feat: default claude with codex switches

Author: ourines

README.md

@@ -1,10 +1,10 @@
 # Worktree Task Plugin
 
-Manage large coding tasks using git worktrees and background Claude Code sessions.
+Manage large coding tasks using git worktrees and background agent sessions（默认启动 Claude Code `claude --dangerously-skip-permissions`，可用 `--codex` 快速切换 Codex CLI，或 `--agent-cmd` 指定任意命令，如 `codex --yolo -m gpt-5.1-codex-max -c model_reasoning_effort="high"`）。
 
 ## Features
 
-- **Launch**: Spawn autonomous Claude Code instances in separate git worktrees
+- **Launch**: Spawn autonomous agent instances (默认 Claude Code，可选 `--codex`/`--agent-cmd`) in separate git worktrees
 - **Status**: Monitor running tasks and their progress
 - **Resume**: Recover interrupted tasks (rate limits, API errors, timeouts)
 - **Cleanup**: Remove completed tasks and worktrees
@@ -81,7 +81,7 @@ worktree-task/
 
 ## How It Works
 
-1. **Launch**: Creates a git worktree, starts a tmux session, runs Claude Code with the task prompt
+1. **Launch**: Creates a git worktree, starts a tmux session, runs Claude Code（或通过 `--codex`/`--agent-cmd` 指定的命令）with the task prompt
 2. **Monitor**: Polls tmux panes to detect completion, errors, or rate limits
 3. **Alert**: Sends macOS notifications via `osascript`
 4. **Resume**: Detects interrupted state and restarts with context

commands/launch.md

@@ -5,7 +5,7 @@ allowed-tools: Bash
 
 # Launch Worktree Task
 
-Launch a background Claude Code session in a separate git worktree to execute a large task autonomously.
+默认启动 Claude Code (`claude --dangerously-skip-permissions`)，可选参数：`--codex` 快速切换到 Codex CLI，或用 `--agent-cmd "codex --yolo -m gpt-5.1-codex-max -c model_reasoning_effort=\"high\""` 自定义命令；在单独的 git worktree 中自主执行大任务。
 
 ## Usage
 

scripts/launch.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 """
-Launch a background Claude Code session in a git worktree.
+Launch a background agent session (Codex CLI) in a git worktree.
 
 Usage: launch.py <branch-name> "<task-description>"
 """
@@ -11,6 +11,11 @@
 import time
 from pathlib import Path
 
+# Default agent command (Claude Code)
+DEFAULT_AGENT_CMD = 'claude --dangerously-skip-permissions'
+# Codex agent command (optional)
+CODEX_AGENT_CMD = 'codex --yolo -m gpt-5.1-codex-max -c model_reasoning_effort="high"'
+
 
 def run(cmd: str, check: bool = True, capture: bool = False) -> subprocess.CompletedProcess:
     """Run a shell command."""
@@ -35,13 +40,13 @@ def session_exists(name: str) -> bool:
     return result.returncode == 0
 
 
-def wait_for_claude_ready(session_name: str, timeout: int = 30) -> bool:
+def wait_for_agent_ready(session_name: str, timeout: int = 30) -> bool:
     """
-    Wait for Claude Code to be ready by detecting the prompt.
+    Wait for the CLI agent to be ready by detecting the prompt.
 
-    Returns True if Claude is ready, False if timeout.
+    Returns True if the agent is ready, False if timeout.
     """
-    print(f"  Waiting for Claude Code to initialize (timeout: {timeout}s)...")
+    print(f"  Waiting for agent to initialize (timeout: {timeout}s)...")
 
     for i in range(timeout):
         result = run(f"tmux capture-pane -t {session_name} -p", capture=True)
@@ -51,11 +56,11 @@ def wait_for_claude_ready(session_name: str, timeout: int = 30) -> bool:
         lines = output.strip().split('\n')
         last_lines = lines[-5:] if len(lines) >= 5 else lines
 
-        # Check for Claude Code ready indicators
+        # Check for agent ready indicators
         for line in last_lines:
             # Look for the prompt (>) or bypass permissions message
             if '>' in line or 'bypass permissions' in line.lower():
-                print(f"  ✓ Claude Code ready (took {i+1}s)")
+                print(f"  ✓ Agent ready (took {i+1}s)")
                 return True
 
         # Show progress every 5 seconds
@@ -64,7 +69,7 @@ def wait_for_claude_ready(session_name: str, timeout: int = 30) -> bool:
 
         time.sleep(1)
 
-    print(f"  ⚠ Warning: Claude Code may not be fully ready after {timeout}s")
+    print(f"  ⚠ Warning: Agent may not be fully ready after {timeout}s")
     return False
 
 
@@ -95,17 +100,20 @@ def load_task_template(script_dir: Path, task_desc: str, worktree_dir: str) -> s
 
 def main():
     if len(sys.argv) < 3:
-        print("Usage: launch.py <branch-name> \"<task-description>\" [--env KEY=VALUE ...]")
+        print("Usage: launch.py <branch-name> \"<task-description>\" [--env KEY=VALUE ...] [--agent-cmd \"<agent command>\"] [--claude] [--codex]")
         print("Example: launch.py feature/my-task \"Implement the new feature\"")
         print("Example: launch.py feature/my-task \"Task\" --env ANTHROPIC_BASE_URL=http://api.codex.markets")
+        print("Example: launch.py feature/my-task \"Task\" --agent-cmd \"codex --yolo -m gpt-5.1-codex-max -c model_reasoning_effort=\\\"high\\\"\"")
+        print("Example: launch.py feature/my-task \"Task\" --codex")
         sys.exit(1)
 
     branch_name = sys.argv[1]
     task_desc = sys.argv[2]
     script_dir = Path(__file__).parent.resolve()
 
