v0.4.2: Multi-AI Consultation Loop

Features

• 🤖 Multi-AI Consultation Loop: Gemini now reviews Claude's work after each iteration and provides feedback for the next run
• 🔄 --loop-until-finish flag: New loop mode that respects completion signals from Gemini or Claude
• ✅ Improved completion detection: Loop now properly detects "Goal achieved!" in Claude's responses

Improvements

• Better Gemini prompt format using structured GOAL_ACHIEVED/CONTINUE/NEXT_FOCUS keys
• Strip markdown formatting for more reliable "Goal achieved!" detection
• Capture Claude's actual output text instead of just git log summary
• Gemini feedback now guides subsequent iterations

Bug Fixes

• Fixed --loop-until-finish not detecting task completion
• Fixed summary not containing Claude's actual response text

What's Changed

• --loop: Runs indefinitely, ignoring completion signals
• --loop-until-finish: Runs until Gemini/Claude determines task is complete
• Gemini reviews each iteration and decides whether to continue or provide specific focus areas

Full changelog: v0.4.1...v0.4.2

v0.4.1

Bug fixes:

• Fix stdin handling to allow user input during Claude execution
• Prevent resource leak by properly closing stdin file descriptor
• Clean dist directory before building in CI

This release fixes issues with auto-accept mode and ensures proper cleanup in the publish workflow.

v0.4.0 - Cost Control with Model Selection

🎉 What's New in v0.4.0

Cost Control & Model Selection

Take control of your AI costs! Choose from cost-effective to premium models.

Breaking Changes

• --auto-gemini-answer renamed to --auto-answer (clearer naming)

New Features

Multiple AI Models:

• OpenAI: gpt-4o-mini (cheapest), gpt-4o, gpt-5.2 (premium reasoning)
• Gemini: gemini-1.5-flash (cheapest), gemini-1.5-pro, gemini-3-pro-preview

Cost-Effective Defaults:

• Auto mode now uses gpt-4o-mini or gemini-1.5-flash (cheapest options)
• Save ~95% on API costs compared to premium models

Model Selection:

# Cost-effective (default)
cook --loop --auto-answer

# Choose specific model
cook --loop --auto-answer --ai-model gpt-4o-mini      # $0.15/$0.60 per 1M tokens
cook --loop --auto-answer --ai-model gpt-4o           # $2.50/$10 per 1M tokens
cook --loop --auto-answer --ai-model gpt-5.2          # $10/$40 per 1M tokens (max reasoning)
cook --loop --auto-answer --ai-model gemini-1.5-flash # $0.075/$0.30 per 1M tokens (cheapest)
cook --loop --auto-answer --ai-model gemini-1.5-pro   # $1.25/$5 per 1M tokens

Cost Comparison

Model	Input	Output	Use Case
gemini-1.5-flash	$0.075/1M	$0.30/1M	💰 Cheapest option
gpt-4o-mini	$0.15/1M	$0.60/1M	💰 Cheapest OpenAI
gemini-1.5-pro	$1.25/1M	$5/1M	⚖️ Balanced
gpt-4o	$2.50/1M	$10/1M	⚖️ Balanced OpenAI
gpt-5.2	$10/1M	$40/1M	🧠 Max reasoning

Documentation

• Added comprehensive cost comparison
• Updated all examples with new flag name
• Included pricing information

Full Changelog: v0.3.0...v0.4.0

Release v0.3.0 - GPT-5.2 & Gemini 3 Pro Support

🚀 Major New Features

Multi-AI Support

We've added support for GPT-5.2 (OpenAI's latest flagship model) and upgraded to Gemini 3 Pro Preview (Google's most intelligent model), giving you access to the most advanced AI reasoning available!

