LLM Workflow Tools with Mise

This repository contains configuration and templates for managing AI-powered code analysis workflows using mise (a development environment management tool) and llm (a CLI tool for interacting with large language models).

Overview

This project provides a streamlined workflow for analyzing code repositories using large language models. It uses:

• mise for environment and task management
• llm for interacting with LLMs (particularly Claude models from Anthropic)
• repomix for bundling repository content for LLM analysis

The project sets up structured tasks to analyze code and generate various types of documentation and insights:

• Code reviews
• GitHub issues based on code analysis
• Prompt engineering for code issues
• Test coverage analysis
• README documentation

Prerequisites

• mise for task and environment management
• llm for interacting with language models
• repomix for bundling repository content
• An Anthropic API key for Claude access

Installation

1. Clone this repository
2. Ensure mise is installed on your system
3. Copy the sample configuration:

   cp config.toml.sample config.toml

4. Edit config.toml to match your environment:
   • Adjust Python version if needed
   • Set your templates directory path in MISE_TEMPLATES_DIR
   • For Linux, change pbcopy to xclip -selection clipboard or wl-copy
   • For Windows, change pbcopy to clip.exe
5. Set up your Anthropic API key as an environment variable:

   export ANTHROPIC_API_KEY=your_api_key_here

Configuration

This project uses two configuration files:

config.toml

Contains task definitions and environment settings for mise:

• Sets up Node.js and Python environments
• Defines tasks for setting up Claude, bundling repos, and generating various analysis outputs
• Configures paths and dependencies between tasks
• Uses environment variables to avoid hardcoded paths

mise.toml

Simple configuration that ensures the latest Python version is used.

Templates

The repository includes several prompt templates for different analysis tasks:

Template	Purpose
code-review-gen.md	Generate thorough code reviews
github-issues-gen.md	Create GitHub-compatible issue descriptions
issue-prompts-gen.md	Create effective prompts for addressing code issues
missing-tests-gen.md	Identify test coverage gaps
readme-gen.md	Generate project documentation

Usage

The repository provides several pre-configured tasks:

Setup

# Set up the Claude integration
mise run llm:setup_claude

# Register all templates with llm
mise run llm:register_templates

Analysis Workflows

# Generate a code review
mise run llm:generate_code_review

# Generate GitHub issues
mise run llm:generate_github_issues

# Generate issue prompts
mise run llm:generate_issue_prompts

# Generate missing tests report
mise run llm:generate_missing_tests

# Generate a README
mise run llm:generate_readme

Intermediate Tasks

# Create repository bundle (if not exists)
mise run llm:clean_bundles

# Copy bundle to clipboard for external use
mise run llm:copy_buffer_bundle

Workflow

1. The workflow typically starts by using repomix to bundle your repository content into a single file (repomix-output.txt)
2. This bundled content is then passed to the appropriate llm template using Claude
3. The analysis output is saved to a corresponding markdown file (e.g., code-review.md, github-issues.md)

Output Files

Each analysis task generates a corresponding output file:

• code-review.md: Detailed code review
• github-issues.md: Potential issues in GitHub format
• issue-prompts.md: Prompts for addressing issues
• missing-tests.md: Test coverage analysis
• README.md: Project documentation

Extending

To add new templates:

1. Create a new markdown file in the templates/ directory
2. Add a new task in config.toml to register and use the template

License

Not specified in the provided code.

Contributing

No contribution guidelines are specified in the provided code.