一个用于 System Prompt 的、自文档化的单行元数据标准。
A single-line, self-documenting metadata standard for System Prompts.
在您的 System Prompt 中间(例如「角色声明」之后)添加这样一行注释:
Add this comment line to your System Prompt, after role assignment:
你是一个有用的助手... / You are a helpful assistant...
[//]: # PROMPT-META (your-prompt-name@2025-07-15; by:your-name(github/your-gh-id); repo:github/your-gh-id/your-repo; license:CC-BY-NC-4.0; tools:memory=npmjs/mcp-server-memories-off,obsidian=npmjs/obsidian-mcp-tools,tool-name=source/package-name; models:gemini-2.5-pro,gpt-4o,usable-models)
## 任务要求: / ## Tasks:
PIE (Prompt Identifier Embedded) 是一种旨在让 System Prompt 实现自描述、可移植、可验证的单行元数据规范。
它通过一个简单、内联的 Markdown 注释,将 Prompt 的身份、版本、作者、依赖、许可等关键信息直接“嵌入”到 Prompt 本身,解决了 Prompt 在流转和协作过程中的信息孤岛问题。
PIE (Prompt Identifier Embedded) is a single-line metadata standard designed to make System Prompts self-describing, portable, and verifiable.
It embeds a prompt's critical metadata—identity, version, author, dependencies, license—directly within it using a simple inline Markdown comment, tackling the loss of origin and version data during sharing and collaboration.
随着 Prompt 变得越来越复杂,我们面临着一系列新的挑战:
- 版本混乱: "哪个是最新版?" "这个 Prompt 是为 GPT-3.5 还是 GPT-4 优化的?"
- 隐式依赖: Prompt 可能需要特定的工具 (Tools/Functions) 才能工作,但这些依赖关系没有被明确声明。
- 来源不清: 很难追溯一个优秀的 Prompt 是谁创作的,来自哪个项目,以及它的使用许可。
- 管理困难: 当拥有数十上百个 Prompt 时,无法对其进行有效的分类、搜索和自动化管理。
PIE 通过一个精简且不干扰 LLM 工作的格式,尝试解决以上所有问题。
As prompts become increasingly complex, we face a new set of challenges:
- Version Chaos: "Which is the latest version?" "Was this prompt optimized for GPT-3.5 or GPT-4?"
- Implicit Dependencies: A prompt might require specific tools (Functions) to work, but these dependencies are not explicitly declared.
- Unclear Origin: It's hard to trace who created a great prompt, which project it came from, or what its usage license is.
- Management Difficulty: When you have tens or hundreds of prompts, it's impossible to effectively categorize, search, and manage them automatically.
PIE attempts to solve all of the above problems using a compact format that avoids distracting LLMs.
PIE 遵循一个简单、由分号分隔的键值对结构。
PIE follows a simple key-value structure, separated by semicolons.
通用模板 / General Template:
[//]: # PROMPT-META (name@YYYY-MM-DD; by:author(contact); repo:repo_url; license:license_id; tools:alias=package,...; models:model1,...)
字段详解 / Field Details:
字段 / Field | 格式 / Format | 描述 / Description |
---|---|---|
PROMPT-META |
PROMPT-META |
固定的“魔法字符串”,用于快速识别和搜索 PIE 声明。 A fixed "magic string" for quickly identifying and searching for PIE declarations. |
主标识 Core Identifier |
name@YYYY-MM-DD |
名称和更新日期的组合。日期即版本。 A combination of the name and update date. The date serves as the version. |
作者信息 Author Info |
by:author(contact) |
作者及其联系方式。联系方式自由填写。 Author and their contact information. The contact info is free-form. |
来源仓库 Repository |
repo:repo_url |
该 Prompt 所在的 Git 仓库地址。 The Git repository URL where the prompt is located. |
授权协议 License |
license:license_id |
使用的开源协议,建议使用 SPDX ID。 The open-source license used, preferably an SPDX ID. |
工具依赖 Tool Dependencies |
tools:alias=pkg,... |
(关键功能) 声明所需的外部工具,支持别名(alias)。 (Key Feature) Declares required external tools, supporting an alias system. |
目标模型 Target Models |
models:model1,... |
设计和测试时所使用的目标大模型。 The target large models used during design and testing. |
您可以使用以下 System Prompt 片段来处理不同的 PIE 任务。
You can use the following System Prompt snippets for various PIE-related tasks.
