Claude Code unable to start with auto enable permissions, fails to indicate user

[warm patrol]

This is a Claude Code permissions issue! When Claude Code needs user approval for commands (like running git, file modifications, etc.), it can't get that input through Telegram — so it hangs.

Solution: Auto-approve permissions

You have a few options:

1. Use --dangerously-skip-permissions flag

Pass this when invoking Claude Code:

bash pty:true command:"claude --dangerously-skip-permissions 'Your task'"

2. Set environment variable

Add CLAUDE_SKIP_PERMISSIONS=true to your shell env or Clawdbot's env config.

3. In Clawdbot config (recommended)

Set the env var globally in your ~/.clawdbot/clawdbot.json:

{
  "env": {
    "CLAUDE_SKIP_PERMISSIONS": "true"
  }
}

Then restart the gateway: clawdbot gateway restart

⚠️ Security note: --dangerously-skip-permissions gives the agent full control. Only use this in trusted environments or when you're OK with the agent running any command without asking.

For the coding-agent skill in Clawdbot, you can also configure it to always pass this flag. Want me to show you how to modify the skill config?

Enable verbose mode for Claude Code:

1. Via command flag

bash pty:true command:"claude --verbose 'Your task'"

2. Via environment variable (persistent)

Add to your Clawdbot config:

{
  "env": {
    "CLAUDE_DEBUG": "1"
  }
}

Or for even more detail:

{
  "env": {
    "CLAUDE_LOG_LEVEL": "debug"
  }
}

Then clawdbot gateway restart.

3. For the coding-agent skill specifically

You can modify the skill call to always include these:

bash pty:true command:"CLAUDE_DEBUG=1 claude --verbose 'Your task'"

Or create a wrapper script that sets these env vars and call that instead.

The verbose output will come through in the process logs, which Clawdbot captures.