Merge branch 'main' into cl-6898-move-live-agents-into-a-chrome-strip-above-the-prompt

Author: TheGreatAxios

CHANGELOG.md

@@ -13,6 +13,16 @@ parallel copies under `docs/` or `scripts/notes/`. At cut time: rename
 
 ## [Unreleased]
 
+### TUI
+
+- **Taller live chain-of-thought preview.** Parent reasoning still paints
+  through the existing thinking row (one fold per turn, settle-to-opener +
+  expand) — no separate mid-turn stream lane. The hard-capped live wrap rises
+  from 3 to 10 inset lines (`LIVE_THINKING_MAX_LINES`) so mid-turn CoT is
+  glanceable; reveal rate stays 28 chars/sec. Sub-agent Task-row thinking is
+  unchanged. Assistant mid-turn text continues to grow the open streaming
+  assistant row from `inference.text.delta`.
+
 ### Fixed
 
 - **Codex Responses no longer sends `reasoning.summary: "auto"`.** ChatGPT

docs/TUI.md

@@ -64,6 +64,16 @@ stays easy to find while scrolling through denser assistant and tool rows —
 the pad is part of the bubble itself, not an extra turn-boundary gap, and
 assistant/tool rows are unchanged.
 
+Parent live reasoning paints through the existing thinking row — never a
+third mid-turn stream lane. While `inference.thinking.delta` arrives,
+`thinkingLivePreviewLines` (`src/tui/thinking.ts`) wraps the newest revealed
+prose into a hard-capped inset paragraph (`LIVE_THINKING_MAX_LINES`, currently 10) at a bounded reveal rate (`REVEAL_CHARS_PER_SEC`). When the turn moves on
+(assistant text, a tool call, or settle), the row collapses to its opening
+clause with the rest behind expand. Mid-turn thinking bursts fold onto that
+same one row per turn (`reasoning-fold`); `inference.text.delta` grows the
+open assistant streaming row in place. Sub-agent Task-row thinking is a
+separate path and is unchanged by this preview.
+
 The prompt box's border carries the metadata that would otherwise cost a
 titlebar row: the model label sits right-aligned in the top rule as
 `profile · model · effort` (empty segments omitted), and a

src/tui/collapse.test.ts

@@ -21,7 +21,7 @@ import {
   type RowLayout,
   type StreamRow,
 } from "./stream";
-import { thinkingLivePreviewLines, thinkingSettledLine } from "./thinking";
+import { thinkingLivePreviewLines, thinkingSettledLine, LIVE_THINKING_MAX_LINES } from "./thinking";
 import { describeView, toolArgsView } from "./tool-args";
 
 const WIDE: RowLayout = { width: 96, multiAgent: false };
@@ -171,7 +171,7 @@ describe("reasoning collapses to a short wrapped preview", () => {
   test("while thinking it wraps a short preview instead of sideways-scrolling", () => {
     const painted = lines({ role: "system", meta: "thinking", text, streaming: true });
     expect(painted.length).toBeGreaterThanOrEqual(1);
-    expect(painted.length).toBeLessThanOrEqual(3);
+    expect(painted.length).toBeLessThanOrEqual(LIVE_THINKING_MAX_LINES);
     // Inset and dim is the whole of reasoning's chrome; it carries no rail.
     expect(painted.every((line) => !line.includes("┆"))).toBe(true);
     expect(painted.join("\n")).toContain("one commit");

src/tui/runtime-bridge.test.ts

@@ -585,6 +585,53 @@ describe("attachSessionBridge", () => {
     );
   });
 
+  test("inference.text.delta opens a live assistant streaming row mid-turn", async () => {
+    await withTestRenderer(
+      async (h) => {
+        const shell = createAppShell(h.renderer, {
+          terminal: { columns: 80, rows: 24 },
+          wireKeys: false,
+          run: "idle",
+        });
+        const bridge = attachSessionBridge(shell, createRecordingPort());
+        try {
+          bridge.handle({ type: "inference.start", data: {} });
+          bridge.handle({
+            type: "inference.thinking.delta",
+            data: { token: "planning the reply" },
+          });
+          bridge.handle({
+            type: "inference.tool_call.end",
+            data: { name: "run_shell", callId: "c1", arguments: "{}" },
+          });
+          bridge.handle({
+            type: "tool.done",
+            data: { result: { callId: "c1", content: "ok", isError: false } },
+          });
+          bridge.handle({
+            type: "inference.text.delta",
+            data: { token: "Here is " },
+          });
+          bridge.handle({
+            type: "inference.text.delta",
+            data: { token: "the answer." },
+          });
+
+          const assistant = shell.streamLog.filter((r) => r.role === "assistant");
+          expect(assistant).toHaveLength(1);
+          expect(assistant[0]?.streaming).toBe(true);
+          expect(assistant[0]?.text).toBe("Here is the answer.");
+          // Still one thinking row for the turn — no third mid-turn stream lane.
+          expect(shell.streamLog.filter((r) => r.meta === "thinking")).toHaveLength(1);
+        } finally {
+          bridge.dispose();
+          shell.dispose();
+        }
+      },
+      { width: 80, height: 24 },
+    );
+  });
+
   test("thinking deltas coalesce and never become plain system rows", async () => {
     await withTestRenderer(
       async (h) => {

src/tui/stream.ts

@@ -401,9 +401,10 @@ function elapsedLabel(ms: number): string {
 }
 
 /**
- * Reasoning body. While text arrives it wraps into a short inset paragraph of
- * the newest revealed prose (no sideways scroll). Once the turn moves on it
- * collapses to the opening clause; the rest is behind the expand key.
+ * Reasoning body. While text arrives it wraps into a bounded inset paragraph of
+ * the newest revealed prose (no sideways scroll; hard line cap). Once the turn
+ * moves on it collapses to the opening clause; the rest is behind the expand
+ * key.
  *
  * A row with no settled thought (a hydrated transcript, a fixture) has no
  * summary to collapse to and keeps the plain block.

