not send if begin does not claim the parent turn", async () => {
+    const records = recordsOf({
+      w1: { status: "done", report: "ok" },
+    });
+    const sends: string[] = [];
+    expect(
+      await driveMailboxMail({
+        parentProcessing: false,
+        isParentProcessing: () => false,
+        mailbox: collectingMailbox(records),
+        lanes: [],
+        beginSystemContinuation: () => undefined,
+        send: (prompt) => {
+          sends.push(prompt);
+          return ACCEPTED_DELIVERY;
+        },
+      }),
+    ).toBe(false);
+    expect(sends).toEqual([]);
+    expect(records.get("w1")?.collected).not.toBe(true);
+  });
+
   test("not-delivered send leaves the wake retryable", async () => {
     const records = await driveMailNotDelivered();
     expect(

src/subagent/mailbox-mail-drive.ts

@@ -179,6 +179,13 @@ async function driveMailboxMailAfterCollect(
       await digestCollectedReports(reports, args.writeBlob),
     );
     args.beginSystemContinuation(prompt);
+    // begin is a no-op while paused (and if disposed/empty). Do not send a
+    // primary, and release the delivering claim so a later explicit send can
+    // still deliver. The mailbox latch unclaims when this drive settles.
+    if (args.isParentProcessing?.() === false) {
+      releaseOccupancyDelivering(args.mailbox, ids);
+      return false;
+    }
   } catch {
     return fail();
   }

src/tui/agent-ask-wake.test.ts

@@ -11,8 +11,10 @@ import {
   createDeliveryGeneration,
   createLeftoverSend,
   createSessionOperationQueue,
+  isPaused,
   routeQueuedDelivery,
 } from "./delivery-queue.js";
+import { handleCtrlC } from "./shell/keys.js";
 import { ingestOperatorPrompt } from "./prompt-attachments.js";
 import {
   armFeedbackCapture,
@@ -623,6 +625,106 @@ describe("agent ask wake delivery", () => {
     });
   }
 
+  test("operator pause does not flush a stashed ask or mailbox occupancy", async () => {
+    await withAppShell(
+      async (shell) => {
+        const sends: string[] = [];
+        const bridge = attachSessionBridge(
+          shell,
+          createLiveSessionPort({
+            send: (text) => {
+              sends.push(text);
+            },
+            deliver: (text) => {
+              sends.push(text);
+            },
+            interrupt: () => undefined,
+          }),
+        );
+        try {
+          const order = trackMailOrder(bridge, true);
+          bridge.handle({ type: "inference.start", data: {} });
+          bridge.handle({
+            type: "inference.text.delta",
+            data: { token: "ok" },
+          });
+          bridge.handle({ type: "agent-ask", asks: [wake("a1", "q1")] });
+          expect(sends).toEqual([]);
+
+          handleCtrlC(shell);
+          expect(isPaused(shell.session)).toBe(true);
+          expect(sends).toEqual([]);
+          expect(order).toEqual([]);
+          expect(bridge.turn.isProcessing).toBe(false);
+
+          bridge.handle({ type: "agent-ask", asks: [wake("a2", "q2")] });
+          expect(sends).toEqual([]);
+          expect(order).toEqual([]);
+          expect(bridge.turn.isProcessing).toBe(false);
+        } finally {
+          bridge.dispose();
+        }
+      },
+      { shell: { run: "busy" } },
+    );
+  });
+
+  test("paused beginSystemContinuation does not mark the turn processing", async () => {
+    await withWakeBridge((bridge, sends, shell) => {
+      handleCtrlC(shell);
+      expect(isPaused(shell.session)).toBe(true);
+      bridge.beginSystemContinuation("mailbox occupancy");
+      expect(bridge.turn.isProcessing).toBe(false);
+      expect(shell.session.run).not.toBe("busy");
+      expect(sends).toEqual([]);
+    });
+  });
+
+  test("in-flight mailbox collect after pause does not start a primary", async () => {
+    await withWakeBridge(async (bridge, sends, shell) => {
+      const order: string[] = [];
+      let release: () => void = () => undefined;
+      const hold = new Promise<void>((resolve) => {
+        release = resolve;
+      });
+      const driver = latchAsyncMailboxMail(bridge, order, hold);
+      bridge.setMailboxMailDriver(driver);
+      bridge.flushMailboxMail();
+      expect(order).toEqual(["mail"]);
+      expect(driver.claimed()).toBe(true);
+      expect(bridge.turn.isProcessing).toBe(false);
+
+      handleCtrlC(shell);
+      expect(isPaused(shell.session)).toBe(true);
+
+      release();
+      await hold;
+      await Promise.resolve();
+      await Promise.resolve();
+
+      expect(order).toEqual(["mail", "begin"]);
+      expect(bridge.turn.isProcessing).toBe(false);
+      expect(sends).toEqual([]);
+      expect(shell.session.run).not.toBe("busy");
+      expect(isPaused(shell.session)).toBe(true);
+      expect(driver.claimed()).toBe(false);
+    });
+  });
+
+  test("unpaused interrupt still begins a system continuation", async () => {
+    await withWakeBridge((bridge, sends, shell) => {
+      bridge.handle({ type: "inference.start", data: {} });
+      expect(bridge.turn.isProcessing).toBe(true);
+      bridge.interrupt();
+      expect(isPaused(shell.session)).toBe(false);
+      expect(bridge.turn.isProcessing).toBe(false);
+      bridge.beginSystemContinuation("mailbox occupancy");
+      expect(bridge.turn.isProcessing).toBe(true);
+      expect(shell.session.run).toBe("busy");
+      expect(sends).toEqual([]);
+    });
+  });
+
   test("a wake question with bracket lines does not spoof attachment-echo matching", async () => {
     await withWakeBridge((bridge, sends, shell) => {
       const ask = {

