| name | attachment-analyzer |
| description | Analyze chat attachments (images, PDF, Word, PowerPoint, Excel/CSV/TXT) using MCP document/image tools and built-in file reading, then return concise findings plus recommended next steps. |
| category | analysis |
| lang | zh,en |
| requires_cli | uvx |
| metadata | {"lunaeclaw":{"emoji":"📎"}} |
Attachment Analyzer
Use this skill when the user sends or references attachments and asks to:
- summarize a file
- extract key points
- read PDF / Word / PPT / spreadsheet
- analyze an image / screenshot / chart
- compare multiple attachments
Goal
Produce a useful result quickly with a stable structure:
- What the file is
- Key findings
- Risks / missing data (if any)
- Suggested next actions
Recommended Tools
Prefer MCP document/image tools when available (for example):
read_document (PDF/DOC/DOCX/PPT/PPTX/XLS/XLSX)
read_image (image analysis / OCR-friendly image loading)
read_file (TXT/MD/LOG/JSON/YAML/CSV/TSV and other plain text files)
If aliases exist, use the alias names configured by the project (for example doc_read, image_read).
Workflow
- Identify attachment paths from the user message/context.
- Group by file type:
- image (
png/jpg/jpeg/webp/gif)
- binary document (
pdf/doc/docx/ppt/pptx/xls/xlsx)
- plain text (
txt/md/log/json/yaml/yml/csv/tsv)
- For each file:
- images: use image tool first
- binary documents: use document tool first
- plain text files: use
read_file first (do not force doc_read)
- if tool unavailable, fall back to built-in capabilities (e.g. image vision) or explain limitation clearly
- Summarize in Chinese by default unless the user asks another language.
- End with a practical next-step suggestion.
Output Template (Chinese, concise)
## 文件识别
- 文件A: 类型 / 主题(推测)
## 关键信息
- ...
## 风险或不确定点
- ...
## 建议下一步
1. ...
2. ...
Guardrails
- Do not fabricate unreadable content.
- If parsing fails, report the exact file and suggest a retry path (another tool / file format conversion).
- For spreadsheets, summarize sheet names / columns / notable values before deep analysis.