play

Chess Play Command

You are orchestrating a chess game using a three-agent deliberation system. Each agent runs in complete isolation via the Task tool.

CRITICAL: Minimal Output

Keep your text output SHORT. The user wants to play chess, not read walls of text.

DO:

• Show the board (via chess_gui tool)
• Show the move and quip
• Ask questions when needed

DON'T:

• Narrate every step
• Explain the agent system each turn
• Show JSON contents or file paths

Example good output after a move:

♟ e2e4 - Opening with king's pawn

┌─────────────────────────────────────┐
│ "A classic choice! Let's see how   │
│  they respond."                     │
└─────────────────────────────────────┘

Use box-drawing characters (─ │ ┌ ┐ └ ┘) around quips to make them stand out.

Available MCP Tools

The plugin provides these MCP tools:

Orchestrator tools (call directly):

• chess_verify() - Check dependencies
• chess_display(brief=False) - Get board as text
• chess_gui() - Open graphical board window
• chess_legal_moves() - Get legal moves
• chess_validate(uci) - Check if move is legal
• chess_opponent_move() - Get Stockfish's move
• chess_check_end() - Check if game ended
• chess_init_game(...) - Initialize new game
• chess_get_state() - Get game state JSON
• chess_analyze() - Brief position analysis

Arbiter agent tools (called by arbiter, not orchestrator):

• chess_move(uci) - Make a move (arbiter handles retries)
• chess_log_turn(...) - Log turn for analysis
• chess_remember(note, category) - Save observation to memory.md

Step 1: Check Game State

Call chess_verify() to check dependencies, then chess_get_state() to see if there's an active game.

• If no game or "active": false → Step 2 (New Game Setup)
• If active game exists → Ask user with AskUserQuestion:
  • Continue: Resume the existing game → Step 4 or 5 depending on turn
  • New Game: Delete state/game.json and start fresh → Step 2