---

name: connect-four description: This skill should be used when the user says "来一盘四子棋", "我要玩四子棋", "玩四子棋", "四连棋", "コネクトフォーで遊ぼう", "四目並べ", "play Connect Four", "let's play Connect Four", "start a Connect Four game", or any request in any language to play a game of Connect Four / 四子棋 / 四目並べ. version: 0.1.0 allowed-tools: Bash, TeamCreate, TaskCreate, TaskUpdate, Agent, SendMessage, TeamDelete

Connect Four Skill

Two AI agents — Blue and Red — play Connect Four on a 6×7 board. You (team-lead) orchestrate the game, display the board after every move, and referee the result.

---

Step 1 — Prerequisites Check

Run this Bash command:

printenv CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS

If the output is NOT exactly 1, output the following message and stop immediately (do not proceed to any further steps):

⚠️  Connect Four requires the experimental Agent Teams feature.

To enable it, add the following to the "env" section of ~/.claude/settings.json:

  "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"

Then restart Claude Code and try again.

---

Step 2 — Create the Team

Call TeamCreate with:

• team_name: "connect-four"
• description: "Connect Four — Blue vs Red"

---

Step 3 — Spawn Both Player Agents (in parallel)

In a single message, make two Agent tool calls simultaneously:

Call 1 — blue-player:

• subagent_type: "general-purpose"
• team_name: "connect-four"
• name: "blue-player"
• prompt: (see Appendix A below)

Call 2 — red-player:

• subagent_type: "general-purpose"
• team_name: "connect-four"
• name: "red-player"
• prompt: (see Appendix B below)

---

Step 4 — Initialize the Board

The board is a 42-character flat string (6 rows × 7 columns, row-major, top to bottom):

• . = empty cell
• B = blue piece
• R = red piece
• Cell index formula: index = row * 7 + col (row 0 = top row, row 5 = bottom row; col 0..6 = left..right)
• Column numbers shown to users: 1–7 (col_index = column_number − 1)

Initial board: ".........................................." (exactly 42 dots)

Render and display the initial empty board to the user (see Board Rendering below), then proceed to the game loop.

---

Step 5 — Game Loop

Track these variables:

• board — the current 42-char board string
• move_count — starts at 0, increments each move
• current_player — starts as "blue-player"
• current_piece — starts as "B" (blue goes first)
• last_move_description — starts as "Game start"

Repeat the following until game_over:

5a — Send board to current player

Call SendMessage with:

type: "message"
recipient: <current_player>
content:
  BOARD: <42-char board string>
  TURN: <move_count>
  YOUR_PIECE: <current_piece>
  OPPONENT_PIECE: <R if current_piece is B, else B>
  LAST_MOVE: <last_move_description>
summary: "Your turn — choose column 1-7"

5b — Receive and parse the response

The player's reply will arrive automatically. Extract from the content:

• MOVE: N — the column chosen (1–7)
• BOARD: <42-char string> — the updated board after placing their piece

5c — Validate the move

Check:

1. The new board string is exactly 42 characters.
2. The new board has exactly one more occurrence of current_piece compared to the old board.
3. All other characters are unchanged.

If validation fails, re-send the original board with an error note:

ERROR: Your board update was invalid. Please re-read the rules and try again.
BOARD: <original 42-char board>
TURN: <move_count>
YOUR_PIECE: <current_piece>
OPPONENT_PIECE: <opponent_piece>
LAST_MOVE: <last_move_description>

5d — Update state and render

Set board to the new board string. Set last_move_description to "Blue played column N" or "Red played column N". Increment move_count. Render the board for the user (see Board Rendering below), including the move description.

5e — Check for win

After each move, check whether current_piece has four in a row anywhere on the board. Use the index formula board[row*7+col].

Horizontal (4 checks per row, 6 rows = 24 checks): For r = 0..5, c = 0..3: → win if board[r7+c], board[r7+c+1], board[r7+c+2], board[r7+c+3] are all current_piece

Vertical (3 starting rows per column, 7 columns = 21 checks): For r = 0..2, c = 0..6: → win if board[r*7+c], board[(r+1)*7+c], board[(r+2)*7+c], board[(r+3)*7+c] are all current_piece

Diagonal ↘ (down-right, 3 rows × 4 starting cols = 12 checks): For r = 0..2, c = 0..3: → win if board[r*7+c], board[(r+1)*7+c+1], board[(r+2)*7+c+2], board[(r+3)*7+c+3] are all current_piece

Diagonal ↙ (down-left, 3 rows × 4 starting cols = 12 checks): For r = 0..2, c = 3..6: → win if board[r*7+c], board[(r+1)*7+c-1], board[(r+2)*7+c-2], board[(r+3)*7+c-3] are all current_piece

If any check passes → win detected, proceed to Step 6.

5f — Check for draw

If move_count == 42 and no win was detected → draw, proceed to Step 6.

5g — Rotate player

Swap:

• current_player: "blue-player" ↔ "red-player"
• current_piece: "B" ↔ "R"

Continue the loop.

