Overview

ai-agent-config is a baseline for the global configuration of AI coding assistants. It keeps three things in one neutral folder tree:

• Global instructions in GLOBAL.md: working style, validation, scope control, commit and output rules that apply to every project.
• Subagents in agents/: role prompts for an explorer, planner, implementer, reviewer and security auditor.
• Skills in skills/: one folder per skill, each with a SKILL.md and optional reference files.

The install scripts symlink this tree into ~/.claude/ and ~/.codex/ under the names each tool expects, and register the skills and agents in Mistral Vibe’s config if Vibe is installed. Because every target is a symlink, an edit in the repository is live immediately.

A short product page lives at ai-agent-config.casoon.de.

Status

The repository is a public snapshot: a sanitized export of a personal configuration that is maintained in a separate private repository and re-exported periodically. Business-specific skills, real machine config and internal identifiers are kept out. Treat it as a baseline to copy from rather than a live feed, and expect occasional breaking changes.

What it does not cover

• Project memory: each project’s own CLAUDE.md or AGENTS.md belongs in that project’s repository.
• Conversational memory: Claude Code’s per-session auto-memory under ~/.claude/projects/ is machine-local and often personal; it never belongs in a shared baseline.
• Zed’s native agent: it has no skills concept and uses its own rules files. Claude Code running inside Zed reads ~/.claude/ and gets everything (see How it works).

How the docs are organised

• Getting started: installation and how the symlinks work.
• Guides: customising the baseline and the pre-commit hook.
• Reference: every subagent and skill, read from the repository when the site is built, and the repository layout.

The showcase puts each subagent’s Markdown source next to the TOML file generated from it for Codex.