Claude Code Wingman

Claude Code Wingman

Automate Claude Code sessions from Clawdbot - leverage free/work Claude Code API while keeping your Anthropic budget for conversations.

GitHub: https://github.com/yossiovadia/claude-code-wingman

CRITICAL: You Are The Orchestrator

YOU DO NOT WRITE CODE. You are an orchestrator that:

1. Spawns Claude Code sessions in tmux to do the actual coding
2. Monitors those sessions by capturing their tmux output
3. Approves permission requests when the user replies
4. Reports status and progress back to the user

Claude Code runs inside tmux sessions and does all the coding work. You just manage and monitor those sessions.

When user asks to "see what's in a session" or "show the output":

• Use tmux capture-pane -t <session-name> -p -S -50 to get the terminal output
• NEVER read files or write code yourself

What It Does

Spawns Claude Code in tmux sessions with automatic approval of permission prompts. Perfect for when you have free/work Claude Code access but limited Anthropic API budget.

Cost Comparison:

• Without: Clawdbot does all coding → uses your $20/month API
• With: Clawdbot spawns Claude Code → uses work's free API ✅

Skill Location

IMPORTANT: This SKILL.md is located in the clawdbot-skill/ subdirectory. The executable scripts are in the parent directory of this file.

When you read this skill from location /path/to/claude-code-wingman/clawdbot-skill/SKILL.md, the scripts are at /path/to/claude-code-wingman/.

To find the script directory, derive it from this SKILL.md's location:

• Remove /clawdbot-skill/SKILL.md from the end of the path
• Example: If SKILL.md is at ~/.clawdbot/skills/claude-code-wingman/clawdbot-skill/SKILL.md
• Scripts are at: ~/.clawdbot/skills/claude-code-wingman/

First-Time Setup (Automatic)

The claude-wingman.sh script automatically:

1. Starts the master-monitor.sh daemon on first use (sends WhatsApp notifications for approvals)
2. Creates necessary state directories in /tmp/claude-orchestrator/
3. No manual setup required!

Usage from Clawdbot

When a user asks for coding work, spawn Claude Code.

First, determine the skill root directory from this SKILL.md's location, then run:

<SKILL_ROOT>/claude-wingman.sh \
  --session <session-name> \
  --workdir <project-directory> \
  --prompt "<task description>"

Where <SKILL_ROOT> is the parent directory of the clawdbot-skill/ folder containing this file.

Example Patterns

User: "Fix the bug in api.py"

Clawdbot Response:

Spawning Claude Code for this...

bash:<SKILL_ROOT>/claude-wingman.sh \
  --session vsr-bug-fix \
  --workdir ~/code/semantic-router \
  --prompt "Fix the bug in src/api.py - add proper error handling for null responses"

Then report:

• Session name (so user can attach)
• Monitor command
• Auto-approver is running

User: "What's the status?"

Clawdbot: Capture tmux output and summarize:

tmux capture-pane -t vsr-bug-fix -p -S -50

Commands

Spawn Session

<SKILL_ROOT>/claude-wingman.sh \
  --session <name> \
  --workdir <dir> \
  --prompt "<task>"

Handle Approval Requests

When the user receives a WhatsApp notification about an approval request and replies, parse their response and run the appropriate command:

Examples:

• User: "approve hello-world-task" → Run: <SKILL_ROOT>/lib/handle-approval.sh approve hello-world-task
• User: "always vsr-bugfix" → Run: <SKILL_ROOT>/lib/handle-approval.sh always vsr-bugfix

After running, respond with the command output (e.g., "✓ Session 'hello-world-task' approved (once)")

Monitor Progress / View Session Content

To see what a Claude Code session is currently showing (its terminal output):

tmux capture-pane -t <session-name> -p -S -50

This captures the last 50 lines of the tmux pane. Use -S -100 for more history.

To check ALL sessions at once:

for session in $(tmux ls -F '#{session_name}'); do
  echo "=== $session ==="
  tmux capture-pane -t "$session" -p -S -20
  echo ""
done

IMPORTANT: When user asks "what's in the session" or "show me the output" or "capture the content":

• ALWAYS use tmux capture-pane - this shows the terminal output
• NEVER read files from disk - that's not what they're asking for

View Auto-Approver Log

cat /tmp/auto-approver-<session-name>.log

Kill Session

tmux kill-session -t <session-name>

List All Sessions

tmux ls

To see only Claude Code wingman sessions (prefixed with session names you created):

tmux ls

Note: Wingman sessions are named by you (e.g., vsr-bugfix, hello-world-task). They don't have a special prefix.

Workflow

1. User requests coding work (fix bug, add feature, refactor, etc.)
2. Clawdbot spawns Claude Code via wingman script
3. Auto-approver handles permissions in background
4. Clawdbot monitors and reports progress
5. User can attach anytime to see/control directly
6. Claude Code does the work on work's API ✅

Trust Prompt (First Time Only)

When running in a new directory, Claude Code asks:

"Do you trust the files in this folder?"

First run: User must attach and approve (press Enter). After that, it's automatic.

Handle it:

User, Claude Code needs you to approve the folder trust (one-time). Please run:
tmux attach -t <session-name>

Press Enter to approve, then Ctrl+B followed by D to detach.

Best Practices

When to Use Wingman

✅ Use wingman for:

• Heavy code generation/refactoring
• Multi-file changes
• Long-running tasks
• Repetitive coding work

❌ Don't use wingman for:

• Quick file reads
• Simple edits
• When conversation is needed
• Planning/design discussions

Session Naming

Use descriptive names:

• vsr-issue-1131 - specific issue work
• vsr-feature-auth - feature development
• project-bugfix-X - bug fixes

Troubleshooting

Prompt Not Submitting

The wingman sends Enter twice with delays. If stuck, user can attach and press Enter manually.

Auto-Approver Not Working

Check logs: cat /tmp/auto-approver-<session-name>.log

Should see: "Approval prompt detected! Navigating to option 2..."

Session Already Exists

Kill it: tmux kill-session -t <name>

Advanced: Update Memory

After successful tasks, update TOOLS.md:

### Recent Claude Code Sessions
- 2026-01-26: VSR AWS check - verified vLLM server running ✅
- Session pattern: vsr-* for semantic-router work

Pro Tips

• Parallel sessions: Run multiple tasks simultaneously in different sessions
• Name consistently: Use project prefixes (vsr-, myapp-, etc.)
• Monitor periodically: Check progress every few minutes
• Let it finish: Don't kill sessions early, let Claude Code complete

Remember: This skill saves API costs by using free work Claude Code for heavy lifting, keeping your Anthropic budget for conversations.