Skip to content

capability-matrix.md

Latest commit

 

History

History
284 lines (215 loc) · 7.5 KB

capability-matrix.md

File metadata and controls

284 lines (215 loc) · 7.5 KB
Raw

Platform Capability Matrix

This document outlines what each AI platform can and cannot do, helping you choose the right platform and prompts for your task.

Quick Reference

Capability ChatGPT Gemini Claude Claude Code Cursor Cline Copilot
Read files ❌ ❌ ❌ ✅ ✅ ✅ ✅*
Write files ❌ ❌ ❌ ✅ ✅ ✅ ✅*
Run commands ❌ ❌ ❌ ✅ ✅ ✅ ❌
Web access ✅** ✅ ❌ ✅ ❌ ❌ ❌
Multi-turn memory ✅ ✅ ✅ ✅ ✅ ✅ Limited
Custom instructions ✅ ✅ ✅ ✅ ✅ ✅ ✅
File upload ✅ ✅ ✅ N/A N/A N/A N/A
Context window 128K 2M 200K 200K 200K Varies 128K
Code execution ✅*** ✅*** ❌ ✅ ✅ ✅ ❌

*Copilot: Within current file context **ChatGPT: With browsing enabled ***Sandboxed code interpreter

Platform Details

ChatGPT / GPT-4

Best For: Brainstorming, planning, learning, general questions

Capabilities:

  • Large context window (128K tokens)
  • Web browsing (when enabled)
  • Code interpreter for Python
  • File upload and analysis
  • Custom GPTs with persistent instructions
  • DALL-E image generation

Limitations:

  • No direct file system access
  • Cannot run arbitrary shell commands
  • No IDE integration
  • Cannot access local projects

Prompt Recommendations:

  • Use portable prompts from prompts/core/
  • Provide all necessary context in the prompt
  • Copy-paste code snippets for review
  • Use for planning before implementation

Configuration: See platforms/chatgpt/

Google Gemini

Best For: Large context analysis, research, documentation review

Capabilities:

  • Massive context window (1M+ tokens)
  • Web access and search
  • File upload (documents, images, video)
  • Code execution (sandboxed)
  • Multi-modal understanding

Limitations:

  • No direct file system access
  • Cannot run shell commands
  • No IDE integration
  • May require different prompting style

Prompt Recommendations:

  • Place instructions AFTER context (Gemini-specific)
  • Use XML-style tags for structure
  • Leverage large context for full file analysis
  • Use for documentation and code review

Configuration: See platforms/gemini/

Claude (Web/API)

Best For: Detailed analysis, nuanced responses, long-form content

Capabilities:

  • Large context window (200K tokens)
  • Excellent instruction following
  • Strong code understanding
  • File upload and analysis
  • Projects with persistent context

Limitations:

  • No web access
  • No file system access
  • Cannot run commands
  • No IDE integration

Prompt Recommendations:

  • Use extended prompts for complex tasks
  • Leverage artifacts for code output
  • Use Projects for persistent context
  • Excellent for security reviews

Configuration: See platforms/claude-code/ (shared patterns)

Claude Code

Best For: Full development workflow, file management, testing

Capabilities:

  • Direct file system access (read/write)
  • Shell command execution
  • Git operations
  • Multi-file editing
  • Tool use and MCP servers
  • Web search and fetch
  • Background task execution

Limitations:

  • Requires terminal environment
  • Permission-based (asks before actions)
  • No GUI interaction

Prompt Recommendations:

  • Use extended prompts for complex workflows
  • Reference skills directly by path
  • Use agents for multi-step tasks
  • Primary platform for this framework

Configuration: See platforms/claude-code/

Cursor

Best For: Code editing, refactoring, in-editor assistance

Capabilities:

  • Full codebase awareness
  • File reading and editing
  • Terminal command execution
  • .cursorrules for persistent instructions
  • Multi-file context
  • Composer for multi-step tasks

Limitations:

  • IDE-specific (VS Code fork)
  • May not follow complex workflows
  • Context can be noisy with large codebases

Prompt Recommendations:

  • Use .cursorrules for project standards
  • Use .cursor/rules/ for domain-specific rules
  • Keep prompts focused on specific tasks
  • Reference file paths explicitly

Configuration: See platforms/cursor/

Cline

Best For: Autonomous coding tasks, guided development

Capabilities:

  • File system access (read/write)
  • Command execution
  • Plan/Act mode separation
  • Multiple AI provider support
  • .clinerules for project guidelines
  • Permission-based safety

Limitations:

  • VS Code extension
  • Requires explicit permission grants
  • May be verbose in explanations

Prompt Recommendations:

  • Use Plan mode for architecture decisions
  • Use Act mode for implementation
  • Configure .clinerules for project standards
  • Set clear scope boundaries

Configuration: See platforms/cline/

GitHub Copilot

Best For: Inline code completion, quick suggestions

Capabilities:

  • Inline code completion
  • Comment-driven generation
  • Chat interface (@workspace)
  • Multi-file awareness (limited)
  • Custom instructions file

Limitations:

  • Limited context window (~8K tokens)
  • Best for small, focused tasks
  • Less reliable for complex logic
  • Cannot execute commands

Prompt Recommendations:

  • Use .github/copilot-instructions.md for standards
  • Write detailed comments before code
  • Keep requests small and focused
  • Use for boilerplate and patterns

Configuration: See platforms/copilot/

Choosing the Right Platform

By Task Type

Task Recommended Platform Why
Project planning ChatGPT, Claude Good for brainstorming
Architecture design Claude, Gemini Long context, nuanced
Setting up testing Claude Code, Cursor Need file access
Writing tests Claude Code, Cline Need to read existing code
Code review Gemini, Claude Large context windows
Refactoring Cursor, Claude Code Direct file editing
Documentation Claude, ChatGPT Good prose generation
Security audit Claude, Claude Code Careful analysis
Quick edits Cursor, Copilot Fast, inline

By Project Stage

Stage Platforms Notes
Brainstorming ChatGPT, Claude, Gemini No file access needed
Planning Claude, Gemini Long-form analysis
Setup Claude Code, Cursor, Cline Need file creation
Development Claude Code, Cursor, Cline File editing + commands
Testing Claude Code, Cline Run test commands
Documentation Claude, Claude Code Write + create files
Launch Any Mostly planning
Maintenance Claude Code, Cursor Ongoing edits

Platform-Specific Tips

ChatGPT

  • Enable web browsing for documentation lookups
  • Use Custom GPTs for persistent WordPress context
  • Upload the skills files as knowledge base

Gemini

  • Put your question AFTER the context/code
  • Use explicit structural tags
  • Leverage the massive context window

Claude

  • Use Projects for persistent context
  • Upload frequently referenced files
  • Excellent for security-sensitive reviews

Claude Code

  • Set up CLAUDE.md in your project root
  • Configure relevant MCP servers
  • Use skills/ directory for reference

Cursor

  • Create comprehensive .cursorrules
  • Use @workspace for codebase questions
  • Use Composer for multi-file changes

Cline

  • Start in Plan mode for new features
  • Use .clinerules for consistent behavior
  • Grant permissions thoughtfully

Copilot

  • Write descriptive function comments
  • Use copilot-instructions.md for standards
  • Best for completion, not generation