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>
This commit is contained in:
Florian BRUNIAUX 2026-01-20 11:23:43 +01:00
parent 360e5203f6
commit c2de35caba
27 changed files with 7304 additions and 12 deletions

View file

@ -0,0 +1,220 @@
# cowork-reference.yaml
# Quick lookup for Claude Cowork - ~1.5K tokens target
# Source: cowork/ documentation
# Purpose: Condensed index for LLMs to quickly answer user questions about Cowork
version: "1.0.0"
updated: "2026-01-20"
status: "research-preview"
# ════════════════════════════════════════════════════════════════
# CORE DOCUMENTATION - File paths for detailed info
# ════════════════════════════════════════════════════════════════
docs:
hub: "cowork/README.md"
overview: "cowork/guide/00-overview.md"
getting_started: "cowork/guide/01-getting-started.md"
capabilities: "cowork/guide/02-capabilities.md"
security: "cowork/guide/03-security.md"
troubleshooting: "cowork/guide/04-troubleshooting.md"
cheatsheet: "cowork/reference/cheatsheet.md"
faq: "cowork/reference/faq.md"
comparison: "cowork/reference/comparison.md"
glossary: "cowork/reference/glossary.md"
# ════════════════════════════════════════════════════════════════
# PROMPTS LIBRARY (60+ ready-to-use)
# ════════════════════════════════════════════════════════════════
prompts:
file_ops: "cowork/prompts/file-ops.md" # 20 prompts
document_creation: "cowork/prompts/document-creation.md" # 15 prompts
data_extraction: "cowork/prompts/data-extraction.md" # 15 prompts
research: "cowork/prompts/research.md" # 17 prompts
# ════════════════════════════════════════════════════════════════
# WORKFLOWS (step-by-step guides)
# ════════════════════════════════════════════════════════════════
workflows:
file_organization: "cowork/workflows/file-organization.md"
expense_tracking: "cowork/workflows/expense-tracking.md"
report_synthesis: "cowork/workflows/report-synthesis.md"
meeting_prep: "cowork/workflows/meeting-prep.md"
team_handoff: "cowork/workflows/team-handoff.md"
# ════════════════════════════════════════════════════════════════
# QUICK REFERENCE - Essential facts
# ════════════════════════════════════════════════════════════════
essentials:
what: "Agentic desktop feature for knowledge workers - file manipulation without code"
platform: "macOS only (Windows planned, Linux not announced)"
subscription: "Pro ($20/mo) or Max ($100-200/mo) - both have access"
usage_limits: "Resets every 5 hours; Pro ~1-1.5h intensive use, Max 5x-20x more"
context: "~200K tokens theoretical, ~165K effective (system overhead)"
vpn_compatible: "NO - #1 issue, VM routing conflict"
# ════════════════════════════════════════════════════════════════
# CAN DO / CANNOT DO
# ════════════════════════════════════════════════════════════════
capabilities:
can_do:
- "Read/write local files in sandbox"
- "Create Office docs (Word, Excel, PowerPoint)"
- "Extract data from images (OCR)"
- "Organize folders/files"
- "Web research via Chrome"
- "Generate PDFs"
cannot_do:
- "Execute code or scripts"
- "Make API calls"
- "Access cloud storage directly"
- "Process audio/video"
- "Work with VPN active"
- "Run on Windows/Linux"
- "Operate in background"
# ════════════════════════════════════════════════════════════════
# PROMPTING FORMULA (CTOC)
# ════════════════════════════════════════════════════════════════
prompting:
framework: "CTOC - Context → Task → Output → Constraints"
context: "Background, files, situation"
task: "Clear single objective"
output: "Exact format and location"
constraints: "Rules, limits, preferences"
example: |
CONTEXT: I have 50 receipts from Paris business trip in ~/Cowork-Workspace/input/receipts/
TASK: Extract all expenses into a spreadsheet
OUTPUT: Save as ~/Cowork-Workspace/output/paris-expenses.xlsx
CONSTRAINTS: Use EUR currency, semicolon formulas, categorize by type
# ════════════════════════════════════════════════════════════════
# COMMON ERRORS - Exact messages + solutions
# ════════════════════════════════════════════════════════════════
errors:
vm_timeout:
message: "Failed to start Claude's workspace — VM connection timeout after 60 seconds"
cause: "VPN active"
solution: "Disconnect VPN completely before using Cowork"
chrome_host:
message: "Chrome native messaging host not found"
cause: "Extension mismatch or missing host"
solution: "Manual installation via bash (see troubleshooting doc)"
context_limit:
message: "Context limit reached"
cause: "System overhead (~25-35K tokens)"
actual_limit: "~165K usable (not 200K)"
solution: "Break into batches of 10-20 files"
access_denied:
message: "Access denied — path outside allowed directories"
cause: "Folder not granted"
solution: "Re-grant folder access in System Preferences"
session_terminated:
message: "Session terminated unexpectedly"
cause: "Sleep/background/VPN"
solution: "Keep app foreground, disable sleep, disconnect VPN"
