Freedom-To-Chatgpt-Claude-Agent: a bilingual prompt library plus account and payment walkthrough

What the repository actually is, and who it is for

The README opens by describing the project as a Chinese-language resource collection for AI beginners: a prompt library covering role prompts, text-to-image, text-to-video, Skill and plugin/MCP categories, combined with a full walkthrough for US-region account registration and Apple gift card payment. The stated goal is to cover three things in one repository, roughly phrased as being able to use AI, being able to get access, and being able to pay for it. Former names listed in the README are ChatGPT 调教指南, 咒语指南 and 聊天提示词指南.

The audience is explicit: 新手友好, beginner friendly, and bilingual. The prompt entries carry difficulty markers such as 新手 and model tags naming GPT, Claude, Gemini, DeepSeek, Kimi and others, so a reader can filter by how much experience they have and which model they use. If you are an engineer looking for a package to add to a dependency file, this is the wrong repository. There is no published package, no installable module and no API surface. The deliverable is Markdown plus a static site.

How the prompt library is built: one data source, generated Markdown

The most interesting structural detail is that the prompt sections are not hand-written into the README. A note in the README states that the section directory and each section's contents are generated by scripts/build.js from a single data source at web/data/prompts.js, and that the generated region should not be edited by hand. The generated area is fenced by an HTML comment marker reading AUTO-PROMPTS:START, with a note telling contributors not to modify it.

That means the real content lives in a JavaScript data file, and the README is an output artifact. For anyone who wants to add a prompt, the correct path is editing web/data/prompts.js and re-running the build, not editing the Markdown. The README also states that non-original entries carry a source field naming where they came from, and that the plugin and MCP section is collected and translated from official documentation and open source collections in the Claude Code, Codex CLI and ChatGPT ecosystems.

The counts given in the README are 280+ prompts split across seven categories: 48 beginner entries, 151 role prompts, 29 text-to-image, 34 text-to-video, 44 Skill recommendations, 33 plugin/MCP entries and 78 prompt technique entries. Those numbers add up to more than 280 because entries can be counted per category listing; treat the totals as the README's own figures rather than an audited count.

Installing nothing: using the online version or cloning the repository

There is no package to install. The README points to a hosted online version at https://wikieden.github.io/Freedom-To-Chatgpt-Claude-Agent/ for searching and one-click copying, and that is the fastest path. If you want the files locally, clone the repository and open the built page.

git clone https://github.com/wikieden/Freedom-To-Chatgpt-Claude-Agent.git
cd Freedom-To-Chatgpt-Claude-Agent

The repository root contains index.html, web/, scripts/, docs/ and the Markdown files, so opening index.html in a browser or serving the directory gives you the same page the GitHub Pages site serves. If you intend to edit prompts, the workflow the README describes is to change the data file and rebuild.

node scripts/build.js

The README does not document the Node version, the dependencies, or what the script prints on success, so treat the command as the documented entry point rather than a guaranteed one-liner. After a rebuild, the region between the AUTO-PROMPTS:START marker and its closing counterpart is what changes.

For the account side, the README gives a four-step sequence: configure a proxy client with TUN mode and verify the exit IP on whoer.net, register a US-region Apple ID with your own Chinese phone number and a tax-free state address with payment set to None, buy a US-region Apple gift card and load the balance, then subscribe to Claude or ChatGPT through in-app purchase on iOS. The full steps live in docs/账号注册与订阅指南.md, which the README links by anchor.

Freeroad: the routing layer the README attaches to the account guide

A separate section of the README covers Freeroad, described as an open source AI routing configuration for Clash Verge Rev, FLClash and Shadowrocket. It is worth being precise about what it is and is not: the README states plainly that Freeroad does not provide proxy nodes. It sits on top of an existing node subscription and adds grouping, routing, DNS and exit verification.

The routing logic splits AI traffic into four isolated groups named in the README as Claude, OpenAI/AI, Google/Gemini/Antigravity and other AI services, so that switching a line for ordinary browsing does not move the exit used by a key service. Country selection comes first, then either automatic speed testing within that country or a pinned node. The README notes that automatic speed testing does not cross countries, and that pinning a single node is preferable when account stability matters. There are also measures aimed at protocol fallback, described as DNS splitting, disabling on UDP failure and client-specific rules.

