📦 Chat Distill — 技能工具
v1.0.0Distill a person's chat style from exported conversation records and generate replies that mimic their voice. Use when (1) analyzing chat history to extract...
0· 18·0 当前·0 累计
by 安全扫描
OpenClaw
可疑
medium confidenceThe skill's files and instructions match its stated purpose, but there is a meaningful privacy/scope mismatch and risk that raw chat content (including sensitive items) will be processed or sent to models without enforced redaction.
评估建议
This skill appears to do what it says, but take care before using it:
- Privacy risk: The parser will output raw chat text and the SKILL.md encourages showing concrete examples. The code does not redact sensitive items. Before running, inspect exports and remove or redact passwords, addresses, financial data, or other sensitive content.
- Model transmission risk: Analysis and mimicry typically involve sending extracted text to an LLM. If you use a cloud LLM, any raw chat content you send may be...详细分析 ▾
✓ 用途与能力
Name, description, SKILL.md, reference docs, and the included Python parser all align: this skill parses exported chats, builds style profiles, and generates mimic replies. No unrelated credentials, binaries, or install steps are requested.
⚠ 指令范围
SKILL.md requires reading full chat export files and instructs the agent to include concrete examples from chats in reports, but also claims 'Respect privacy: Never echo sensitive content' — that privacy requirement is a policy instruction only. The included parser (scripts/extract_messages.py) will output raw message text and does not implement automatic redaction or sensitive-data detection. That mismatch means the agent or user could accidentally expose passwords, addresses, or other private data when following the workflow (or when the agent sends data to an external model).
✓ 安装机制
No install spec, no external downloads, only a local Python script and markdown references are included. Lowest-risk installation footprint.
✓ 凭证需求
The skill does not request environment variables, credentials, or config paths. Its need for access is limited to user-provided chat export files (sensitive but expected for this purpose).
✓ 持久化与权限
The skill does not request persistent/always-on privileges. Defaults (no always:true) are used and there is no code that modifies system-wide agent settings.
安全有层次，运行前请审查代码。
运行时依赖
无特殊依赖
版本
latestv1.0.02026/4/16
Initial release: chat style analysis + mimicry from exported chat records
● 无害
安装命令
点击复制官方npx clawhub@latest install chat-distill
镜像加速npx clawhub@latest install chat-distill --registry https://cn.longxiaskill.com镜像同步中
技能文档
Workflow
- Parse → extract messages per speaker from raw export (see
references/format-parsers.md) - Analyze → build style profile (see
references/style-dimensions.md) - Report → output analysis report using template in
references/output-template.md - Mimic → generate replies on demand using the profile
Quick Start
Given a chat export file:
- Read the file and identify the format (WeChat export, plain text, JSON array, TG export).
- Normalize into
{ speaker, text, time? }messages using parsing rules inreferences/format-parsers.md. - Pick the target speaker — the one whose style to learn. If multiple speakers exist, ask which one.
- Run analysis following
references/style-dimensions.md. - Output the report per
references/output-template.md§ Analysis Report. - When the user asks for a mimicked reply, use the profile +
references/output-template.md§ Mimic Reply.
Key Principles
- Show, don't tell: Include concrete examples from the actual chat when reporting style traits.
- Preserve quirks: Capture tics the speaker doesn't notice — repeated filler words, capitalization habits, punctuation style.
- Respect privacy: Never echo sensitive content (passwords, addresses, financials) from chats into reports. Anonymize if needed.
- Minimum sample: Require at least 20 messages from the target speaker. If fewer, warn that analysis may be unreliable.