# Caliber — Auto-Refresh CLAUDE.md and Rules

> Caliber keeps CLAUDE.md, Cursor rules, AGENTS.md, and Copilot instructions accurate by regenerating them from your current repo state. Bootstrap via npx.

## Install

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

# Caliber — Auto-Refresh CLAUDE.md and Rules

> Caliber keeps CLAUDE.md, Cursor rules, AGENTS.md, and Copilot instructions accurate by regenerating them from your current repo state. Bootstrap via npx.

## Quick Use

1) Install

```bash
npx @rely-ai/caliber bootstrap
```

2) Run

```bash
In your agent session: /setup-caliber
```

3) Verify

```bash
caliber score --compare main
```

---

## Intro

**Best for**: teams whose agent context files drift after refactors and want deterministic, local updates

**Works with**: Node.js >= 20 (repo), npx workflow, repos that use CLAUDE.md / AGENTS.md / Cursor rules

**Setup time**: 9 minutes

### Quant Data

- Deterministic scoring (no LLM/API) shown in README table (repo)
- Requires Node.js >= 20 (repo)

---

## How to Use It Well

Use Caliber as a guardrail after every meaningful refactor: regenerate configs, review diffs, and keep only the minimal instructions that improve outcomes.

### Adoption Checklist

- Start with one real task and keep the scope narrow
- Capture a baseline: time-to-first-success and output quality
- Version your config/skills so teammates stay in sync

### Guardrails

Do not let auto-generated instructions become noise. If a section doesn't change behavior, delete it.

### FAQ

**Q: Does it send code to the cloud?**
A: The README says bootstrap is local; generation uses your own AI provider key. Treat any provider call like normal agent usage.

**Q: What should I run first?**
A: Start with `bootstrap`, then use `/setup-caliber` inside your agent workflow to generate configs.

**Q: How do I measure ROI?**
A: Track config score improvements, number of path hallucinations eliminated, and how often rules stay in sync after merges.

---

## Source & Thanks

> GitHub: https://github.com/caliber-ai-org/ai-setup
> Owner avatar: https://avatars.githubusercontent.com/u/258568152?v=4
> License (SPDX): MIT
> GitHub stars (verified via `api.github.com/repos/caliber-ai-org/ai-setup`): 1,038


---

<!-- ZH -->

# Caliber——自动刷新 Claude/Cursor 配置

> Caliber 会根据仓库现状自动生成并持续刷新 CLAUDE.md、Cursor rules、AGENTS.md 与 Copilot 指令，避免重构后路径过期导致幻觉，并支持本地评分/对比。npx 一步启动。

## 快速使用

1）安装

```bash
npx @rely-ai/caliber bootstrap
```

2）运行

```bash
In your agent session: /setup-caliber
```

3）验证

```bash
caliber score --compare main
```

---

## 简介

**适合谁**：重构后上下文文件经常过期的团队，希望用确定性、本地化的方式自动更新

**适用环境**：仓库要求 Node.js >= 20、npx 工作流、使用 CLAUDE.md / AGENTS.md / Cursor rules 的代码库

**安装耗时**：9 分钟

### 量化信息

- README 给出确定性评分表（不依赖 LLM/API，仓库）
- 要求 Node.js >= 20（仓库）

---

## 用好它的方式

把 Caliber 当成重构后的护栏：重生成配置、review diff，只保留真正能提升产出的最小指令集。

### 推广清单

- 先选一个真实任务，小范围试跑
- 记录基线：首次成功耗时与输出质量
- 配置/技能要版本化，避免团队漂移

### 风险与护栏

别让自动生成的指令变成噪音：如果某段不会改变行为，就删掉。

### FAQ

**Q: 会把代码发到云端吗？**
A: README 表示 bootstrap 为本地流程；生成阶段使用你自己的 AI Provider Key。把它当作正常 agent 调用来做权限与审计。

**Q: 先跑什么？**
A: 先 `bootstrap`，再在 agent 会话里运行 `/setup-caliber` 生成配置。

**Q: 怎么量化收益？**
A: 记录评分提升、路径幻觉减少次数，以及合并/重构后配置保持同步的频率。

---

## 来源与感谢

> GitHub：https://github.com/caliber-ai-org/ai-setup
> Owner avatar：https://avatars.githubusercontent.com/u/258568152?v=4
> 许可证（SPDX）：MIT
> GitHub stars（已通过 `api.github.com/repos/caliber-ai-org/ai-setup` 核验）：1,038


---
Source: https://tokrepo.com/en/workflows/caliber-auto-refresh-claude-md-and-rules
Author: Script Depot