ATLAS-v0.3.5

New Features

Atlas now supports two modes of tool orchestration:

Deterministic Mode (Default)

• System-controlled tool selection
• One tool at a time based on edit format
• Tools: replace_lines, write_file

LLM-Controlled Mode (New)

• Enable with --llm-controlled-tools flag
• LLM chooses which tools to use
• Multiple tools available simultaneously
• Supports complex multi-step workflows

Usage

# Enable LLM-controlled tool orchestration
atlas --llm-controlled-tools

Security Fixes

• urllib3>=2.6.3: Fixed decompression-bomb safeguards bypass vulnerability (CVE)
• aiohttp>=3.13.3: Fixed multiple security vulnerabilities including:
  • Zip bomb vulnerability
  • Denial of Service (DoS) vulnerabilities
  • Other security issues

These updates address Dependabot security alerts and improve the overall security posture of the application.

Full Changelog: v0.3.4...v0.3.5

ATLAS-v0.3.4

New Features

• Added /whole command: Switch to wholefile edit format during an active session
• Support for latest OpenAI models: Added metadata and settings for newest OpenAI models

Improvements

• Enhanced patch prompts for better code generation
• Updated CodeBLEU visualization with improved instructions
• Updated CA (Correctness Analysis) visualization to compare Atlas and LLM-controlled orchestration approaches
• Added CA evaluation framework

Documentation

• Refactored architecture documentation styles
• Added command instructions to README

Maintenance

• Updated GitHub Actions dependencies (checkout v6, upload-artifact v6)
• Removed unnecessary files

Full Changelog: v0.3.4...v0.3.4

ATLAS-v0.3.3

Security fixes

• Fixed command injection vulnerabilities in run_cmd_subprocess, editor.py, and commands.py (#92)
• Replaced insecure tempfile.mktemp() with NamedTemporaryFile in tests (#92)

Version updates

• Updated Python requirement to 3.14+ across all configuration files
• Updated CI/CD workflows to use Python 3.14
• Bumped outdates dependencies with dependabot (#84, #85)

CI/CD improvements

• Fixed Dependabot configuration
• Added permissions to GitHub Actions workflows (#89)
• Removed obsolete conda workflow (#88)

Full Changelog: v0.3.3...v0.3.3

ATLAS-v0.3.2

Features

• Legacy language support: Added tree-sitter tags for 13 languages to improve repo map generation
  • Legacy: COBOL, Fortran, Ada, Pascal, Perl, Tcl, Assembly, Objective-C
  • Hardware: VHDL, Verilog
  • Modern: Nix, PHP, Scala, TypeScript
• Total: 44 languages now supported with repo map tags

Bug Fixes

• Fixed ValueError when using regular OpenAI API keys (non-Copilot format) (#80)

Other Changes

• Removed deprecated remove_reasoning model setting (#64)
• Updated documentation URLs to point to GitHub README
• Added L2M CLI screenshot to README

Full Changelog: v0.3.2...v0.3.2

ATLAS-v0.3.1

What's New

Features:

• Added scatter plot visualizations with regression analysis for CodeBLEU evaluation
• Enhanced data structure and documentation

What's Changed

• Fix Discord link in README.md by @nolanlwin in #56
• Fixed LiteLLM API Connection Error by @nolanlwin in #60
• Fixed Python 3.13 compatibility with an audioop error by @nolanlwin in #61

Full Changelog: v0.3.0...v0.3.1

ATLAS-v0.3.0

What's New

Evaluation Benchmarks

• CodeBLEU Benchmark: Added comprehensive CodeBLEU evaluation module for measuring code similarity between generated and reference code

  • Supports multiple programming languages (Python, C, C#, C++, Java, JavaScript, PHP, Go, Ruby, Rust)
  • Configurable weights for n-gram match, weighted n-gram, AST match, and data-flow match
  • Automated batch evaluation with summary statistics
  • Integration with L2M modernization workflow

• Harbor Benchmark: Added Harbor (Terminal) benchmarking integration

  • L2M agent implementation for containerized agent evaluation
  • Automated installation template for Docker containers
  • Support for iterative refinement workflows

Fixed

Bug Fixes

• API Key Priority: Improved API key selection priority and JSON serialization handling
• Test Failures: Fixed failing tests across multiple test files
• Git Configuration: Fixed Git commit attribution to use correct user name/email
• Repository URL: Updated remote URL to new repository location (legacy2modern → l2m)

Error Handling

• Rate Limit Errors: Enhanced error messages for free-tier vs paid model rate limits
• Tree-sitter Compatibility: Added compatibility checks and better error messages for tree-sitter version issues
• Import Errors: Improved error handling for missing dependencies

Code Quality

• Indentation Errors: Fixed indentation issues in rate limiting code
• Type Hints: Normalized type annotation formatting across codebase
• Unused Imports: Cleaned up unused imports and whitespace

Changed

Infrastructure

• Dependencies: Updated requirements.txt with explicit tree-sitter>=0.24.0 requirement
• Evaluation Module: Reorganized evaluation module structure
  • Renamed evaluate_all.py → modernize_and_evaluate.py for clarity
  • Renamed evaluate_l2m.py → l2m_helpers.py for better naming
  • Improved module exports and documentation

Documentation

• Package Documentation: Enhanced evals/__init__.py to document both CodeBLEU and Harbor modules
• Harbor Module: Added proper exports and error handling for optional dependencies
• Code Comments: Improved code comments and documentation throughout

Developer Experience

• Output Formatting: Changed free-tier rate limit messages to use io.tool_output instead of print for consistent formatting
• Git Configuration: Better handling of local vs global Git configuration

Removed

• CHANGELOG.md: Removed as it was not being maintained or referenced
• Release Notes Argument: Removed --show-release-notes CLI argument and related code
• Unused Files: Cleaned up unused evaluation and helper files

Migration Notes

For Users

• No breaking changes - this is a minor version bump with new features
• Free-tier OpenRouter users will experience faster failure times (30s vs 3min) when rate limited
• Enhanced error messages will guide you to switch models or add API keys

For Developers

• If using CodeBLEU evaluation, ensure tree-sitter>=0.24.0 is installed
• Update imports if using evals.codebleu.evaluate_all → use modernize_and_evaluate instead
• If using evals.codebleu.evaluate_l2m → use l2m_helpers instead

Full Changelog: v0.2.1...v0.3.0

ATLAS-v0.2.1

Changes

• Fixed invalid exception handling syntax in repo.py
• Corrected inconsistent styling in TUI
• Cleaned up benchmark
• Added model display name conversion

Full Changelog: v0.2.1...v0.2.1

ATLAS-v0.2.0

Added

• PyPI publishing workflow for automated releases
• PyPI installation support (pip install l2m)

Changed

• Updated installation documentation in README

ATLAS-v0.1.0

Major Changes

Architecture Simplification

• Single-agent architecture - Simplified from complex multi-agent system
• Better maintainability - Easier to understand and modify
• Improved performance - Reduced overhead from agent coordination
• Removed legacy components - Cleaned up old multi-agent code

TUI Redesign

• Minimalist design - Clean, professional terminal interface
• Brand color integration - Added #278ef5 throughout UI elements
• Improved UX - Helpful command examples on startup
• Status indicators - Brand-colored waiting spinner and indicators

Bug Fixes

• Fixed exception handling TypeError
• Fixed LiteLLM exception catching issues
• Improved error handling throughout

Migration Notes

This release represents a major architectural shift. The multi-agent system has been replaced with a streamlined single-agent approach.