Anoma LLM Configuration

This repository represents LLM configurations that all of our projects can merge in to bring the shared values of the engineering department into our codebases. This is important as without these configurations LLM agents may be misaligned, leading to code that may not be maintainable or review that misses key insights.

Currently the repository only supports Claude.

Including these configurations into your repository

To include this repository into your project run the following commands:

git remote add llm https://github.com/anoma/llm.git

git merge llm/<latest-tag> --allow-unrelated-histories

Then tell your LLM Agent to add the following skills to their init file

❯ /init
❯ add the following skills to your CLAUDE.md file
  @.claude/skills/general-conventions/SKILL.md
  @.claude/skills/elixir-conventions/SKILL.md

where elixir can be replaced with the language of the project.

This repository does not offer a file like Claude.md, as it’s expected to provide repository specific information we can not know about beforehand.

Shared LLM Values

For code that we wish to preserve here are some values this repository brings in:

1. Code Principles
   • Minimize code
   • Generalize over special-casing
   • No unnecessary indirection
   • Flatten control flow
2. Live Exploration
   • Inspect process/object state at runtime
   • Test hypotheses interactively
   • Trace actual call paths
3. Example-Driven Development
   • Each example should reveal something new about the system
4. Anti-Patterns (never do)
   • Add features beyond the request
   • Create abstractions for single use
   • Add flexibility or configurability not requested
   • Handle errors that can’t happen
   • Improve adjacent code unprompted
5. Failure Protocol
   • Stop after 2 failed attempts at the same approach
   • Preserve partial work
   • Diagnose what failed and why
   • Change approach or ask for guidance
6. Scope Awareness
   • Before starting: restate what is needed, surface unstated assumptions, flag ambiguity

Different kinds of configurations for different workflows (like LLMs that generate an entire codebase for you) will be added over time, always check out the newest release to see what may benefit you.

Useful skills

/code-review <file>

review a file with a subagent that specializes in reviewing code.

Project setup

This repository offers skills set up in such a way to provide core values to projects written in any language. We do this by providing our general values as a generic skills file. Conventions for specific languages are added in their own skills file.

These skills files do the heavy lifting for any custom agents you may wish to set up for your repository. Meaning that you just have to specify what is special for your project in a custom agent file, e.g. in this repository always run XYZ.