agent-config

A centralized repository for managing AI agent configurations, skills, commands, and subagents across multiple AI CLI platforms: Claude, Codex, Copilot, and Gemini.

Security notice: The contents of this repository flow directly into each agent's system prompt. Review all files — especially agents/_GLOBAL.md and agent-specific stubs — before use. Anyone with write access to this repo can influence the behavior of all configured agents.

Overview

This project provides a centralized location for storing and managing:

• Agent Configurations - Per-agent instruction files (merged from shared + agent-specific source)
• Reusable Skills - Specialized capabilities that agents can leverage
• Custom Commands - Slash commands available to agents (Markdown source; auto-converted to TOML for Gemini)
• Subagents - Specialized AI agents that primary agents can delegate work to

Agent Capability Matrix

Not all features are supported by every agent. The table below shows what each agent supports and where each resource is installed on disk.

Feature	Claude	Codex	Copilot	Gemini
Config file	~/.claude/CLAUDE.md	~/.codex/AGENTS.md	~/.copilot/copilot-instructions.md	~/.gemini/GEMINI.md
Skills directory	~/.claude/skills/	❌ no standard path	~/.copilot/skills/	❌ no convention
Commands directory	~/.claude/commands/ (.md)	~/.codex/prompts/ (.md)	❌ not supported	~/.gemini/commands/ (.toml)
Subagents directory	~/.claude/agents/ (.md)	❌ not supported	⚠️ P1: .agent.md ext required	~/.gemini/agents/ (.md)
Command format	Markdown	Markdown	—	TOML (auto-converted at build time)

Scripts emit an explicit skip message for any unsupported feature rather than silently skipping.

Directory Structure

agent-config/
├── README.md                    # This file
├── LICENSE                      # Project license
├── Makefile                     # Build and setup targets
├── agents/                      # Agent configurations
├── skills/                      # Reusable skills for agents
├── commands/                    # Custom commands available to agents
├── subagents/                   # Custom agents for delegation
├── prompts/                     # Custom prompts for specific tasks
└── scripts/                     # Setup and installation scripts

Folders

agents/

Contains configuration files for four AI agents: Claude, Codex, Copilot, and Gemini.

Each agent has:

• A shared source file: agents/_GLOBAL.md
• An agent-specific source file: agents/<agent>/_<AGENT_NAME>.md
• A generated merged file: agents/<agent>/<AGENT_NAME>.md (gitignored; symlinked into the agent's home dir)

Generation: make generate-agent-files (runs as part of make all)

See agents/README.md for symlink targets and details.

skills/

Reusable skills and specialized capabilities that agents can access and leverage.

Contents should include:

• Well-defined capabilities agents can perform
• Domain expertise and knowledge bases
• Procedural workflows and processes
• Specialized knowledge resources

commands/

Slash commands available to agents. Source files are Markdown (.md) with YAML frontmatter.

• Claude reads them directly from ~/.claude/commands/
• Codex reads them from ~/.codex/prompts/
• Gemini requires TOML format — files are auto-converted via make generate-gemini-commands to build/gemini-commands/ and symlinked to ~/.gemini/commands/
• Copilot does not support custom commands

subagents/

Configurations for custom AI agents that primary agents can delegate work to.

Contents should include:

• Specialized agent configurations
• Agents designed for specific domains or tasks
• Agent instructions for delegated work
• Subagent capabilities and limitations

scripts/

Setup and configuration scripts for initializing the AI environment.

All scripts use utils.sh for shared helpers. Existing files at symlink destinations are automatically backed up with a .old extension before replacement.

Available scripts:

Script	Purpose
generate-agent-files.sh	Merges _GLOBAL.md + _<AGENT>.md → <AGENT>.md for each agent
generate-gemini-commands.sh	Converts commands/*.md → build/gemini-commands/*.toml (TOML format for Gemini)
symlink-agents.sh	Symlinks per-agent config to the correct filename in each agent's home dir
symlink-skills.sh	Symlinks skills/ to Claude and Copilot (skipped for Codex and Gemini)
symlink-commands.sh	Symlinks commands/ to Claude and Codex; defers Gemini to symlink-gemini-commands.sh
symlink-gemini-commands.sh	Symlinks build/gemini-commands/ → ~/.gemini/commands/
symlink-subagents.sh	Symlinks subagents/ to Claude and Gemini (skipped for Codex; P1 for Copilot)

Setup

git clone https://github.com/coreydaley/agent-config.git
cd agent-config
make all

make all generates merged agent files, converts commands to TOML for Gemini, and creates all symlinks in one step. It is idempotent — safe to run multiple times.

Individual targets

make generate             # Generate all artifacts (agent files + Gemini TOML)
make symlinks             # Create all symlinks
make symlink-agents       # Agent config symlinks only
make symlink-skills       # Skills symlinks only (Claude + Copilot)
make symlink-commands     # Command symlinks only (Claude + Codex + Gemini)
make symlink-subagents    # Subagent symlinks only (Claude + Gemini)
make help                 # Show all available targets

Optional: Codex skills registration

Codex does not support a global skills directory. To register skills from this repo with Codex via ~/.codex/config.toml, run this separately (not part of make all):

make configure-codex-skills

Optional: YOLO mode aliases

Each agent has a flag to skip permission prompts and run fully autonomously. If you want this behavior by default, add these aliases to your shell config (~/.zshrc, ~/.bashrc, etc.):

alias claude='claude --dangerously-skip-permissions'
alias codex='codex --full-auto'
alias copilot='copilot --allow-all'
alias gemini='gemini --yolo'

Warning: These aliases disable all permission prompts. The agent will execute file writes, shell commands, and other actions without asking first. Only use this if you trust the agent configuration in this repo and understand the risks.

Backup behavior

Existing regular files or directories at symlink destinations are renamed to .old before the symlink is created. Existing symlinks are replaced directly.

Usage

After running make all, each agent will load its merged config file automatically. The capability matrix above shows what each agent can access.

Reference the individual README files for details:

• agents/README.md - Agent configuration and symlink targets

Contributing

When adding new resources:

• Place them in the appropriate folder (skills, commands, subagents, or prompts)
• Follow the naming conventions established in each directory
• Update the relevant README files with documentation
• Test that symlinks and configurations work correctly

If you encounter a problem, please open an issue. If you'd like to fix an issue or add new functionality, feel free to fork this repository and submit a pull request.

Disclaimer

Please note: Content in this project is likely generated using AI language models. While efforts have been made to ensure quality and accuracy, AI-generated content can contain errors, outdated information, or unintended biases.

Use at your own risk. Always:

• Review AI-generated content before using it in production
• Test configurations and commands thoroughly
• Verify information against authoritative sources
• Consider the limitations and potential issues of AI-generated code and instructions
• Take responsibility for any issues that may arise from using this project

The creators and maintainers assume no liability for problems caused by following instructions or using resources from this repository.