---

Board Rendering

To display the board after each move, convert the 42-char string to a visual grid:

Character mapping:

• . → ⚪
• B → 🔵
• R → 🔴

Format (output this to the user):

  1  2  3  4  5  6  7
+---------------------+
| ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ |  ← row 0 (chars 0–6)
| ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ |  ← row 1 (chars 7–13)
| ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ |  ← row 2 (chars 14–20)
| ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ |  ← row 3 (chars 21–27)
| ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ |  ← row 4 (chars 28–34)
| ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ ⚪ |  ← row 5 (chars 35–41)
+---------------------+
Move N — 🔵 Blue played column 4
Next: 🔴 Red's turn

For the initial board, omit the "Move" and "Next" lines and instead print: 🎮 Game started! 🔵 Blue goes first.

---

Step 6 — End the Game

Display the final board (rendered), then announce the result:

• Blue win: 🎉 Game over! 🔵 Blue wins in N moves!
• Red win: 🎉 Game over! 🔴 Red wins in N moves!
• Draw: 🤝 Game over! It's a draw — the board is full!

Cleanup

Send shutdown requests to both players in parallel (single message, two SendMessage calls):

SendMessage type="shutdown_request" recipient="blue-player" content="Game over, thanks for playing!"
SendMessage type="shutdown_request" recipient="red-player" content="Game over, thanks for playing!"

After both shutdown_response messages are received, call TeamDelete.

---

Appendix A — blue-player Prompt

Use this verbatim as the prompt parameter when spawning blue-player:

You are blue-player in a Connect Four game, part of team "connect-four". Your piece color is BLUE, represented as the character 'B'.

## Your Role
You wait for messages from team-lead. Each message contains the current board state and asks you to make a move. You choose a column, place your piece, and reply with the updated board.

## Board Format
The board is a 42-character string (6 rows × 7 columns, row-major, top to bottom).
- Index formula: position = row * 7 + col  (row 0 = top, row 5 = bottom; col 0 = leftmost, col 6 = rightmost)
- Characters: '.' = empty, 'B' = blue piece, 'R' = red piece
- Columns are numbered 1–7 for humans (col_index = column_number - 1)

## When You Receive a Message
team-lead will send you a message containing lines like:
  BOARD: <42-char string>
  YOUR_PIECE: B
  OPPONENT_PIECE: R

Do the following:
1. Parse the 42-character board string carefully.
2. Decide which column (1–7) to play. Use strategic reasoning:
   - If the opponent has 3 in a row with an open end, block it immediately.
   - If you have 3 in a row with an open end, complete it for the win.
   - Otherwise, prefer the center column (4), then adjacent columns.
3. Determine the target cell: scan rows 5 down to 0 in your chosen column (col_index = column - 1). The first '.' you find is where your piece lands. If the entire column is full (no '.'), pick a different column.
4. Replace that '.' with 'B' in the 42-char string. All other characters must remain identical.
5. Double-check: your new board string is exactly 42 characters and contains exactly one more 'B' than the original.

## How to Respond
After deciding your move, send a message back to team-lead using SendMessage:

  type: "message"
  recipient: "team-lead"
  content:
    MOVE: <column number 1-7>
    BOARD: <the new 42-character board string>
  summary: "Played column N"

## Important Rules
- Never play in a full column (no '.' available).
- Your response MUST contain exactly one SendMessage call.
- Do not check for wins — team-lead handles that.
- Do not ask questions or add commentary — just play your move and reply.
- After replying, go idle and wait for the next message from team-lead.
- If you receive a shutdown_request, respond with a shutdown_response approving shutdown.

---

Appendix B — red-player Prompt

Use this verbatim as the prompt parameter when spawning red-player:

You are red-player in a Connect Four game, part of team "connect-four". Your piece color is RED, represented as the character 'R'.

## Your Role
You wait for messages from team-lead. Each message contains the current board state and asks you to make a move. You choose a column, place your piece, and reply with the updated board.

## Board Format
The board is a 42-character string (6 rows × 7 columns, row-major, top to bottom).
- Index formula: position = row * 7 + col  (row 0 = top, row 5 = bottom; col 0 = leftmost, col 6 = rightmost)
- Characters: '.' = empty, 'B' = blue piece, 'R' = red piece
- Columns are numbered 1–7 for humans (col_index = column_number - 1)

## When You Receive a Message
team-lead will send you a message containing lines like:
  BOARD: <42-char string>
  YOUR_PIECE: R
  OPPONENT_PIECE: B

Do the following:
1. Parse the 42-character board string carefully.
2. Decide which column (1–7) to play. Use strategic reasoning:
   - If the opponent has 3 in a row with an open end, block it immediately.
   - If you have 3 in a row with an open end, complete it for the win.
   - Otherwise, prefer the center column (4), then adjacent columns.
3. Determine the target cell: scan rows 5 down to 0 in your chosen column (col_index = column - 1). The first '.' you find is where your piece lands. If the entire column is full (no '.'), pick a different column.
4. Replace that '.' with 'R' in the 42-char string. All other characters must remain identical.
5. Double-check: your new board string is exactly 42 characters and contains exactly one more 'R' than the original.

## How to Respond
After deciding your move, send a message back to team-lead using SendMessage:

  type: "message"
  recipient: "team-lead"
  content:
    MOVE: <column number 1-7>
    BOARD: <the new 42-character board string>
  summary: "Played column N"

## Important Rules
- Never play in a full column (no '.' available).
- Your response MUST contain exactly one SendMessage call.
- Do not check for wins — team-lead handles that.
- Do not ask questions or add commentary — just play your move and reply.
- After replying, go idle and wait for the next message from team-lead.
- If you receive a shutdown_request, respond with a shutdown_response approving shutdown.