playbooks

home / skills / mixedbread-ai / bernd / user-files

user-files skill

/backend/skills/user-files

This skill helps you search, read, and summarize your uploaded files from /files/ across PDFs, markdown, images, and documents.

npx playbooks add skill mixedbread-ai/bernd --skill user-files

Review the files below or copy the command above to add this skill to your agents.

Files (1)
skill.md
1.5 KB
---
name: user-files
description: Search and read user's uploaded files (PDFs, markdown, images, documents)
when_to_use: When the user asks about their files, documents, or wants to find/summarize content from uploaded materials
allowed_tools:
  - files
---

# User Files Access

You have access to the user's uploaded files stored in `/files/`. These can include:
- PDF documents
- Markdown files (.md)
- Text files (.txt)
- Word documents (.docx)
- Images (PNG, JPG, GIF, WebP)

## How to access files

Use the `files` tool with these commands:

### Search files semantically
```
files(command="search", query="your search query", path="/files/")
```
This performs semantic search across all uploaded files. Use natural language queries.

### List files in a folder
```
files(command="list", path="/files/")
files(command="list", path="/files/subfolder/")
```

### Read a specific file
```
files(command="read", path="/files/document.md")
```

## Tips

1. **Start with search** - If the user asks about a topic, search first to find relevant files
2. **Check file types** - The search results include metadata with mime_type
3. **For images** - You can describe what's in an image path but cannot view images directly
4. **Folder structure** - Users can organize files in folders, use list to explore

## Example queries

- "What did that PDF say about quarterly results?" → Search for "quarterly results" in /files/
- "Summarize my notes on machine learning" → Search for "machine learning" then read matching files
- "What files do I have?" → List /files/

Overview

This skill lets me search and read your uploaded files stored under /files/. I can semantically search PDFs, markdown, text, Word docs, and images, then open and summarize or extract content. Use it to quickly locate information across your documents without manual browsing.

How this skill works

I use the files tool to perform semantic searches across the /files/ folder, list folder contents, and open specific files for reading. Search returns results with metadata (including mime_type) so I can choose the best files to read. For images I provide descriptions based on available metadata and file context rather than pixel-level inspection.

When to use it

  • You need a quick summary of a long PDF or Word document.
  • You want to find notes or references across many files by topic.
  • You need to list or inspect files in a specific folder.
  • You want extraction of specific facts, dates, or figures from documents.
  • You need descriptions or context for uploaded images.

Best practices

  • Start by running a semantic search for the topic to narrow results before reading full files.
  • Check the mime_type in results to prioritize readable formats (text, markdown, PDF).
  • If multiple files match, ask for summaries or comparisons to save time.
  • Provide folder paths when you want me to restrict the scope (e.g., a subfolder).
  • Tell me the specific questions or the exact data you want extracted to get focused answers.

Example use cases

  • "Summarize any slides or PDFs mentioning quarterly revenue." — search for “quarterly revenue” and read matching files.
  • "List all files in /files/projectX/ so I can pick which to review." — list folder contents.
  • "Extract action items from meeting notes in my uploaded markdown files." — search and read matching notes.
  • "Compare two proposal documents and highlight key differences." — read both files and summarize contrasts.
  • "Describe what’s in the uploaded images folder and identify likely screenshots vs photos." — list and describe image files by metadata and context.

FAQ

Can you read images like text files?

I can describe images using available metadata and context, but I do not perform pixel-level OCR unless the image file yields embedded text; if you need OCR, request that explicitly and I will attempt to extract readable text.

How do I narrow searches to a specific folder?

Include the folder path when requesting searches or file lists (for example, specify /files/subfolder/) so I only inspect that location.