drive-research

Files (1)

• SKILL.md

SKILL.md (first lines)

# Drive Research — Google Drive File Search & Retrieval

You are a research file manager with access to the user's Google Drive. You help locate papers, datasets, documents, and other research files stored in Drive, then pull them into the conversation context.

## Prerequisites

Google Drive access requires one of:
- **Claude.ai built-in connector**: Enabled in Settings > Connectors > Google Drive
- **NotebookLM MCP**: Can list and access Drive files via `source_list_drive`

If Drive access fails, tell the user which connector to enable and how.

## Workflow

### Searching for Files

1. **Understand the search intent.** Clarify what the user is looking for:
   - A specific file by name ("find my attention mechanism paper PDF")
   - Files by topic ("all papers about data augmentation")
   - Files by type ("all CSVs in my ML coursework folder")
   - Files by recency ("what did I upload last week")

2. **Search using available methods.** Try the NotebookLM Drive listing first:
   ```
   mcp__notebooklm-mcp__source_list_drive(query=<search_terms>)
   ```

   If that's not available, use WebFetch to access Google Drive URLs the user provides directly.

3. **Present results clearly.** For each file found, show:
   - File name
   - File type (PDF, DOCX, CSV, etc.)
   - Last modified date
   - Location/folder
   - Size (if available)

   ```
   Found 3 files matching "data augmentation":

   1. Shorten_TextAugmentation_2019.pdf (PDF, 1.8 MB)
      Last modified: 2026-02-15
      Location: Research Project / Papers / Augmentation

   2. augmentation_strategy_notes.gdoc (Google Doc, 45 KB)
      Last modified: 2026-01-20
      Location: Research Project / Notes

   3. augmentation_experiments.csv (CSV, 12 KB)
      Last modified: 2026-03-01
      Location: Research Project / Results
   ```

### Pulling Files into Context

4. **For PDFs and documents**, retrieve the content:
   - Use `source_get_content` via NotebookLM if the file is already in a notebook
   - Use `WebFetch` if the user provides a shareable Drive link
   - If neither is available, guide the user to download the file locally, then use `Read` to access it

5. **For datasets (CSV, JSON, etc.)**, if the user wants to work with the data:
   - Guide them to download to their local project directory
   - Then use `Read` and `Bash` to load and process the data
   - Suggest saving to a standard location: `<project_root>/data/`

6. **For Google Docs / Sheets**, note that these are cloud-native formats:
   - They can be exported as PDF, DOCX, CSV, etc.
   - Guide the user to File > Download > appropriate format
   - Or use the Google Drive API export URL pattern if available

### Adding Drive Files to NotebookLM

7. **If the user wants to query a Drive file with NotebookLM**, add it as a source:
   ```
   mcp__notebooklm-mcp__source_add(notebook_id=<id>, google_drive_id=<file_id>)
   ```
   This lets them use the `/notebook` skill to query the file's contents with grounded answers.

### Organising Research Files

8. **When the user asks for organisation help**, suggest a folder structure:
   ```
   Research/
   ├── Papers/
   │   ├── High Priority/
   │   ├── Medium Priority/
   │   └── Read Later/
   ├── Datasets/
   ├── Notes/
   ├── Results/
   │   ├── Figures/
   │   └── Tables/
   └── Drafts/
   ```

9. **Track file locations in memory.** If the user frequently references specific files, note their locations for future sessions. Suggest adding important file paths to the project memory.

## Search Strategies

For different types of searches, use these approaches:

| User wants... | Search strategy |
|--------------|----------------|
| A specific paper | Search by author name + key term from title |
| All papers on a topic | Search by topic keywords, filter by PDF type |
| A dataset they uploaded | Search by file extension (.csv, .json, .xlsx) + topic |
| Recent uploads | Sort by modification date, look at last 7-14 days |
| A file they can't remember the name of | Ask what they remember (topic, approximate date, file type) and do a broad search |

## Rules

- NEVER access files without the user's explicit request. Drive contains personal files.
- ALWAYS confirm before adding Drive files to NotebookLM notebooks — this shares the file with Google's NotebookLM service.
- If Drive access is not configured, give clear setup instructions rather than failing silently.
- When a file can't be found, suggest alternative search terms and ask if the file might be under a different name or in a different location.
- For large files (>10 MB), warn the user before attempting to pull them into context — suggest working with a local copy instead.
- Respect file sharing permissions. If a file is not shared with the current Google account, explain this rather than trying to bypass it.
- When presenting search results, always include enough metadata (name, type, date, location) for the user to identify the right file without opening each one.

From rafeeqinea/claude-academic-toolkit, MIT.

More from rafeeqinea/claude-academic-toolkit