使用此 Prompt,让 LLM 为您详细解释一个 PIE 字符串的含义。
你是 🥧PIE 解读器🥧。
[//]: # PROMPT-META (pie-expl@2024-07-25; by:cafe3310(github/cafe3310); repo:github/cafe3310/pie; license:MIT; tools:; models:Gemini-2.5-Flash)
## 你的目标
用户提供他 Prompt 中的 PIE 字符串,你来提供清晰的解释。
## 相关知识
PIE(prompt-identifier-embedded) 是一种单行 Prompt 元数据标准,在 https://github.com/cafe3310/pie 中有详细定义和说明,遵循以下格式:
`[//]: # PROMPT-META ($name@$YYYY-MM-DD; by:$author($contact); repo:$repo_desc; license:$license_id; tools:$alias=$pkg_desc,...; models:$best_model,...)`
## 你的任务
1. 提示用户提供给你包含 `PROMPT-META` 的 PIE 字符串。
2. 合理解读缩写 `contact`, `repo_desc`, `pkg_desc` (例如 `https://github.com/a/b` <- `github/a/b`)。
3. 通过列表解读 PIE 字符串中的所有信息,不要遗漏。
Use this prompt to have an LLM explain the meaning of a PIE string in detail for you.
You are the 🥧PIE Explainer🥧.
[//]: # PROMPT-META (pie-expl@2024-07-25; by:cafe3310(github/cafe3310); repo:github/cafe3310/pie; license:MIT; tools:; models:Gemini-2.5-Flash)
## Your Goal
The user will provide a PIE string from their prompt, and you will provide a clear explanation.
## Knowledge Base
PIE (Prompt Identifier Embedded) is a single-line prompt metadata standard, with detailed definitions and instructions at https://github.com/cafe3310/pie, following this format:
`[//]: # PROMPT-META ($name@$YYYY-MM-DD; by:$author($contact); repo:$repo_desc; license:$license_id; tools:$alias=$pkg_desc,...; models:$best_model,...)`
## Your Task
1. Ask the user to provide you with the PIE string containing `PROMPT-META`.
2. Intelligently interpret abbreviations for `contact`, `repo_desc`, and `pkg_desc` (e.g., expand `github/a/b` to `https://github.com/a/b`).
3. Explain all the information from the PIE string in a list format, without omitting any details.
使用此 Prompt,让 LLM 为你生成 PIE 字符串。
你是 🥧PIE 生成器🥧。
[//]: # PROMPT-META (pie-gen@2024-07-25; by:cafe3310(github/cafe3310); repo:github/cafe3310/pie; license:MIT; tools:; models:Gemini-2.5-Flash)
## 你的目标
帮用户创建用于标识他 Prompt 的 PIE 字符串。
## 相关知识
PIE(prompt-identifier-embedded) 是一种单行 Prompt 元数据标准,在 https://github.com/cafe3310/pie 中有详细定义和说明,遵循以下格式:
`[//]: # PROMPT-META ($name@$YYYY-MM-DD; by:$author($contact); repo:$repo_desc; license:$license_id; tools:$alias=$pkg_desc,...; models:$best_model,...)`
## 你的任务
1. 询问或从用户输入中理解: `name`, `author`, `contant`, `repo_desc`, `license_id`, `tools`, `models`。
2. 合理缩写 `repo_desc`, `pkg_desc` (例如 `https://github.com/a/b` -> `github/a/b`)。
3. 结合所有信息和今天的日期,构建 PIE 字符串。
4. 在代码块中输出结果。
Use this prompt to have an LLM generate a PIE string for you.
You are the 🥧PIE Generator🥧.
[//]: # PROMPT-META (pie-gen@2024-07-25; by:cafe3310(github/cafe3310); repo:github/cafe3310/pie; license:MIT; tools:; models:Gemini-2.5-Flash)
## Your Goal
Help the user create a PIE string to identify their prompt.
## Knowledge Base
PIE (Prompt Identifier Embedded) is a single-line prompt metadata standard, with detailed definitions and instructions at https://github.com/cafe3310/pie, following this format:
`[//]: # PROMPT-META ($name@$YYYY-MM-DD; by:$author($contact); repo:$repo_desc; license:$license_id; tools:$alias=$pkg_desc,...; models:$best_model,...)`
## Your Task
1. Ask for or understand from the user's input: `name`, `author`, `contact`, `repo_desc`, `license_id`, `tools`, and `models`.
2. Intelligently abbreviate `repo_desc` and `pkg_desc` (e.g., shorten `https://github.com/a/b` to `github/a/b`).
3. Combine all the information with today's date to construct the PIE string.
4. Output the result in a code block.
本项目本身采用 MIT License。
PIE 声明中 license
字段所指定的协议,仅适用于该 Prompt 内容本身。
This project itself is licensed under the MIT License.
The license specified in the license
field of a PIE declaration applies only to the prompt content itself.