# code2prompt — Turn Repos into LLM-Ready Prompts

> code2prompt ingests a codebase and outputs structured, token-aware prompts, helping you move from ad-hoc copy/paste to repeatable context engineering.

## Install

Copy the content below into your project:

## Quick Use

1. Install (Cargo):
   ```bash
   cargo install code2prompt
   ```
2. Generate a prompt from the current repo:
   ```bash
   code2prompt .
   ```
3. Verify:
   - Save output to a file once and confirm it includes the files you expected (and respects `.gitignore`).

## Intro

code2prompt ingests a codebase and outputs structured, token-aware prompts, helping you move from ad-hoc copy/paste to repeatable context engineering.

- **Best for:** developers building agent workflows who need consistent, sharable repo context
- **Works with:** any Git repo; outputs text you can paste into ChatGPT/Claude/Cursor/Codex
- **Setup time:** 5–15 minutes

## Practical Notes

- Quant: run on 2 repos and compare output size (chars/tokens) before/after adding ignore rules.
- Quant: keep a baseline prompt snapshot and diff it weekly to spot context drift.

## Pattern: context as an artifact

Treat the generated prompt like a build artifact:

- Check it into a scratch folder (or attach to PRs) when you need reproducibility.
- Add ignore rules aggressively to keep noise down.
- Prefer smaller, task-scoped prompts over one giant dump.

## Agent workflow tip

Use `code2prompt` as a pre-step before:
- running an agent to refactor,
- generating docs,
- or doing security review of changes.

If you record the prompt file, you can replay the same context across models and compare results.

### FAQ

**Q: Is this only a CLI?**
A: No. The repo describes a wider ecosystem, but the CLI alone is already useful for most workflows.

**Q: How do I reduce prompt size?**
A: Add ignores and scope to a subdirectory; prioritize only the files needed for the task.

**Q: Can I use it for reviews?**
A: Yes—generate a snapshot prompt and attach it to a PR so reviewers/agents share the same context.

## Source & Thanks

> Source: https://github.com/mufeedvh/code2prompt
> License: MIT
> GitHub stars: 7,342 · forks: 422

---

<!-- ZH -->

## 快速使用

1. 安装（Cargo）：
   ```bash
   cargo install code2prompt
   ```
2. 从当前仓库生成 prompt：
   ```bash
   code2prompt .
   ```
3. 验证：
   - 用一次 `--output-file` 保存输出，确认包含期望文件并遵守 `.gitignore`。

## 简介

code2prompt 会遍历并结构化输出你的代码仓库，把内容整理成更适合 LLM 的提示词上下文；相比手工复制粘贴，它更可复用、可对比，也更适合做系统化的 Context Engineering 流程。

- **适合谁：** 在做 agent 工作流、需要稳定可分享“仓库上下文”的开发者
- **可搭配：** 任意 Git 仓库；输出可直接粘贴到 ChatGPT/Claude/Cursor/Codex
- **准备时间：** 5–15 分钟

## 实战建议

- 量化建议：在 2 个仓库上运行，对比加入 ignore 规则前后的输出体积（字符/Token）。
- 量化建议：保留一份基线输出快照，每周 diff 一次，观察上下文漂移。

## 常用打法：把上下文当作可复用产物

把生成的 prompt 当成构建产物来管理：

- 需要可复现时，把输出落盘（或随 PR 附件）便于回放。
- 主动维护 ignore 规则，减少噪音。
- 更推荐“按任务切片”的小 prompt，而不是一次性大 dump。

## Agent 工作流提示

把 `code2prompt` 当作很多任务的前置步骤：重构、生成文档、安全审查等。

保留 prompt 文件后，你可以用同一份上下文在不同模型间重放，对比效果与成本。

### FAQ

**它只是 CLI 吗？**
答：不止。仓库还提到更完整的生态，但单独 CLI 就能覆盖大多数场景。

**怎么降低输出体积？**
答：加 ignore 并限制到子目录；只保留任务必需文件。

**能用于评审吗？**
答：可以。生成一份快照并随 PR 附上，让评审/agent 共享同一上下文。

## 来源与感谢

> Source: https://github.com/mufeedvh/code2prompt
> License: MIT
> GitHub stars: 7,342 · forks: 422


---
Source: https://tokrepo.com/en/workflows/code2prompt-turn-repos-into-llm-ready-prompts
Author: Script Depot