-    # Parse custom environment variables
+    # Parse custom environment variables and agent command override
     custom_env = {}
+    agent_cmd_override = None
     i = 3
     while i < len(sys.argv):
         if sys.argv[i] == "--env" and i + 1 < len(sys.argv):
@@ -114,6 +122,15 @@ def main():
                 key, value = env_pair.split("=", 1)
                 custom_env[key] = value
             i += 2
+        elif sys.argv[i] == "--agent-cmd" and i + 1 < len(sys.argv):
+            agent_cmd_override = sys.argv[i + 1]
+            i += 2
+        elif sys.argv[i] == "--codex":
+            agent_cmd_override = CODEX_AGENT_CMD
+            i += 1
+        elif sys.argv[i] == "--claude":
+            agent_cmd_override = DEFAULT_AGENT_CMD
+            i += 1
         else:
             i += 1
 
@@ -173,28 +190,34 @@ def main():
     # Wait for shell to initialize
     time.sleep(1)
 
-    # Launch Claude Code
-    print("Launching Claude Code...")
+    # Launch agent
+    print("Launching agent...")
 
     # Build command with custom environment variables
+    agent_cmd_base = agent_cmd_override if agent_cmd_override else DEFAULT_AGENT_CMD
     if custom_env:
         env_exports = " && ".join([f"export {k}='{v}'" for k, v in custom_env.items()])
-        claude_cmd = f"{env_exports} && claude --dangerously-skip-permissions"
+        agent_cmd = f"{env_exports} && {agent_cmd_base}"
         print(f"  Using custom environment variables: {', '.join(custom_env.keys())}")
     else:
-        claude_cmd = "claude --dangerously-skip-permissions"
+        agent_cmd = agent_cmd_base
+
+    if agent_cmd_override:
+        print(f"  Agent command override: {agent_cmd_base}")
+    else:
+        print(f"  Agent command (default): {agent_cmd_base}")
 
-    run(f"tmux send-keys -t {session_name} '{claude_cmd}' Enter")
+    run(f"tmux send-keys -t {session_name} '{agent_cmd}' Enter")
 
-    # Wait for Claude Code to be ready (with timeout protection)
-    if not wait_for_claude_ready(session_name, timeout=30):
+    # Wait for agent to be ready (with timeout protection)
+    if not wait_for_agent_ready(session_name, timeout=30):
         print("  Proceeding anyway, but task may not start correctly...")
 
     # Extra buffer time to ensure input is ready
     time.sleep(1)
 
     # Load and send task prompt
-    print("Sending task to Claude...")
+    print("Sending task to agent...")
     task_prompt = load_task_template(script_dir, task_desc, str(worktree_dir))
 
     # Use a temp file to avoid shell escaping issues

skills/worktree-task/SKILL.md

@@ -1,11 +1,11 @@
 ---
 name: worktree-task
-description: Manage large coding tasks using git worktrees and background Claude Code sessions. Use this when users want to execute large, complex tasks (like implementing a new service, major refactoring, or multi-step features) in isolation without blocking the current session. Spawns autonomous Claude Code instances via tmux.
+description: 使用 git worktree + 背景代理会话管理大型任务（默认启动 Claude Code；可用 --codex 快速切到 Codex CLI，或通过 --agent-cmd 指定任意命令）。适合用户希望在不阻塞当前会话的情况下执行大型或多步骤任务。
 ---
 
 # Worktree Task Manager
 
-This skill manages large coding tasks by spawning autonomous Claude Code instances in separate git worktrees via tmux sessions.
+This skill manages large coding tasks by spawning autonomous agent instances (默认 Claude Code，可选 `--codex`/`--agent-cmd`) in separate git worktrees via tmux sessions.
 
 ## When to Use
 
@@ -42,7 +42,7 @@ The script will:
 1. Verify git status is clean (or prompt to commit/stash)
 2. Create a git worktree with the specified branch
 3. Create a tmux session
-4. Launch Claude Code with `--dangerously-skip-permissions`
+4. Launch Claude Code by default (use `--codex` for default Codex command or `--agent-cmd` to switch, e.g., `codex --yolo -m gpt-5.1-codex-max -c model_reasoning_effort="high"`)
 5. Send the task with instructions to use Task tool for each phase
 
 ### 2. Monitor Progress
@@ -82,7 +82,7 @@ Configure alerts in `hooks/hooks.json`.
 
 ## Critical Instructions for Spawned Claude
 
-The spawned Claude Code instance receives these critical instructions:
+The spawned agent receives these critical instructions:
 
 1. **MUST use Task tool** - Each major phase must be executed via `Task` tool to prevent context overflow
 2. **Silent mode** - No confirmations needed, user has pre-approved all operations