docs(backlog): add semantic segmentation requirement

Add requirement for improving SRT semantic segmentation:
- Current rule-based approach cannot identify topic transitions
- Document 4 potential solutions with trade-offs
- Recommended approach: two-stage with optional LLM

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

.planning/backlog/semantic-segmentation.md

@@ -0,0 +1,60 @@
+# Semantic Segmentation for SRT Processing
+
+**Status:** Backlog
+**Priority:** Medium
+**Source:** User feedback during v1.5 testing
+
+## Problem
+
+Current `SemanticSegmenter` uses simple heuristic rules:
+- Time gap > 3 seconds
+- Paragraph length > 500 characters
+
+This approach cannot identify **semantic topic transitions**.
+
+### Example Case
+
+Video: BV1MiPSz2ErY (科学透镜 - 5 topics in one video)
+1. 银河系中心的超大黑洞人马座A*其实不是黑洞
+2. 银河系和仙女座星系等构成的本星系群陷入巨大物质薄片
+3. 科学家为世界上97%的建筑生成3D模型
+4. 雷暴的形成
+5. 闪电的形成
+
+Current segmentation mixes these topics together.
+
+## Potential Solutions
+
+### Option 1: Transition Keyword Detection (Simple)
+- Detect common transition phrases: "接下来", "另外", "第二个", etc.
+- **Pros:** Fast, no API calls
+- **Cons:** Low accuracy, many videos lack explicit transitions
+
+### Option 2: LLM-Based Segmentation (Accurate)
+- Ask LLM to analyze entire subtitle and identify topic boundaries
+- **Pros:** Highest accuracy, understands semantics
+- **Cons:** Extra API cost, token limits for long videos
+
+### Option 3: Two-Stage Approach (Balanced)
+- Stage 1: Simple rule-based segmentation (current)
+- Stage 2: LLM checks paragraph beginnings for new topics
+- **Pros:** Most cases use simple rules, LLM only for key decisions
+- **Cons:** More complex implementation
+
+### Option 4: Embedding-Based Similarity
+- Calculate semantic similarity between adjacent paragraphs
+- Low similarity = topic boundary
+- **Pros:** No LLM API calls, runs locally
+- **Cons:** Requires embedding model, implementation complexity
+
+## Recommendation
+
+**Phase 1 (Quick Win):** Transition keyword detection + improve prompt
+
+**Phase 2 (Full Solution):** Two-stage approach with optional LLM semantic segmentation
+
+## Success Criteria
+
+1. Topics are separated correctly
+2. Each paragraph covers a single coherent topic
+3. No false splits within a topic

kb_workflow/deep_analyzer/__init__.py

@@ -116,6 +116,14 @@ class SimpleAIClient:
         data = response.json()
         message = data.get('choices', [{}])[0].get('message', {})
 
+        # 调试：打印完整的 API 响应结构
+        print(f"[DEBUG] disable_reasoning={disable_reasoning}", file=sys.stderr)
+        print(f"[DEBUG] message keys: {list(message.keys())}", file=sys.stderr)
+        if 'content' in message:
+            print(f"[DEBUG] content 长度: {len(message['content'])}, 前200字符: {message['content'][:200]}", file=sys.stderr)
+        if 'reasoning_content' in message:
+            print(f"[DEBUG] reasoning_content 长度: {len(message['reasoning_content'])}, 前200字符: {message['reasoning_content'][:200]}", file=sys.stderr)
+
         # 返回 content 字段
         # 当禁用思考模式时，content 包含直接答案
         # 当启用思考模式时，content 可能包含思考过程和分析

kb_workflow/srt/corrector.py

@@ -26,7 +26,11 @@ class AIClientProtocol(Protocol):
 
 # 纠错 Prompt 模板
 CORRECTION_PROMPT = """你是一个文本校对专家。以下是语音识别生成的字幕，可能包含识别错误。
-请只纠正明显的错误（如同音字、漏字、错字），不要改变原意和风格。
+
+请执行以下任务：
+1. 纠正明显的错误（如同音字、漏字、错字）
+2. 添加适当的标点符号（逗号、句号、顿号等）
+3. 不要改变原意和风格
 
 原文：
 {text}