feat(telemetry,ui): reasoning span attribute + collapsed thought summary (Phase 1 of #162)

packages/cli/src/ui/components/messages/ConversationMessages.test.tsx

@@ -0,0 +1,73 @@
+/**
+ * @license
+ * Copyright 2025 Qwen Team
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { describe, expect, it } from 'vitest';
+import { render } from 'ink-testing-library';
+import { ThinkMessage, ThinkMessageContent } from './ConversationMessages.js';
+
+describe('ThinkMessage', () => {
+  it('renders the streaming text expanded while pending', () => {
+    const { lastFrame } = render(
+      <ThinkMessage
+        text="Let me consider this carefully and weigh the options."
+        isPending={true}
+        contentWidth={80}
+      />,
+    );
+    const output = lastFrame() ?? '';
+    expect(output).toContain('Let me consider this');
+    // Streaming render uses the existing ⟡ glyph, not the ▸ summary marker.
+    expect(output).toContain('⟡');
+    expect(output).not.toContain('thinking (');
+  });
+
+  it('renders compact "thinking (N chars)" summary once stream finalizes', () => {
+    const text = 'reasoning '.repeat(10).trim(); // 99 chars
+    const { lastFrame } = render(
+      <ThinkMessage text={text} isPending={false} contentWidth={80} />,
+    );
+    const output = lastFrame() ?? '';
+    expect(output).toContain('▸');
+    expect(output).toContain(`thinking (${text.length} chars)`);
+    // Underlying reasoning text is not rendered inline post-stream.
+    expect(output).not.toContain('reasoning reasoning');
+  });
+
+  it('formats large char counts with thousands separator', () => {
+    const text = 'x'.repeat(12_345);
+    const { lastFrame } = render(
+      <ThinkMessage text={text} isPending={false} contentWidth={80} />,
+    );
+    const output = lastFrame() ?? '';
+    expect(output).toContain('thinking (12,345 chars)');
+  });
+});
+
+describe('ThinkMessageContent', () => {
+  it('renders the continuation text while pending', () => {
+    const { lastFrame } = render(
+      <ThinkMessageContent
+        text="continued reasoning text"
+        isPending={true}
+        contentWidth={80}
+      />,
+    );
+    const output = lastFrame() ?? '';
+    expect(output).toContain('continued reasoning text');
+  });
+
+  it('renders nothing once stream finalizes (the parent ThinkMessage owns the summary)', () => {
+    const { lastFrame } = render(
+      <ThinkMessageContent
+        text="continued reasoning text"
+        isPending={false}
+        contentWidth={80}
+      />,
+    );
+    const output = lastFrame() ?? '';
+    expect(output.trim()).toBe('');
+  });
+});

packages/cli/src/ui/components/messages/ConversationMessages.tsx

@@ -227,35 +227,70 @@ export const AssistantMessageContent: React.FC<
   />
 );
 
+// Post-stream summary line ("▸ thinking (N chars)"). Phase 1 of the reasoning
+// rendering work (see #162): full text remains live in Langfuse, ACP
+// `agent_thought_chunk` notifications, and ChatRecord for back-compat. An
+// in-TUI expand affordance is a follow-up. Note: when a long thought was
+// split mid-stream into a gemini_thought + gemini_thought_content pair, this
+// counts only the first chunk — the continuation renders nothing (see below).
+// True total requires post-finalize coalescing in useGeminiStream and is
+// deferred since splits are rare and the count is a hint, not a contract.
+const ThinkSummary: React.FC<{ text: string }> = ({ text }) => {
+  const charCount = text.length;
+  return (
+    <PrefixedTextMessage
+      text={`thinking (${charCount.toLocaleString()} chars)`}
+      prefix="▸"
+      prefixColor={theme.text.secondary}
+      textColor={theme.text.secondary}
+    />
+  );
+};
+
 export const ThinkMessage: React.FC<ThinkMessageProps> = ({
   text,
   isPending,
   availableTerminalHeight,
   contentWidth,
-}) => (
-  <PrefixedMarkdownMessage
-    text={text}
-    prefix="⟡"
-    prefixColor={theme.text.secondary}
-    isPending={isPending}
-    availableTerminalHeight={availableTerminalHeight}
-    contentWidth={contentWidth}
-    textColor={theme.text.secondary}
-  />
-);
+}) => {
+  if (!isPending) {
+    return <ThinkSummary text={text} />;
+  }
+  return (
+    <PrefixedMarkdownMessage
+      text={text}
+      prefix="⟡"
+      prefixColor={theme.text.secondary}
+      isPending={isPending}
+      availableTerminalHeight={availableTerminalHeight}
+      contentWidth={contentWidth}
+      textColor={theme.text.secondary}
+    />
+  );
+};
 
 export const ThinkMessageContent: React.FC<ThinkMessageContentProps> = ({
   text,
   isPending,
   availableTerminalHeight,
   contentWidth,
-}) => (
-  <ContinuationMarkdownMessage
-    text={text}
-    isPending={isPending}
-    availableTerminalHeight={availableTerminalHeight}
-    contentWidth={contentWidth}
-    basePrefix="⟡"
-    textColor={theme.text.secondary}
-  />
-);
+}) => {
+  // When the stream has finalized, suppress the continuation block. The
+  // adjacent ThinkMessage already renders the summary line; rendering this
+  // continuation as another summary would double-count and drop chars across
+  // the split boundary. Streaming-time renders unchanged so live thoughts
+  // still appear.
+  if (!isPending) {
+    return null;
+  }
+  return (
+    <ContinuationMarkdownMessage
+      text={text}
+      isPending={isPending}
+      availableTerminalHeight={availableTerminalHeight}
+      contentWidth={contentWidth}
+      basePrefix="⟡"
+      textColor={theme.text.secondary}
+    />
+  );
+};

packages/core/src/agents/runtime/agent-core.ts

@@ -471,10 +471,14 @@ export class AgentCore {
               messagesAfter: masked.length,
             });
           }
-          const compacted =
-            estimateTokens(masked) <= targetTokens
-              ? masked
-              : compactMessages(masked, targetTokens);
+          let compacted: Content[];
+          if (estimateTokens(masked) <= targetTokens) {
+            compacted = masked;
+          } else {
+            const taskStore = this.runtimeContext.getTaskStore?.();
+            const result = compactMessages(masked, targetTokens, { taskStore });
+            compacted = result instanceof Promise ? await result : result;
+          }
           if (compacted.length < historyBefore.length) {
             chat.setHistory(compacted);
             const tokensAfter = estimateTokens(compacted);