# RuleForge — Generate AI Assistant Rules from Repo

> RuleForge is a CLI that scans a repo and generates rule files (CLAUDE.md, .cursorrules, Copilot instructions) so agents follow consistent boundaries.

## Install

Save the content below to `.claude/skills/` or append to your `CLAUDE.md`:

## Quick Use

1. Install / run:
   ```bash
   pip install ruleforge
   ```
2. Start / smoke test:
   ```bash
   ruleforge scan .
   ```
3. Verify:
   - Run `ruleforge preview .` and confirm it produces a non-empty rule draft that matches your repo structure and toolchain.

## Intro

RuleForge is a CLI that scans a repo and generates rule files (CLAUDE.md, .cursorrules, Copilot instructions) so agents follow consistent boundaries.

- **Best for:** Teams who want repo-specific agent rules that stay fresh as the codebase evolves
- **Works with:** Any repo with a detectable structure; integrates well with CI via `ruleforge audit --min-score ...`
- **Setup time:** 15 minutes

## Practical Notes

- Setup time ~15 minutes (install + scan + preview)
- Quantitative gate from README: `ruleforge audit . --min-score 80` can fail CI when rules are too thin
- GitHub stars + forks (verified): see Source & Thanks

Rule files only work when they reflect reality. The best pattern is to regenerate rules on every meaningful repo change (new framework, new folder layout, new CI pipeline), then review the diff like code.

Suggested routine:

1. Run `ruleforge generate .` (without overwrite) and review the output.
2. Add `ruleforge audit . --min-score 80` to CI to prevent regressions in guidance quality.
3. Keep “forbidden actions” and “review-required operations” explicit in the rule text so agents don’t guess.

### FAQ

**Q: Will this overwrite my files?**
A: Only if you pass `--overwrite`. Use preview/audit first to validate output.

**Q: What formats does it generate?**
A: The README lists CLAUDE.md, .cursorrules, and copilot instructions as primary outputs.

**Q: How do I keep rules up to date?**
A: Regenerate on changes and review diffs; add CI audits so guidance doesn’t drift.

## Source & Thanks

> Source: https://github.com/he-yufeng/RuleForge
> License: MIT
> GitHub stars: 53 · forks: 13

---

<!-- ZH -->

## 快速使用

1. 安装 / 运行：
   ```bash
   pip install ruleforge
   ```
2. 启动 / 冒烟测试：
   ```bash
   ruleforge scan .
   ```
3. 验证：
   - 运行 `ruleforge preview .`；确认生成的规则草稿不为空，并且能正确反映仓库结构与工具链。

## 简介

RuleForge 是一个规则生成 CLI：扫描仓库结构与工具链，自动生成 CLAUDE.md、.cursorrules 与 Copilot 指令文件，让 Agent 在一致的边界与安全约束下工作。

- **适合谁：** 希望规则随代码库演进而自动更新的团队（而不是手工维护一堆文件）
- **可搭配：** 适用于大多数有清晰结构的仓库；可用 `ruleforge audit --min-score ...` 接入 CI
- **准备时间：** 15 分钟

## 实战建议

- 准备时间约 15 分钟（安装 + scan + preview）
- README 给出可量化闸门：`ruleforge audit . --min-score 80` 可在规则过薄时让 CI 失败
- GitHub stars / forks（已核验）：见「来源与感谢」

规则文件只有在“贴近现实”时才有效。最佳模式是：每次仓库出现关键变化（新框架、新目录、新 CI）就重新生成一次规则，并像代码一样审查 diff。

建议例行流程：

1. 先跑 `ruleforge generate .`（不覆盖），审阅输出内容。
2. 把 `ruleforge audit . --min-score 80` 加进 CI，防止指导质量回退。
3. 在规则里明确“禁止动作”和“必须人工复核的操作”，避免 Agent 靠猜。

### FAQ

**会直接覆盖文件吗？**
答：只有加 `--overwrite` 才会覆盖。建议先 preview/audit 验证输出。

**主要生成哪些格式？**
答：README 列出 CLAUDE.md、.cursorrules 与 Copilot 指令作为主要输出。

**怎么保持规则常新？**
答：随变更重新生成并审 diff；再用 CI audit 防止指导漂移。

## 来源与感谢

> Source: https://github.com/he-yufeng/RuleForge
> License: MIT
> GitHub stars: 53 · forks: 13


---
Source: https://tokrepo.com/en/workflows/ruleforge-generate-ai-assistant-rules-from-repo
Author: Prompt Lab