claude-code-ultimate-guide/cowork/guide/02-capabilities.md
Florian BRUNIAUX c2de35caba docs: complete Cowork documentation v1.0 (Phase 2)
- Create comprehensive Cowork docs (23 files in cowork/)
  - 6 guides: overview, getting started, capabilities, security, troubleshooting
  - 60+ ready-to-use prompts across 4 categories
  - 5 step-by-step workflows
  - Reference materials: cheatsheet, FAQ, comparison, glossary

- Integrate Perplexity research (P0/P1/P2)
  - Exact error messages with solutions (VPN, Chrome host, context limits)
  - Competitive analysis (vs Copilot/Gemini/ChatGPT/Apple Intelligence)
  - Enterprise validation (TELUS, Rakuten, Zapier stats)
  - OCR accuracy benchmarks (97% field, 63% line-item)
  - Token budget planning per task type

- Document critical limitations
  - VPN incompatibility (#1 community issue)
  - Context limit reality (165K vs 200K theoretical)
  - Platform constraints (macOS only)
  - Usage limits and pricing (Pro $20, Max $100-200)

- Update central files
  - README.md: detailed Cowork section with tables
  - VERSION: 3.9.6 → 3.9.7
  - machine-readable/reference.yaml: add cowork_reference entry
  - machine-readable/cowork-reference.yaml: new LLM-optimized index (~1.5K tokens)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2026-01-20 11:23:43 +01:00

