RA.Aid/CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.14.8] - 2025-02-25

Changed

• Improved programmer.py tool prompts for better clarity on related files visibility
• Enhanced programmer tool to remind users to call emit_related_files on any new files created
• Updated README.md to use media queries for showing different logos based on color scheme preference

[0.14.7] - 2025-02-25

Added

• Windows compatibility improvements
  • Add error handling for Windows-specific modules
  • Add Windows-specific tests for compatibility

Changed

• Improve cross-platform support in interactive.py
• WebUI improvements
  • Improve message display
  • Add syntax highlighting
  • Add animations
• Expert tool prompt improvements

Fixed

• WebUI improvements
  • Fix WebSocket communication
• Interactive command handling improvements
  • Fix interactive history capture
  • Fix command capture bugs
  • Multiple fixes for interactive command execution on both Linux and Windows
  • Enhance error handling for interactive processes

[0.14.6] - 2025-02-25

Added

• Added --no-git flag to aider commands to prevent git operations

Changed

• Updated aider-chat dependency from 0.75 to 0.75.1
• Improved prompts for better tool effectiveness
• Enhanced emit_key_snippet documentation to focus on upcoming work relevance

[0.14.5] - 2025-02-24

Changed

• Optimized prompts

[0.14.4] - 2025-02-24

Changed

• Updated aider-chat dependency from 0.74.2 to 0.75
• Improved tool calling performance by minimizing tool return values
• Replaced emit_key_snippets with emit_key_snippet for simpler code snippet management
• Simplified return values for multiple tools to improve tool calling accuracy
• Updated tool prompts to remove unnecessary context cleanup references
• Reorganized order of tools in read-only tools list

Fixed

• Fixed tests to align with updated tool return values
• Updated test assertions to match new simplified tool outputs

[0.14.3] - 2025-02-24

Added

• Added support for Claude 3.7 Sonnet model
• Added version display in startup configuration panel

Changed

• Updated language library dependencies (langgraph, langchain-core, langchain, langchain-openai, langchain-google-genai)
• Changed default Anthropic model from Claude 3.5 Sonnet to Claude 3.7 Sonnet

Fixed

• Fixed f-string syntax error in write_file.py
• Fixed bug where model selection on Anthropic was always using default instead of respecting user selection
• Fixed Anthropic key error message to reference the correct variable
• Added test for user-specified Anthropic model selection

[0.14.2] - 2025-02-19

Added

• Added automatic fallback mechanism to alternative LLM models on consecutive failures
• Added FallbackHandler class to manage tool failures and fallback logic
• Added console notification for tool fallback activation
• Added detailed fallback configuration options in command line arguments
• Added validation for required environment variables for LLM providers

Changed

• Enhanced CiaynAgent to handle chat history and improve context management
• Improved error handling and logging in fallback mechanism
• Streamlined fallback model selection and invocation process
• Refactored agent stream handling for better clarity
• Reduced maximum tool failures from 3 to 2

Fixed

• Fixed tool execution error handling and retry logic
• Enhanced error resilience and user experience with fallback handler
• Improved error message formatting and logging
• Updated error handling to include base message for better debugging

[0.14.1] - 2025-02-13

Added

• Added expected_runtime_seconds parameter for shell commands with graceful process shutdown
• Added config printing at startup (#88)

Changed

• Enforce byte limit in interactive commands
• Normalize/dedupe related file paths
• Relax aider version requirement for SWE-bench compatibility
• Upgrade langchain/langgraph dependencies

Fixed

• Fixed aider flags (#89)
• Fixed write_file_tool references

[0.14.0] - 2025-02-12

Added

• Status panel showing tool/LLM status and outputs
• Automatic detection of OpenAI expert models
• Timeouts on LLM clients

Changed

• Improved interactive TTY process capture and history handling
• Upgraded langgraph dependencies
• Improved prompts and work logging
• Refined token/bytes ratio handling
• Support default temperature on per-model basis
• Reduced tool count for more reliable tool calling
• Updated logo and branding assets
• Set environment variables to disable common interactive modes

Fixed

• Various test fixes
• Bug fixes for completion message handling and file content operations
• Interactive command input improvements
• Use reasoning_effort=high for OpenAI expert models
• Do not default to o1 model (#82)
• Make current working directory and date available to more agents

[0.13.2] - 2025-02-02

• Fix temperature parameter error for expert tool.

[0.13.1] - 2025-01-31

Added

• WebUI (#61)
• Support o3-mini

Changed

• Convert list input to string to handle create-react-agent tool calls correctly (#66)
• Add commands for code checking and fixing using ruff (#63)

Fixed

• Fix token estimation
• Fix tests
• Prevent duplicate files (#64)
• Ensure default temperature is set correctly for different providers
• Do not incorrectly give temp parameter to expert model
• Correcting URLs that were referencing ai-christianson/ra-aid - should be ai-christianson/RA.Aid (#69)

Improved

• Integrate litellm to retrieve model token limits for better flexibility (#51)
• Handle user defined test cmd (#59)
• Run tests during Github CICD (#58)
• Refactor models_tokens to be models_params so we can track multiple parameters on a per-model basis.

