📘 OFFICIAL DOCUMENTATION & SPECIFICATIONS

RulesForge Documentation

The Universal Single Source of Truth (SSOT) for AI Coding Assistants and Agentic Workflows.

Welcome to the official documentation for RulesForge. This guide covers how RulesForge turns your codebase architecture, build routines, and coding guidelines into standard-compliant, deterministic instruction files for Claude Code, Cursor IDE, GitHub Copilot, Windsurf Cascade, and open agentic frameworks.

🚀 Getting Started ✳ Anthropic Claude Code ⚡ Cursor IDE (.mdc) ⌘ OpenAI & AGENTS.md 🐙 GitHub Copilot 🌊 Windsurf Cascade 🎯 Context Engineering 🔒 Security & Privacy

1. Getting Started: How RulesForge Works

In modern software engineering, developers work across diverse AI tools simultaneously. However, each assistant expects its steering files at different paths and with unique formatting rules. RulesForge eliminates this friction by providing a Single Source of Truth.

Step 1: Declare Your Specs

Enter your project description, tech stack, testing frameworks, coding standards, and safety boundaries in the interactive form.

Step 2: Inspect Live Preview

Review standard-compliant tabs for each target file. RulesForge dynamically compiles the instructions into native syntax.

Step 3: Export or Bundle

Copy individual files to your clipboard or click Download Rules Bundle (.ZIP) to extract directly into your project root.

2. Supported AI Platforms & File Specifications

Anthropic Claude Code (`CLAUDE.md`)

Anthropic recommends keeping CLAUDE.md concise (under 200 lines) so that it functions as an efficient "router" rather than an exhaustive codebase dump. Oversized system prompts degrade attention and inflate per-turn latency. RulesForge structures CLAUDE.md with:

High-level architecture outline and repository layout.
Explicit command instructions (e.g. npm run test, cargo check).
Compact style conventions and defensive prohibitions.
Target Location: Project root directory (/CLAUDE.md).

Cursor IDE (`.cursor/rules/*.mdc`)

Cursor uses modern .mdc files equipped with YAML frontmatter to support modular, path-scoped rules. This ensures an assistant only activates frontend guidelines when editing frontend files:

---
description: React & TypeScript coding standards
globs: src/components/**/*.{tsx,ts}
alwaysApply: false
---
- Use functional components with strict TypeScript prop types.
- Never use direct inline styles; prioritize design tokens.

Target Location: .cursor/rules/rules.mdc or scoped files under .cursor/rules/*.mdc.

OpenAI Codex & AAIF (`AGENTS.md`)

AGENTS.md is the open community standard promoted by the Agentic AI Foundation. It provides multi-agent execution guidelines, tool capabilities, and verification steps that autonomous agents follow before executing tasks.

Declares environment pre-requisites and CLI tooling.
Defines subagent hand-off boundaries and output contracts.
Target Location: Project root directory (/AGENTS.md).

GitHub Copilot (`.github/copilot-instructions.md`)

GitHub Copilot automatically consumes custom repository instructions located in the .github directory. RulesForge optimizes this file to focus on code quality, testing conventions, and prohibited operations.

Enforces test verification before committing changes.
Target Location: .github/copilot-instructions.md.

Windsurf Cascade (`.windsurfrules`)

Windsurf's Cascade agent utilizes .windsurfrules to establish project constraints, multi-step confirmation rules, and safe command execution boundaries.

Prevents accidental destructive terminal commands.
Target Location: Project root directory (/.windsurfrules).

3. Context Engineering, Token Economy & Budgeting

Why do explicit, version-controlled instruction files and token budgeting matter?

Prevent Context Dilution: Injecting oversized system prompts causes the model to ignore critical instructions in the middle of long files. RulesForge provides a real-time Token Calculator with configurable budgets (1,000, 2,000, or 4,000 tokens) to ensure your rulebooks stay within optimal performance boundaries.
Dedicated TOKEN_BUDGET.md Report: To keep your actual instruction files (CLAUDE.md, AGENTS.md, .mdc) 100% clean and free of confusing comments, RulesForge bundles a dedicated TOKEN_BUDGET.md audit report inside every downloaded ZIP archive.
Eliminate Hallucinations: Large Language Models are probabilistic. Without explicit rules, assistants guess outdated dependencies or invent non-existent test runners.
Enforce Verification Workflows: Instructing the assistant to run npm test, pytest, or go test ./... guarantees code is validated before you review it.
Defensive Guardrails: Protect critical configuration files (e.g., .env, database migration files, protected Git branches) from unauthorized modification.

4. Local-First Security & Client Privacy

At RulesForge, we believe developer tooling must respect source code confidentiality:

100% In-Browser Execution: All parsing, generation, live previewing, and ZIP bundling occur strictly inside your web browser using vanilla JavaScript.
Zero Server Logging: Your project ideas, internal command paths, and architectural rules are never uploaded to remote databases or used for model training.
Free & Open Access: RulesForge requires no account creation, no API keys, and no subscription fees.