358 lines
10 KiB
Markdown

# Cowork Capabilities
> **Reading time**: ~8 minutes
>
> **Purpose**: Understand exactly what Cowork can and cannot do
---
## Capability Matrix
### File Operations
| Operation | Can Do | Notes |
|-----------|--------|-------|
| **Read files** | ✅ Yes | Any format in granted folder |
| **Create files** | ✅ Yes | Any text-based format |
| **Move files** | ✅ Yes | Within granted folders |
| **Copy files** | ✅ Yes | Within granted folders |
| **Rename files** | ✅ Yes | Bulk renaming supported |
| **Delete files** | ✅ Yes | ⚠️ Permanent, no trash |
| **Create folders** | ✅ Yes | Nested structures supported |
| **Extract archives** | ❌ No | Cannot run unzip/tar |
| **Compress files** | ❌ No | Cannot create archives |
### Document Generation
| Format | Can Create | Features |
|--------|------------|----------|
| **Plain text** (.txt) | ✅ Yes | Any content |
| **Markdown** (.md) | ✅ Yes | Full formatting |
| **Word** (.docx) | ✅ Yes | Headers, tables, formatting |
| **Excel** (.xlsx) | ✅ Yes | Formulas, multiple sheets, formatting |
| **PowerPoint** (.pptx) | ✅ Yes | Slides, basic formatting |
| **PDF** | ✅ Yes | Generated from content |
| **HTML** | ✅ Yes | Full HTML/CSS |
| **CSV** | ✅ Yes | Data export |
| **JSON** | ✅ Yes | Structured data |
### Input Processing
| Input Type | Can Process | Notes |
|------------|-------------|-------|
| **Text files** | ✅ Yes | Any encoding |
| **Office docs** | ✅ Yes | Word, Excel, PowerPoint |
| **PDFs** | ✅ Yes | Text extraction |
| **Images** | ✅ Yes | OCR for text extraction |
| **Screenshots** | ✅ Yes | Receipt/document scanning |
| **Markdown** | ✅ Yes | Full parsing |
| **CSV/JSON** | ✅ Yes | Structured data parsing |
| **Audio** | ❌ No | Cannot process |
| **Video** | ❌ No | Cannot process |
| **Encrypted files** | ❌ No | Cannot decrypt |
### Web Capabilities (via Chrome)
| Action | Can Do | Notes |
|--------|--------|-------|
| **Search web** | ✅ Yes | Via Chrome integration |
| **Read pages** | ✅ Yes | Extract content |
| **Save content** | ✅ Yes | To local files |
| **Fill forms** | ⚠️ Limited | Requires explicit approval |
| **Make purchases** | ❌ No | Security restriction |
| **Login to sites** | ❌ No | Security restriction |
| **API calls** | ❌ No | No direct network access |
---
## Detailed Capabilities
### 1. File Organization
Cowork excels at organizing large numbers of files:
```
INPUT: 500 files in Downloads folder
OUTPUT: Organized structure by type/date/project
```
**What Cowork Does**:
- Analyzes file names, types, content
- Proposes organizational schema
- Creates folder structure
- Moves files (with your approval)
- Generates organization report
**Limitations**:
- Cannot access file metadata (creation date, etc.) reliably
- Cannot read file content for some binary formats
- Relies on file names/extensions for categorization
### 2. Document Synthesis
Combine multiple sources into structured outputs:
```
INPUT: 15 meeting notes, 3 reports, 5 emails (as text files)
OUTPUT: Executive summary with key decisions and action items
```
**What Cowork Does**:
- Reads all source documents
- Identifies key themes and information
- Structures into requested format
- Generates polished output document
**Limitations**:
- Context window limits (~200K tokens)
- Cannot access original email/calendar systems
- Synthesis quality depends on source clarity
### 3. Data Extraction
Extract structured data from unstructured sources:
```
INPUT: 20 receipt images (photos, screenshots)
OUTPUT: Excel spreadsheet with Date, Vendor, Amount, Category
```
**What Cowork Does**:
- OCR on images
- Identifies relevant fields
- Normalizes data format
- Creates Excel with formulas
**Limitations**:
- **Field extraction**: ~97% accuracy (vendor, date, totals)
- **Line-item extraction**: ~63% accuracy (table rows) — verify manually
- Handwritten text is challenging
- Some receipt formats may not parse correctly
- Budget 30-50% of "saved" time for error correction
> ⚠️ **No independent benchmarks exist** for AI file organization tools. Productivity claims are vendor marketing, not peer-reviewed research. Set realistic expectations.
### 4. Report Generation
Create formatted reports from raw data:
```
INPUT: CSV with sales data
OUTPUT: Formatted report with charts and analysis
```
**What Cowork Does**:
- Analyzes data patterns
- Creates summary statistics
- Generates visual representations
- Formats for specified output
**Limitations**:
- Chart capabilities limited in Excel output
- Complex visualizations need manual refinement
- Statistical analysis is basic
### 5. Research Compilation
Gather and organize research from multiple sources:
```
INPUT: Topic: "Remote work productivity tools"
OUTPUT: Research document with comparison matrix
```
**What Cowork Does**:
- Uses Chrome for web research
- Extracts relevant information
- Organizes into structured format
- Cites sources
**Limitations**:
- Cannot access paywalled content
