This document outlines the planned features and enhancements for the ClaudeCoder GitHub Action. This living document will be updated as industry trends evolve and user feedback is incorporated.

Description: Allow users to describe a feature in natural language, and have ClaudeCoder implement it end-to-end across multiple files and systems.

Implementation Details:

• Develop advanced planning capabilities to break down feature requirements
• Implement changes across multiple related files
• Generate appropriate tests and documentation
• Create a PR with the complete implementation

Description: Modify how commit messages are created so they are not a single standard message but contextually aware of what has been changed

Implementation Details:

• Evaluate the best architecture to get the commit message together with the code changes, without making a secondary request just for the commit message
• Commit messages obey a .claudecoder stylefile with details on how the message is generated (example: - add emojis at the end; - use semver rules) etc

Description: Add API stats to the final PR message response, with details on the in/out tokens used, the cost and total context size utilized during the operation

Implementation Details:

• Use the API response structure to find information on the API cost and add it to the response message on the PR conversation

Description: ClaudeCoder will automatically select from Haiku or Sonnet based on the complexity of the task to save on API calls

Implementation Details:

• Evaluate the code and user request first on Haiku, asking for a complexity estimate
• Based on some industry standard complexity measurement, the model will either complete the request and answer what the user wants, or return an answer with the complexity score and requesting to ask to a smarter model.
• Ask to respond with some parameter or keyword so its easy to apply a regex and detect if a new model request is needed

Description: Allow repository maintainers to configure code style preferences that ClaudeCoder will respect when suggesting changes. This will ensure that AI-generated code follows team conventions.

Implementation Details:

• Support reading from .editorconfig, .prettierrc, and similar files
• Allow providing style guidelines in the action configuration
• Enhance Claude prompts to respect these guidelines
• Add examples of style configuration to documentation

Description: Enable ClaudeCoder to add inline comments on specific lines of code in the PR, rather than only modifying files or adding general comments.

Implementation Details:

• Use GitHub's Pull Request Review API to add line-specific comments
• Include reasoning for suggested changes in the comments
• Add support for code suggestions using GitHub's suggested changes format
• Allow users to accept suggestions with a single click

Description: Improve ClaudeCoder's understanding of the repository's context by analyzing project structure, dependencies, and existing patterns.

Implementation Details:

• Extend repository scanning to analyze package.json, requirements.txt, and similar files
• Identify common patterns and coding conventions in the existing codebase
• Use this context to generate more relevant and consistent code suggestions
• Implement memory of previous interactions within the same PR

Description: Allow users to specify which files or directories should be included or excluded from ClaudeCoder's analysis.

Implementation Details:

• Add configuration options for file inclusion/exclusion patterns
• Support .claudecodeignore file (similar to .gitignore) for repository-specific settings
• Optimize performance by only analyzing relevant files

Description: Can read and write the context of the pull requests in the issues

Implementation Details:

• TBD

Description: Enable ClaudeCoder to automatically generate or update tests for code changes, and validate that tests pass before suggesting changes.

Implementation Details:

• Analyze existing test patterns in the repository
• Generate appropriate unit, integration, or end-to-end tests for suggested changes
• Integrate with CI workflows to run tests on suggested changes
• Only suggest changes that pass tests

Description: Scan code for potential security vulnerabilities and suggest secure alternatives.

Implementation Details:

• Integrate with security scanning tools or APIs
• Train Claude to recognize common security patterns and anti-patterns
• Provide context-aware security recommendations
• Link to relevant security best practices in comments

Description: Help teams refactor large sections of code or entire systems by understanding the existing architecture and gradually transforming it.

Implementation Details:

• Analyze codebase interdependencies and architecture
• Plan complex refactoring operations that maintain functionality
• Execute refactoring in small, reviewable chunks
• Maintain comprehensive test coverage throughout the process

Description: Support multiple AI models beyond Claude, allowing users to choose or ensemble different models for different types of tasks.

Implementation Details:

• Add support for other LLM providers like OpenAI, Mistral, and Anthropic
• Implement adapter patterns to normalize interactions across providers
• Develop benchmarks to recommend optimal models for specific tasks
• Allow combining strengths of different models

Description: Develop a system for ClaudeCoder to learn and adapt to the specific patterns and preferences of each repository over time.

Implementation Details:

• Store anonymized usage data and feedback in a repository-specific way
• Analyze acceptance patterns of previous suggestions
• Adapt prompts and generation parameters based on historical data
• Provide analytics on suggestion acceptance rates and areas for improvement