Learn to customize your AI by editing text files.
You edit a file. The AI reads it. That's all that happens.
No build step — change a file, restart the session, done. Every instruction is git-diffable, inspectable without running anything, and works across AI tools (Claude Code, Gemini CLI, Codex, OpenCode). There is no framework to learn, no API to call, no schema to compile. The files are the system.
When you use an AI coding tool like Claude Code, every conversation starts from scratch — the AI doesn't know your name, your preferences, or what you're working on. You repeat yourself every session.
forge-learn fixes that. It's a collection of plain text files that your AI reads automatically. You tell it who you are once, and it remembers. You add skills — small instruction files — and it learns new abilities.
Everything here is a text file you can open, read, and edit. No magic, no hidden configuration, no compilation. If you can edit a document, you can customize your AI.
forge-learn is built for Claude Code — skills are auto-discovered and slash commands (/Tour, /Explain, etc.) work out of the box.
The identity files (rules/Identity.md, Goals.md) and CLAUDE.md are plain markdown, so they're useful with other AI coding tools too — but skill discovery varies:
| Tool | Reads CLAUDE.md | Skills as /commands | Install |
|---|---|---|---|
| Claude Code | Yes | Yes (full plugin support) | curl -fsSL https://claude.ai/install.sh | bash |
| OpenCode | Yes (fallback) | Partial (.claude/skills/) |
curl -fsSL https://opencode.ai/install | bash |
| Codex CLI | No (uses AGENTS.md) | No (Uses $Skill "command" instead) | See Codex docs |
| Gemini CLI | No (uses GEMINI.md) | No | See Gemini CLI docs |
| GitHub Copilot | No (uses .github/copilot-instructions.md) | No | Via IDE |
For tools that don't read CLAUDE.md, you can copy the content into their instruction file format (e.g., AGENTS.md for Codex, GEMINI.md for Gemini CLI).
rules/ <- Who you are and what you care about
skills/ <- What your AI can do (commands you can run)
agents/ <- AI personas (e.g., CodeHelper)
Steering is your identity. Open rules/Identity.md, change the name to yours, and save. Next session, your AI greets you by name.
Skills are actions. Type /Tour and your AI walks you through everything. Type /Explain and it breaks down any file or error in plain language.
Agents are personas. The starter agent CodeHelper explains broken code in plain language. You can edit it or create your own.
Mac / Linux:
curl -fsSL https://claude.ai/install.sh | bashWindows (PowerShell):
irm https://claude.ai/install.ps1 | iexMac (Homebrew):
brew install --cask claude-codeWindows (WinGet):
winget install Anthropic.ClaudeCodeOpen a terminal (Mac: Terminal app, Windows: PowerShell or Git Bash) and paste:
git clone https://github.com/N4M3Z/forge-learn.git
cd forge-learnThis creates a forge-learn directory on your computer and moves into it.
Don't have Git? Mac: type
git --versionin Terminal and macOS will prompt you to install it. Windows: download from git-scm.com.
Let your AI do it (recommended): Start Claude Code and ask it to follow INSTALL.md — it checks prerequisites, personalizes your identity files, deploys skills, and runs /Tour.
claude
# then tell it: "Set up forge-learn for me. Follow INSTALL.md."Or do it yourself:
- Open
rules/Identity.mdin any text editor — changeYour Nameto your actual name and save - Open
rules/Goals.md— replace the example goals with yours - Deploy agents and skills:
- Mac/Linux/WSL/Git Bash: run
make install - Windows PowerShell fallback: run the Windows block in INSTALL.md (
cargo build --release+install-*.exe)
- Mac/Linux/WSL/Git Bash: run
- Start Claude Code:
claude - Type
/Tour
That's it. You're set up.
By default, skills only work inside the forge-learn directory. To use /Explain, /GitHelp, and other skills in any project:
make install SCOPE=userThis installs skills globally so they're available whenever you run claude, regardless of which project you're in.
| Skill | What it does |
|---|---|
/Tour |
Walks you through your setup and available skills |
/Progress |
Shows your current level and suggests what to try next |
/Explain |
Explains any file, error, or concept in plain language |
/FixIt |
Diagnoses problems and proposes fixes |
/GitHelp |
Translates plain English into git commands |
/Kickstart |
Turns "I want to build X" into a plan with starter files |
/Summarize |
Extracts key points into a structured summary |
forge-learn has a 7-level progression from your first skill modification to ecosystem contributor. Type /Progress anytime to see where you are and what to try next.
| Level | Name | What you learn |
|---|---|---|
| 1 | Discover | Run a skill, read its source, change one line |
| 2 | Personalize | Edit identity files, see the AI adapt to you |
| 3 | Navigate | Work with files and save your progress |
| 4 | Build | Create projects from ideas |
| 5 | Author | Write custom rules, skills, and agents |
| 6 | Connect | Expand with optional modules |
| 7 | Forge | Contribute improvements to the ecosystem |
See rules/Levels.md for the full roadmap with progress tracking.
When you start a session, your AI tool reads these files before you say anything:
CLAUDE.md— ground rules and conventions (read automatically by Claude Code)rules/Identity.md— your name, language, experience level (frontmatter parsed at session start)rules/Goals.md— what you're working toward- Any other
.mdfiles inrules/— custom rules you add
Skills (skills/*/SKILL.md) are registered as commands during plugin discovery. When you invoke a skill (e.g., /Explain in Claude Code, $Explain in Codex), the tool loads that skill's SKILL.md and follows its instructions. No runtime interpretation — the file IS the instruction.
Agents (agents/*.md) are specialist personas deployed by make install into provider-specific directories.
You change a file, the AI's behavior changes. There is no hidden layer between the file and the behavior.
The rules/ files are yours to edit with personal information (your name, goals, preferences). If you plan to push this repository to a public GitHub account, be aware that anything you write in these files will be visible. Consider keeping your fork private, or review your rules files before pushing.
Once you're comfortable, expand your setup with optional modules:
| Module | What it adds |
|---|---|
| forge-text | 12 text processing skills — translate, simplify, grammar fix, expand, condense, and more |
| forge-council | Multi-specialist code review — a panel of experts debates your changes |
| forge-avatar | Deep identity — digital avatar, beliefs, strategies, communication preferences |
| forge-core | Developer toolkit — build your own skills, agents, and modules |
Clone a module and ask your AI to help set it up. For automated module management, see forge-user.
- Git — to download and manage this repository
- Claude Code — the AI tool that reads these files
- A text editor — anything works (VS Code, Notepad, TextEdit, Sublime, vim)
- Rust toolchain (
cargo) for first-time local binary builds (lib/) - Windows: Works natively on Windows 10+.
makerecipes are POSIX-style; use WSL/Git Bash formake install, or use the PowerShell fallback in INSTALL.md.