[0.13.0] - 2025-01-22

Added

• Added Deepseek Provider Support and Custom Deepseek Reasoner Chat Model (#50)
• Added Aider config File Argument Support (#43)
• Added configurable --recursion-limit argument (#46)
• Set Default Max Token Limit with Provider/Model Dictionary (#45)

Changed

• Updated aider-chat version from 0.69.1 to 0.72.1 (#47)

Fixed

• Fixed Issue 42 related to Levenshtein (#44)

Improved

• Various prompt improvements
• Better handling of 429 errors on openrouter
• Improved project info handling and token usage optimization
• Extracted tool reflection functionality
• Improved work log handling
• Added tests for CiaynAgent._does_not_conform_to_pattern

[0.12.1] - 2025-01-08

• Fix bug where directories are added as related files.

[0.12.0] - 2025-01-04

Added

• Google Gemini AI provider support
• Dependency check functionality in ra_aid/dependencies.py
• Test coverage reporting to pytest commands

Changed

• Updated minimum Python requirement to 3.9
• Updated OpenAI model defaults
• Modified test files to support new Gemini provider
• Updated SWE-bench dataset generation script with UV package management

Fixed

• Date-based assertions in directory listing tests

[0.11.3] - 2024-12-30

• MacOS fixes.

[0.11.2] - 2024-12-30

• Fix SyntaxError: f-string expression part cannot include a backslash.

[0.11.1] - 2024-12-29

• Improve prompts.
• Fix issue #24.

[0.11.0] - 2024-12-28

• Add CiaynAgent to support models that do not have, or are not good at, agentic function calling.
• Improve env var validation.
• Add --temperature CLI parameter.

[0.10.3] - 2024-12-27

• Fix logging on interrupt.
• Fix web research prompt.
• Simplify planning stage by executing tasks directly.
• Make research notes available to more agents/tools.

[0.10.2] - 2024-12-26

• Add logging.
• Fix bug where anthropic is used in chat mode even if another provider is specified.

[0.10.0] - 2024-12-24

• Added new web research agent.

[0.9.1] - 2024-12-23

• Fix ask human multiline continuation.

[0.9.0] - 2024-12-23

• Improve agent interruption UX by allowing user to specify feedback or exit the program entirely.
• Do not put file ID in file paths when reading for expert context.
• Agents log work internally, improving context information.
• Clear task list when plan is completed.

[0.8.2] - 2024-12-23

• Optimize first prompt in chat mode to avoid unnecessary LLM call.

[0.8.1] - 2024-12-22

• Improved prompts.

[0.8.0] - 2024-12-22

• Chat mode.
• Allow agents to be interrupted.

[0.7.1] - 2024-12-20

• Fix model parameters.

[0.7.0] - 2024-12-20

• Make delete_tasks tool available to planning agent.
• Get rid of implementation args as they are not used.
• Improve ripgrep tool status output.
• Added ask_human tool to allow human operator to answer questions asked by the agent.
• Handle keyboard interrupt (ctrl-c.)
• Disable PAGERs for shell commands so agent can work autonomously.
• Reduce model temperatures to 0.
• Update dependencies.

[0.6.4] - 2024-12-19

• Added monorepo_detected, existing_project_detected, and ui_detected tools so the agent can take specific actions.
• Prompt improvements for real-world projects.
• Fix env var fallback when base key is given, expert and base provider are different, and expert key is missing.

[0.6.3] - 2024-12-18

• Fix one shot completion signaling.
• Clean up error outputs.
• Update prompt for better performance on large/monorepo projects.
• Update programmer prompt so we don't use it to delete files.

[0.6.2] - 2024-12-18

• Allow shell commands to be run in read-only mode.
• When asking for shell command approval, allow cowboy mode to be enabled.
• Update prompt to suggest commands be run in non-interactive mode if possible, e.g. using --no-pager git flag.
• Show tool errors in a panel.

[0.6.1] - 2024-12-17

Added

• When key snippets are emitted, snippet files are auto added to related files.
• Add base task to research subtask prompt.
• Adjust research prompt to make sure related files are related to the base task, not just the research subtask.
• Track tasks by ID and allow them to be deleted.
• Make one_shot_completed tool available to research agent.
• Make sure file modification tools are not available when research only flag is used.
• Temporarily disable write file/str replace as they do not work as well as just using the programmer tool.

[0.6.0] - 2024-12-17

Added

• New file_str_replace tool for performing exact string replacements in files with unique match validation
• New write_file_tool for writing content to files with rich output formatting and comprehensive error handling