# ════════════════════════════════════════════════════════════════
# TOKEN BUDGET - Task costs
# ════════════════════════════════════════════════════════════════
tokens:
simple_qa: "5K-10K"
file_inventory: "20K-30K"
small_file_org: "30K-50K (10-20 files)"
large_file_org: "80K-150K (50+ files)"
multi_doc_synthesis: "50K-100K"
ocr_batch: "60K-100K (10+ images)"
agentic_overhead: "15-30% (Plan→Execute→Check cycles)"
# ════════════════════════════════════════════════════════════════
# OCR ACCURACY (from Perplexity research)
# ════════════════════════════════════════════════════════════════
ocr:
field_extraction: "~97% (vendor, date, totals)"
line_items: "~63% (table rows)"
warning: "Budget 30-50% of 'saved' time for error correction"
best_for: "Single-currency, clear fonts, standard formats"
# ════════════════════════════════════════════════════════════════
# COMPETITIVE POSITIONING
# ════════════════════════════════════════════════════════════════
competition:
unique_strength: "Only assistant with true multi-step autonomy + local files + browser automation"
vs_copilot: "Better autonomy, worse OCR accuracy"
vs_gemini: "Better autonomy, smaller context (165K vs 1M+)"
vs_chatgpt: "Better autonomy, macOS only vs cross-platform"
vs_apple_intelligence: "Better capabilities, cloud vs on-device"
best_ocr: "Microsoft Copilot"
best_context: "Gemini Advanced (1M+ tokens)"
best_privacy: "Apple Intelligence (on-device)"
best_autonomy: "Cowork"
# ════════════════════════════════════════════════════════════════
# ENTERPRISE VALIDATION (Claude general, not Cowork-specific)
# ════════════════════════════════════════════════════════════════
enterprise:
telus: "$90M value, 500K hours saved"
rakuten: "87.5% reduction in processing time"
zapier: "89% employee adoption"
note: "Stats for Claude in general - validates underlying AI capabilities"
# ════════════════════════════════════════════════════════════════
# LIMITATIONS & CONSTRAINTS
# ════════════════════════════════════════════════════════════════
limitations:
environment:
- "VPN incompatible (VM routing conflict)"
- "macOS only (Windows planned, Linux not announced)"
- "Requires app foreground (no background operation)"
- "No session persistence across restarts"
technical:
- "No code execution"
- "No API calls"
- "No cloud storage direct access"
- "No audio/video processing"
- "Context limit hits at 165K (not 200K)"
workflow:
- "Pro tier exhausts in ~1-1.5h intensive use"
- "Usage resets every 5 hours (not daily)"
- "No cross-session memory"
- "Connector reliability inconsistent"
# ════════════════════════════════════════════════════════════════
# DECISION FLOWCHART
# ════════════════════════════════════════════════════════════════
decide:
use_cowork_when:
- "Need to organize/rename files"
- "Create Office documents"
- "Extract data from images/receipts"
- "Compile research → documents"
- "Batch file operations"
- "Browser automation tasks"
use_alternative_when:
- "Need code execution → Claude Code"
- "Need best OCR → Microsoft Copilot"
- "Need 1M+ context → Gemini Advanced"
- "Privacy critical → Apple Intelligence"
- "On Windows/Linux → Copilot or ChatGPT"
- "VPN required → Claude Code"
# ════════════════════════════════════════════════════════════════
# QUICK START (5 steps)
# ════════════════════════════════════════════════════════════════
quick_start:
1: "Enable Cowork in Claude Desktop → Settings → Features"
2: "Create ~/Cowork-Workspace/{input,output}"
3: "Grant folder access when prompted"
4: "Use CTOC framework for prompts"
5: "Review plan before approval"
# ════════════════════════════════════════════════════════════════
# GOLDEN RULES
# ════════════════════════════════════════════════════════════════
rules:
1: "Always review plan before approval"
2: "Dedicated workspace only (never Documents/Desktop)"
3: "No credentials in workspace"
4: "Verify file sources (prompt injection risk)"
5: "Budget 30-50% time for OCR error correction"
6: "Disconnect VPN before using"
7: "Backup before destructive operations"

View file

@ -3,8 +3,8 @@
# Source: guide/ultimate-guide.md
# Purpose: Condensed index for LLMs to quickly answer user questions about Claude Code
version: "3.9.6"
updated: "2026-01"
version: "3.9.7"
updated: "2026-01-20"
# ════════════════════════════════════════════════════════════════
# DEEP DIVE - Line numbers in guide/ultimate-guide.md
@ -119,7 +119,8 @@ deep_dive:
ai_ecosystem_voice_to_text: "guide/ai-ecosystem.md:449"
ai_ecosystem_alternative_providers: "guide/ai-ecosystem.md:959"
voice_refine_skill: "examples/skills/voice-refine/SKILL.md"
# Cowork documentation (expanded - see cowork/)
# Cowork documentation (v1.0 - complete)
cowork_reference: "machine-readable/cowork-reference.yaml" # Dedicated YAML index
cowork_hub: "cowork/README.md"
cowork_summary: "guide/cowork.md"
cowork_overview: "cowork/guide/00-overview.md"