foxhound is a powerful tool designed to assist Large Language Model (LLM) agents in navigating unfamiliar codebases effectively. LLMs can struggle with understanding the structure of a new repository and often lack the ability to consult with knowledgeable team members. foxhound addresses these challenges through two main components:

How foxhound Works

1. wikigen constructs a detailed wiki for the repository, which contains skill files (CLAUDE.md / AGENTS.md) that inform the LLM on how to navigate the codebase effectively.
2. The LLM utilizes this wiki to grasp the repository’s structure, dependencies, and common modification practices.
3. In scenarios where the LLM encounters unfamiliar code or faces potentially risky changes, consult helps identify the appropriate human expert based on recent activity in the relevant code sections.
4. Communication occurs in Slack where the LLM awaits responses to its inquiries, allowing it to incorporate human feedback seamlessly.

This dual functionality efficiently addresses the common issues (80% of the cases) related to code structure comprehension while reserving human expertise for specialized queries (20% of the cases).

Project Structure

The repository is organized as follows:

foxhound/
  wikigen/             → Refer to wikigen/README.md for more information
  consult/             → Refer to consult/README.md for more information
  docs/sprints/        → Sprint planning documents
  go.mod               → Shared Go module

Documentation

Each tool comes with detailed documentation, including full usage instructions, available flags, examples, and skill file details:

• wikigen/README.md — Detailed information on the wiki generation tool.
• consult/README.md — Complete guide for the expert consultation tool.

Sample Skill Files

The sample-skills/ directory includes example skill files that illustrate how foxhound generates training data for LLM agents, enabling them to navigate the generated wiki and effectively consult with human experts:

• sample-skills/CLAUDE.md — Sample skill file tailored for Claude Code.
• sample-skills/AGENTS.md — Sample skill file tailored for Codex.

With foxhound, LLMs gain crucial insights into repository structures while maintaining access to human knowledge, ultimately leading to more efficient and accurate code modifications.