src/tui/delivery-queue-pause.test.ts

@@ -0,0 +1,56 @@
+import { describe, expect, test } from "bun:test";
+import {
+  createSessionQueue,
+  enqueue,
+  enqueueSteer,
+  interrupt,
+  isPaused,
+  pause,
+  resumeForSend,
+  badgeCount,
+} from "./delivery-queue";
+
+/**
+ * Phase 1 (CL-10149) pins the `paused` state and its pure helpers. The drain
+ * gate that consumes this flag ships in Phase 2; here we assert only the state
+ * transitions — pause never drops items and resumeForSend clears the flag
+ * without touching the queue.
+ */
+describe("delivery-queue pause", () => {
+  test("createSessionQueue defaults to not paused", () => {
+    const s = createSessionQueue("busy");
+    expect(isPaused(s)).toBe(false);
+  });
+
+  test("pause after interrupt holds queue items and sets paused", () => {
+    let s = createSessionQueue("busy");
+    s = enqueue(s, "follow-up");
+    s = enqueueSteer(s, "steer");
+    s = interrupt(s);
+    s = pause(s);
+    expect(isPaused(s)).toBe(true);
+    // NG2: pause keeps every queued item.
+    expect(badgeCount(s)).toBe(2);
+    expect(s.items.map((i) => i.text)).toEqual(["follow-up", "steer"]);
+  });
+
+  test("resumeForSend clears paused and keeps items intact", () => {
+    let s = createSessionQueue("busy");
+    s = enqueue(s, "follow-up");
+    s = pause(s);
+    expect(isPaused(s)).toBe(true);
+    const resumed = resumeForSend(s);
+    expect(isPaused(resumed)).toBe(false);
+    // No item is dropped by the resume.
+    expect(resumed.items.map((i) => i.text)).toEqual(["follow-up"]);
+    expect(badgeCount(resumed)).toBe(1);
+  });
+
+  test("enqueue while paused accepts items and does not auto-resume", () => {
+    let s = createSessionQueue("busy");
+    s = pause(s);
+    s = enqueue(s, "queued-while-paused");
+    expect(isPaused(s)).toBe(true);
+    expect(badgeCount(s)).toBe(1);
+  });
+});