Two constraints matter. First, the README says different clients use different configuration files and they cannot be mixed: Clash Verge Rev and FLClash use a global extension script, while Shadowrocket uses a native .conf file. Second, the README states directly that routing configuration only manages routing behaviour and cannot guarantee an account will never trigger risk controls, since the outcome depends on node quality, IP reputation, account region, client environment and usage. That disclaimer is the honest part of the section.

Where this project is the wrong tool

The account and payment guide is the fragile half of the repository. It depends on third-party services the project does not control: the README names a specific proxy provider, a code-receiving platform, a gift card shop, and links to specific client release pages and an App Store listing. Any of those can change pricing, availability, registration rules or regional policy without the repository being updated. The README's own disclaimer about risk controls applies equally here. If you need a stable, documented signup procedure for a company, this is not where you should be reading it, because you cannot pin a vendor's onboarding flow the way you pin a dependency version.

On the prompt side, the limitation is different. The entries are short, single-block prompt templates with tags, not a curated evaluation of model behaviour. Nothing in the README indicates that any prompt has been measured against a baseline, so the difficulty and model tags are editorial labels. Several entries reference model versions and product features with dates attached, such as a 2026 selection guide and entries describing Gemini's Daily Brief and DeepSeek V3.1. Those entries age. A prompt library that tracks product releases needs continuous editing, and the last push to this repository was on 2026-09-01, which is recent enough that the repository is not stale, but the update cadence is not documented anywhere in the README.

Alternatives and how they differ in approach

The obvious alternative is the prompt collection maintained by the model vendors themselves, such as the official documentation and prompt guides referenced in this repository's plugin and MCP section. The difference is scope and language: vendor documentation is authoritative about a single model and is written in English first, while this project is Chinese first, bilingual per entry, and deliberately spans GPT, Claude, Gemini and Chinese models in one place. If you only use one vendor's model, its own guide will be more accurate about that model's current capabilities.

A second alternative is a general prompt-sharing site. Those are larger and continuously updated by a crowd, but they are not versioned, not licensed as a repository, and not reproducible. This project's prompts live in web/data/prompts.js under an MIT licence, so you can fork the data file, strip the entries you do not want, and rebuild the site with your own selection. That is a real difference: the content is structured data, not a feed.

A third option is to skip prompt collections entirely and write your own system prompts. That is more work but avoids the tagging problem, since you know which model and which version you are targeting. The README's own universal prompt template entry, a fill-in-the-blanks role/goal/background/format template, is arguably the only entry you need if you go that route.

Licence, maintenance and what an upgrade costs you

The repository is MIT licensed, with a LICENSE file at the root. For a documentation and data repository, MIT is permissive: you can copy prompts, modify them, and redistribute, provided the licence and copyright notice are retained. The README asks for donations through an Alipay image, which is a request, not a licence term, and does not change your rights under MIT. Nothing here suggests any restriction on commercial reuse of the prompt text; if you plan to ship the prompts inside a product, read the LICENSE file yourself rather than relying on this summary, and note that the README states non-original entries carry a source field, so some entries may be adapted from other projects whose terms you would need to check separately.

Upgrade cost is low in the software sense and non-trivial in the content sense. There is no dependency graph to maintain, no breaking API, and no migration. Pulling the latest changes gets you new prompt entries and revised guides. The cost is that the generated Markdown is not hand-editable, so if you fork and customise, you must keep your changes in web/data/prompts.js or they will be overwritten the next time scripts/build.js runs. The README's warning not to hand-edit the generated region is the single most important operational fact in the repository.

Editorial conclusion

Adopt it if you want a searchable, copy-paste bilingual prompt set and you are comfortable reading Chinese, since the README, docs and prompt entries are primarily Chinese with an English README_EN.md. Skip it if you need a maintained software library, an API, or English-first documentation: the repository ships Markdown, a static site and a build script, and the English README is a translation rather than the source. Before relying on it, open web/data/prompts.js and confirm the prompt you want is still there, and check whether the account and payment guide in docs/账号注册与订阅指南.md still matches the current signup flow, because that guide describes third-party services and regional requirements that change without notice.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. Project website
4. README
5. wikieden/Freedom-To-Chatgpt-Claude-Agent on GitHub