- Web research is slower than direct API
- Quality depends on available public information
---
## Excel Capabilities (Detailed)
Since Excel output is a major Cowork strength:
### Supported Features
| Feature | Supported | Example |
|---------|-----------|---------|
| **Basic formulas** | ✅ Yes | `=SUM(A1:A10)` |
| **Conditional formulas** | ✅ Yes | `=IF(A1>100, "High", "Low")` |
| **VLOOKUP/HLOOKUP** | ✅ Yes | Cross-reference sheets |
| **Multiple sheets** | ✅ Yes | Summary + detail sheets |
| **Cell formatting** | ✅ Yes | Bold, colors, borders |
| **Number formatting** | ✅ Yes | Currency, percentages |
| **Data validation** | ⚠️ Limited | Basic dropdown lists |
| **Pivot tables** | ❌ No | Must create manually |
| **Macros/VBA** | ❌ No | Security restriction |
| **Charts** | ⚠️ Limited | Basic chart types |
### Regional Considerations
Excel formula syntax varies by region:
- **US/UK**: `=SUM(A1,A2)` (comma separator)
- **EU**: `=SUM(A1;A2)` (semicolon separator)
**Tip**: Specify your regional setting in prompts:
```
Create an Excel file using European formula syntax (semicolon separators)
```
---
## Context Window Usage
Cowork shares Claude's ~200K token context limit, but **effective capacity is ~165K**.
### Why 165K, Not 200K?
System overhead consumes tokens before your task starts:
- Tool definitions: ~10K tokens
- Safety instructions: ~10K tokens
- Execution logs: ~5-15K tokens (varies)
**Effective limit**: ~165-175K usable tokens
### Practical Limits
| Content Type | Approximate Capacity |
|--------------|---------------------|
| Plain text pages | 100-400 pages |
| Documents | 40-80 typical docs |
| Spreadsheet rows | 8,000-40,000 rows |
| Images (OCR) | 40-80 images |
### When You Hit Limits
**Error message**:
```
Context limit reached
```
**Symptoms**:
- Cowork stops mid-task (often around 165K, not 200K)
- Results are incomplete
- Silent failure without clear message
**Solutions**:
- Break work into batches of 10-20 files
- Process files in groups
- Save intermediate results to checkpoint files
- Start fresh conversation for new tasks
### Token Budget by Task Type
| Task | Tokens | Pro Sessions |
|------|--------|--------------|
| Simple Q&A | 5K-10K | Many |
| File inventory | 20K-30K | 6-8 |
| Small file org (10-20 files) | 30K-50K | 3-5 |
| Large file org (50+ files) | 80K-150K | 1-2 |
| OCR batch (10+ images) | 60K-100K | 2-3 |
**Agentic overhead**: Plan→Execute→Check cycles add 15-30% tokens.
---
## What Cowork CANNOT Do
### Code Execution
```
❌ Cannot run: Python, JavaScript, shell scripts
❌ Cannot execute: Installed applications
❌ Cannot use: Command-line tools
```
**Workaround**: Use Claude Code for code execution tasks.
### Network Operations
```
❌ Cannot make: API calls, HTTP requests
❌ Cannot access: Remote databases
❌ Cannot sync: Cloud storage directly
```
**Workaround**: Download cloud files locally first, or use Chrome for web access.
### System Operations
```
❌ Cannot change: System settings
❌ Cannot install: Software
❌ Cannot access: Other applications' data
```
**Workaround**: These operations must be done manually.
### Security-Sensitive Operations
```
❌ Cannot handle: Passwords, credentials
❌ Cannot process: Encrypted files
❌ Cannot access: Protected system folders
```
**Workaround**: Keep sensitive data out of Cowork workspace.
### Environment Constraints
```
❌ Cannot work: With VPN active (VM routing conflict)
❌ Cannot run: On Windows or Linux (macOS only)
❌ Cannot operate: In background (requires app foreground)
❌ Cannot persist: Sessions across app restarts
```
**VPN Issue**: Cowork's VM conflicts with VPN network routing. This is the #1 reported issue. Solution: Disconnect VPN before using Cowork. See [Troubleshooting](04-troubleshooting.md#vm-connection-issues) for details.
---
## Capability Decision Tree
```
Need to do something with files?
├─ Yes → Is it within your granted folder?
│ ├─ Yes → Can Cowork read the file type?
│ │ ├─ Yes → Cowork can help ✅
│ │ └─ No (audio/video/encrypted) → Manual process needed
│ └─ No → Grant access or move files first
└─ No → What do you need?
├─ Execute code → Use Claude Code
├─ API integration → Manual or Claude Code
├─ System changes → Manual operation
└─ Web research → Cowork + Chrome ✅
```
---
## Best Practices for Capabilities
### Maximize Success
1. **Match task to capability** — Check matrix before starting
2. **Prepare inputs** — Ensure files are in readable formats
3. **Specify formats** — Be explicit about output requirements
4. **Test small first** — Verify with few files before batch
### When to Choose Alternatives
| If You Need | Use Instead |
|-------------|-------------|
| Code execution | Claude Code |
| API integration | Claude Code + scripts |
| Cloud file sync | Native cloud apps |
| Audio/video | Specialized tools |
| Real-time data | Manual process |
---
*[← Getting Started](01-getting-started.md) | [Cowork Documentation](../README.md) | [Security →](03-security.md)*