Douyin Analyzer

内容分析模块。对转录文本进行语义分段、提取要点、生成总结。由 agent 直接完成，不依赖脚本。

Security Scan

The skill's code, instructions, and requirements are consistent with a transcript analysis helper; nothing requests unrelated credentials or external installs, though a bundled optional script reads a local config file which you should review before running.

Purpose & Capability

Name and description (semantic segmentation, keypoint extraction, summarization) match the SKILL.md and the included analyzer.py which implements basic cleaning, segmentation, extraction and prompt generation. Minor inconsistency: SKILL.md says the agent performs analysis and does not require scripts, yet a helper script (scripts/analyzer.py) is bundled as an optional fallback (the README states it's currently unused). This is plausible but worth noting.

Instruction Scope

SKILL.md instructs the LLM to operate on Whisper transcripts and defines specific, constrained processing steps and an output format. It does not tell the agent to read unrelated files, call external endpoints, or exfiltrate data. The script is referenced only as a fallback and is not mandated by the instructions.

Install Mechanism

No install spec is provided (instruction-only skill) and no external downloads or package installs are required. Lowest-risk install posture.

Credentials

The skill declares no required env vars or credentials. The bundled script does attempt to load a config from the user's home (~/.openclaw/skills/douyin-config.json and fallback ~/.openclaw/config.json) when executed; reading a local skill config is reasonable but you should review those config files before running the script to confirm they do not contain sensitive values you don't want accessed.

Persistence & Privilege

The skill does not request permanent/always-on installation and does not modify other skills or system-wide settings. Autonomous model invocation is allowed (default) but the skill's scope is limited to text analysis.

Assessment

This skill appears coherent for transcript analysis. Two practical actions before installing or running it: 1) Inspect scripts/analyzer.py (provided) yourself — it is a local, non-networked text-processing helper but it reads ~/.openclaw/skills/douyin-config.json (and a fallback). Confirm those config files are safe and do not contain secrets you don't want accessed. 2) If you only want LLM-only behavior, you can avoid executing the bundled script; the SKILL.md states the agent performs the high-quality analysis directly. If you plan to run the script locally, run it in a restricted environment and review the remainder of the file (the provided listing was truncated) to ensure there are no unexpected operations.

Douyin Analyzer - 内容分析模块

职责

• 输入：Whisper 转录的原始文本
• 输出：结构化的转录稿（分段、修正、要点提取）
• 由 agent（LLM）直接完成，不需要脚本

分析步骤

agent 拿到转录文本后，按以下顺序处理：

1. 修复转录错误

• 修复明显的语音识别错误（如"知固减锁"→"知识库检索"）
• 修复专有名词（如"曼学ai"→"慢学AI"）
• 保留原意，不改变内容

2. 去除口语化表达

• 去除语气词（"嗯"、"啊"、"这个"、"那个"）
• 去除重复啰嗦
• 转为书面语，但保持自然

3. 语义分段

• 按主题/话题分段
• 每段加小标题
• 保持逻辑连贯

4. 提取核心要点

• 列出关键观点
• 标注重要结论

5. 生成标签

• 提取 3-5 个主题标签

输出格式

# {视频标题}

**作者**: {作者名}
**链接**: {原链接}
**转录时间**: {日期}

---

## 完整转录稿

### 一、{第一段主题}

{内容}

### 二、{第二段主题}

{内容}

---

## 核心要点

- 要点1
- 要点2

---

## 标签

#标签1 #标签2

备用脚本

scripts/analyzer.py 提供基础文本清理功能，当前未使用。

python scripts/analyzer.py /path/to/transcript.txt