🤖 GPT-5.2 Integration

• Model: gpt-5.2 - OpenAI's best general-purpose model
• Reasoning: xhigh (maximum reasoning effort)
• Output: 65,536 max tokens
• Best for: Complex reasoning, broad world knowledge, multi-step agentic tasks

🧠 Gemini 3 Pro Preview

• Model: gemini-3-pro-preview - Google's most intelligent multimodal model
• Output: 65,536 max tokens
• Best for: State-of-the-art reasoning and multimodal understanding
• Automatic fallback: Used if GPT-5 fails or no OpenAI key available

⚡ Smart AI Selection

The automator now intelligently selects the best AI model:

1. Try GPT-5.2 first (if OPENAI_API_KEY is set)
2. Fall back to Gemini 3 Pro (if GEMINI_API_KEY is set)
3. Fall back to auto-answer "y" (if no keys available)

📦 Installation

pip install --upgrade let-claude-code

🔧 Setup

Add one or both API keys to your .env file:

# Option 1: Use GPT-5.2 (preferred for max reasoning)
OPENAI_API_KEY=sk-...

# Option 2: Use Gemini 3 Pro (fallback)
GEMINI_API_KEY=...

# Optional: Telegram notifications
TG_BOT_TOKEN=...
TG_CHAT_ID=...

🚀 Usage

# Run with AI auto-answer (tries GPT-5, then Gemini)
cook --loop -m fix_bugs --auto-gemini-answer -y

# Or use the wrapper script
./cook --loop -m fix_bugs --auto-gemini-answer -y

🔑 Get API Keys

• GPT-5: https://platform.openai.com/api-keys
• Gemini: https://aistudio.google.com/app/apikey

🐛 Bug Fixes (from v0.2.2)

• Fixed stdin handling for Claude CLI - prompts are now sent and closed properly
• Resolved loop hang issues when Claude requests input
• Added safety checks for closed stdin in resumed sessions
• Improved error handling for AI API failures

📝 What's Changed

Full Changelog: v0.2.2...v0.3.0

When Claude asks a question during automation, the most advanced AI models will now answer it with maximum reasoning capabilities! 🎉

Release v0.2.2

🐛 Bug Fixes

Loop Hang Issues Resolved

• Fixed loop hang when Claude asks questions: Now auto-answers 'y' when Gemini is not enabled
• Fixed stdin handling: Properly closes stdin when not needed to prevent process hangs
• Added loop delay: Implements 10-second delay when runs complete quickly (< 30s) to prevent rapid failures

Gemini Integration Improvements

• Enhanced error logging: Added detailed HTTP/URL error messages for easier debugging
• Better error handling: Improved fallback behavior when Gemini API fails

UX Improvements

• Added --auto-yes flag: Skip lock file confirmation prompts for automated workflows
• Created 'cook' wrapper script: Bypasses Python module caching issues for reliable execution
• Fixed module loading: Resolved init.py import caching problems

📦 Installation

pip install --upgrade let-claude-code

🚀 Usage

Use the new wrapper script to avoid caching issues:

# Basic usage
./cook --loop -m fix_bugs -y

# With Gemini auto-answer
./cook --loop -m fix_bugs --auto-gemini-answer -y

Or use the installed command:

cook --loop -m fix_bugs -y

🔧 Technical Details

This release resolves a critical issue where the automator would hang indefinitely when Claude requested user input and Gemini integration was not enabled. The fix ensures all input_required messages are handled regardless of the Gemini configuration.

Release v0.2.1

feat: prompt to remove stale lock file

fix: re-add input_required handling for --auto-gemini-answer

Release v0.2.0

fix: remove input_required handling that was causing hangs

The input_required handling was causing the loop to hang when Claude asked questions. Removed to restore original behavior.

Release v0.1.9

fix: support TELEGRAM_BOT_TOKEN env var

feat: load .env file for environment variables

Release v0.1.8

feat: Add --auto-gemini-answer for autonomous Claude operation

• Auto-answer Claude's questions using Gemini API
• Sends TG notifications when Claude asks questions and Gemini answers
• Prompts for GEMINI_API_KEY if not set in env
• Support TELEGRAM_CHAT_ID env var

Release v0.1.7

Fix: Expand ~ in --claude flags