src/tui/thinking-reveal.test.ts

@@ -7,7 +7,7 @@
 
 import { describe, expect, test } from "bun:test";
 
-import { advanceRevealChars, thinkingLivePreviewLines } from "./thinking";
+import { advanceRevealChars, LIVE_THINKING_MAX_LINES, thinkingLivePreviewLines } from "./thinking";
 import { withTestRenderer } from "./harness";
 import { attachSessionBridge, createRecordingPort } from "./runtime-bridge";
 import { createAppShell } from "./shell";
@@ -83,12 +83,23 @@ describe("thinkingLivePreviewLines with a reveal position", () => {
 
   test("omitting revealChars wraps whatever has arrived, capped to max lines", () => {
     const lines = thinkingLivePreviewLines(text, 10);
-    expect(lines.length).toBeLessThanOrEqual(3);
+    expect(lines.length).toBeLessThanOrEqual(LIVE_THINKING_MAX_LINES);
     expect(lines.length).toBeGreaterThan(0);
     expect(lines.every((line) => line.length <= 10)).toBe(true);
     expect(lines.join(" ")).toContain("running");
   });
 
+  test("a long burst fills more than three lines and still respects the hard cap", () => {
+    const long = Array.from({ length: 40 }, (_, i) => `clause-${i}`).join(" ");
+    const lines = thinkingLivePreviewLines(long, 20);
+    expect(lines.length).toBeGreaterThan(3);
+    expect(lines.length).toBeLessThanOrEqual(LIVE_THINKING_MAX_LINES);
+    expect(lines.every((line) => line.length <= 20)).toBe(true);
+    // Newest prose wins when the wrap exceeds the cap.
+    expect(lines.join(" ")).toContain("clause-39");
+    expect(lines.join(" ")).not.toContain("clause-0");
+  });
+
   test("sample frames across a few rates, printed for eyeballing", () => {
     const sample = "we need to check whether the cache key already accounts for the locale";
     for (const rate of [15, 20, 28, 40, 60]) {

src/tui/thinking.ts

@@ -4,8 +4,9 @@
  *
  * Reasoning is not the answer, so it never owns the screen. Live text used to
  * ride a single sideways-scrolling row; that was unreadable. Now the newest
- * revealed prose wraps into a few inset lines. Once the turn moves on the row
- * collapses to its opening clause — same expand path as before.
+ * revealed prose wraps into a bounded inset paragraph (hard-capped — never an
+ * unbounded dump). Once the turn moves on the row collapses to its opening
+ * clause — same expand path as before.
  */
 
 import { sliceToWidth, stringWidth, wrapLines } from "./view/height.js";
@@ -28,12 +29,18 @@ export function flattenReasoningText(text: string): string {
  * Characters per second the reveal position advances at while reasoning
  * streams. Picked by printing sample frames at 15/20/28/40/60 chars/sec and
  * reading them back: below ~20 the line feels laggy against a fast model,
- * above ~40 it is back to unreadable. 28 landed as fast-but-legible.
+ * above ~40 it is back to unreadable. 28 landed as fast-but-legible and still
+ * reads well against the taller live preview.
  */
 export const REVEAL_CHARS_PER_SEC = 28;
 
-/** How many wrapped lines a live reasoning preview may claim. */
-export const LIVE_THINKING_MAX_LINES = 3;
+/**
+ * How many wrapped lines a live reasoning preview may claim. Hard bound — the
+ * preview never paints unbounded CoT into the transcript. Raised into the
+ * 8–12 band so mid-turn chain-of-thought is glanceable without inventing a
+ * separate stream lane.
+ */
+export const LIVE_THINKING_MAX_LINES = 10;
 
 /**
  * Advance a reveal position toward the text that has actually arrived, capped

src/tui/turn-state.test.ts

@@ -38,6 +38,10 @@ describe("turnStateFromEvent", () => {
     expect(
       fold([{ type: "inference.start" }, { type: "inference.thinking.delta" }]).streamingType,
     ).toBe("thinking");
+    // Canonical bridge alias — fixtures may emit thinking.delta directly.
+    expect(fold([{ type: "inference.start" }, { type: "thinking.delta" }]).streamingType).toBe(
+      "thinking",
+    );
   });
 
   test("text deltas accumulate a live token count, thinking deltas do not", () => {

src/tui/turn-state.ts

@@ -561,6 +561,7 @@ export function turnStateFromEvent(
       return streaming(state, "text", nowMs, deltaText(event));
 
     case "inference.thinking.delta":
+    case "thinking.delta":
       return streaming(state, "thinking", nowMs, deltaText(event));
 
     case "inference.tool_call.delta":