AgentMe
Learn AI coding tools through play
A strategy card game where you build a deck of AI coding tools and battle other developers. Every card teaches real best practices — flip it to learn.
MODEL
1M context -- draw 2 cards when played
The most capable model. Period.
Opus 4.6
Claude Opus 4.6 is Anthropic's most intelligent model, released February 5, 2026. It features a 1M token context window, 128k max output tokens, and adaptive thinking that dynamically determines reasoning depth. It outperforms GPT-5.2 on GDPval-AA by ~144 Elo points.
- Use adaptive thinking (the default) -- at high effort, Claude almost always thinks; at lower effort, it may skip for simpler problems
- Use effort levels (low, medium, high, max) to control intelligence/speed/cost tradeoffs -- max is exclusive to Opus 4.6
- Leverage the 1M context window for large codebases, but note premium pricing kicks in above 200k tokens
- Use context compaction for longer-running agentic tasks to manage the context window efficiently
AGENT
While you control a Workflow, this has +1 attack
Your AI pair programmer, right in the editor.
VS Code Copilot
The GitHub Copilot extension for VS Code provides AI-powered code completion, inline chat, and a full Agent Mode that can autonomously plan, edit files, run terminals, and execute multi-step tasks. Agent plugins and Copilot Memory were introduced in VS Code 1.109-1.110 (Feb-March 2026).
- Use Agent Mode (Ctrl+Alt+I) for multi-file tasks; use inline chat (Ctrl+I) for single-location edits
- Install agent plugins from Extensions view for domain-specific skills
- Configure MCP servers in .vscode/mcp.json to extend Copilot's tool capabilities
- Use /autoApprove in agent chat to auto-approve tool calls for trusted tasks
MODEL
Battlecry: play a Command from your hand for free
The model that unified Codex and GPT.
GPT-5.4
GPT-5.4 is OpenAI's frontier model, released March 5, 2026. It features a 1.05M token context window, native computer-use capabilities, and configurable reasoning effort levels (none/low/medium/high/xhigh). Pricing is $2.50/$15.00 per million tokens.
- Use reasoning.effort (low/medium/high/xhigh) for complex multi-step reasoning tasks
- Leverage Tool Search for systems with many tools -- GPT-5.4 can look up tool definitions on demand
- For inputs exceeding 272K tokens, be aware of the 2x input / 1.5x output pricing uplift
- Use structured outputs and function calling for reliable, parseable responses in production
Goal of the Game
Reduce your opponent's health to zero by playing agents, commands, workflows, and artifacts. Each card represents a real AI coding tool — and every match teaches you something new.
Agents
Your fighters. They have attack and health, and stay on the board until destroyed.
Commands
One-shot effects. Play them, trigger the effect, they're gone. Like running a CLI command.
Workflows
Persistent effects that stay on the board and boost your team every turn. Your CI/CD pipeline.
Artifacts
Passive tools that provide ongoing bonuses. Your config files and documentation.
How to Play
Coming SoonEach turn you get mana to spend on cards. Play agents to attack, commands for instant effects, and workflows to control the board. The developer who best combines their tools wins.
Pick your clan
Choose Claude Code, Copilot, or Codex as your primary agent. Add neutral cards from any faction.
Build a 30-card deck
Balance your mana curve. Mix agents for combat, commands for removal, and workflows for long-term value.
Battle & learn
Play against friends. Flip any card to read its best practices, commands, and docs link. Every match is a learning session.
Reading a Card
Every card has two sides. The front is for gameplay — the back is for learning. Right-click any card to flip it.
Front (Gameplay)
- Mana cost — how much it costs to play this turn
- Attack / Health — combat stats (agents only)
- Type — Agent, Command, Workflow, or Artifact
- Phase tags — where in the dev lifecycle this tool shines
- Keywords — Battlecry, Rush, Taunt, Deathrattle
- Flavor text — a memorable one-liner
Back (Education)
- Description — what this tool actually does
- Best Practices — 4 verified tips from real usage
- Commands — copy-paste ready CLI commands
- QR code — scan to open the official documentation
- Sources — verified links, no hallucinated content
Controls
Mouse
- Left click — open card fullscreen
- Right click — flip card (front / back)
Keyboard
- ⏎ Enter / Space — open fullscreen or flip
- Arrow Left / Right — navigate cards in presentation mode
- Escape — exit fullscreen or presentation
Choose Your Clan
Claude Code
26 cards. Terminal-native agent with sub-agents, hooks, MCP, and the Superpowers skill suite.
Microsoft
10 cards. GitHub Actions, Copilot, Azure CLI, Codespaces — the GitHub ecosystem.
Codex
14 cards. OpenAI's Codex CLI, GPT-5.4 models, Agents SDK, sandboxing, and function calling.
Neutral
23 cards. Aider, LangGraph, CrewAI, Gemini CLI, Docker Compose — works in any deck.
Game Modes
Battle Mode
Coming SoonTwo players, 30-card decks, mana that grows each turn. Play agents, cast commands, deploy workflows. The developer who destroys their opponent's health pool wins. Every card you play teaches a real tool — even when you lose, you learn.
Presentation Mode
Available NowBuild a deck of tools you want to present, then go fullscreen. Navigate card by card with keyboard arrows. Perfect for tech talks, workshops, and team onboarding. Each card flip reveals the education side.
Ready to build your deck?
73 cards across 4 clans. Verified best practices from 2026. Free and open-source.