Refactoring Summary: Direct Claude Code Integration

Completed: 2026-01-27

Objective

Remove Gemini dependency and simplify bot architecture to pass all Telegram messages directly to Claude Code CLI.

Architecture Change

Before:

Telegram Message
  ↓
Intent Parser (Gemini) → Skill Extraction
  ↓
Workflow Orchestrator → Confirmation Flow
  ↓
Claude Code Executor (skill-specific)
  ↓
Results

After:

Telegram Message
  ↓
Claude Code CLI --print
  ↓
Results (with progress updates)

Files Modified

File	Lines Changed	Description
parsers/intent_parser.py	115 → 34	Removed Gemini LLM logic, only extracts URLs
workflows/orchestrator.py	264 → 159	Removed confirmation flow, direct execution
executors/claude_executor.py	+84 lines	Added execute_direct() method
handlers/message_handler.py	98 → 77	Simplified, removed confirmation UI
requirements.txt	-1 dep	Removed google-generativeai
config.py	-1 field	Removed gemini_api_key
.env.example	-1 line	Removed GEMINI_API_KEY
README.md	Updated	New architecture diagram & features

Benefits

1. Simpler Architecture: Single integration point (Claude Code CLI)

2. No External Dependencies: Removed Gemini API requirement

3. No API Management: No keys, quotas, or rate limits to manage

4. More Predictable: Claude Code handles all interpretation consistently

5. Fewer Dependencies: 4 Python packages instead of 5

Key Changes

Intent Parser (Simplified)

# Before: 115 lines with Gemini LLM
# After: 34 lines - just URL extraction
class IntentParser:
    def parse(self, user_message: str) -> WorkflowIntent:
        url = self._extract_url(user_message)
        return WorkflowIntent(raw_text=user_message, url=url)

Claude Executor (New Method)

async def execute_direct(self, user_message: str, progress_callback=None):
    """Execute Claude Code directly with user's message"""
    cmd = [self.claude_bin, "--print", user_message]
    # Streams output and extracts created files

Message Handler (Simplified)

# Before: Confirmation flow with inline keyboards
# After: Direct execution
workflow_id = await orchestrator.start_workflow(
    user_message=user_message,
    chat_id=chat_id,
    progress_callback=progress_callback
)

Testing Status

✅ Configuration validated ✅ All imports successful ✅ Bot token configured ✅ Orchestrator initialized

Next Steps for Users

1. Start the bot:

   ./start_bot.sh

2. Test with simple message: Send via Telegram: "create a simple hello world Python script"

3. Test with URL: Send via Telegram: "analyze https://www.anthropic.com"

4. Test with complex task: Send via Telegram: "analyze article https://example.com/article, create infographic, commit to github"

Rollback Instructions

If needed, revert to Gemini-based version:

git log --oneline | head -5  # Find commit hash
git revert e88d9aa  # Revert the refactoring commit
pip install google-generativeai
# Add GEMINI_API_KEY back to .env

Commit Details

Commit: e88d9aa Message: refactor: Remove Gemini dependency, use direct Claude Code CLI integration Files Changed: 8 files changed, 171 insertions(+), 320 deletions(-) Author: Co-Authored-By: Claude Sonnet 4.5 [e-mail]