Freedom-To-Chatgpt-Claude-Agent: A Bilingual Prompt Library With Account and Payment Guides

What Freedom-To-Chatgpt-Claude-Agent actually contains

This is a resource collection, not a program. The README describes four modules: a prompt library of 280+ bilingual entries, a full account registration walkthrough for US Apple ID, Claude and OpenAI ChatGPT/Codex, a payment path built around US Apple gift cards and iOS in-app purchase, and a proxy setup guide. A fifth module points to a separate repository, freeroad, which holds the routing configuration.

The prompt library is split into seven sections: beginner (48 entries), role prompts (151), text-to-image (29), text-to-video (34), skill recommendations (44), plugins/MCP (33) and prompt techniques (78). Each entry is labelled with a difficulty marker, the models it targets, and an attribution field. The README states that non-original entries carry a source field, and that the plugin/MCP section is compiled and translated from official documentation and open source collections rather than written from scratch.

Who it is for is stated plainly in the README: Chinese-speaking AI beginners. The framing is a single chain from using the tools to being able to access them to being able to pay for them. That framing is also the main constraint. Nothing here is an executable artifact, so evaluation has to be about editorial quality and link hygiene, not runtime behaviour.

How the prompt library is generated from web/data/prompts.js

The build model is the most technically interesting part. According to the README, the section index and all section content in the README are generated by scripts/build.js from a single data source at web/data/prompts.js. The generated region is fenced by an AUTO-PROMPTS:START marker, and the README explicitly says not to hand-edit it.

That means the repository has one source of truth and at least two outputs: the README and the online site served from index.html. Editing a prompt means editing the JavaScript data file and re-running the build, not editing Markdown. For anyone maintaining a prompt collection, this is the pattern worth copying, because it removes the usual drift between a site and a repository.

The cost is that the README is not a normal README. A large share of it is generated output, so reading it on the repository page gives you the whole prompt set inline. The repository root also carries .nojekyll, which is what lets the site be served as-is from the branch. The entry list shows .github/, .kiro/, .vscode/, ChatGpt-receipt.md, docs/, images/, scripts/ and web/ alongside the two READMEs, so the layout matches that description.

Using the prompt library and rebuilding it locally

There are no published releases, and the README does not give an install command for a package. The project is consumed two ways: through the hosted site, or by cloning the repository and running the build yourself. The README points to the online version for search and one-click copy at https://wikieden.github.io/Freedom-To-Chatgpt-Claude-Agent/.

To work on it locally, clone the repository. The README names the build script and the data file but does not document the exact invocation, so read scripts/build.js before assuming how it is meant to be called. The repository's primary language is JavaScript, and the build script lives in scripts/ while the data file lives in web/data/.

After the build, the generated block between the AUTO-PROMPTS markers in README.md should reflect whatever is in web/data/prompts.js. If you edit the README directly inside that block, the next build will overwrite it, which is the failure mode the README warns about.

For first real use, the practical path is the online site rather than a local build. Open it, search for a scenario, and copy the prompt. A representative beginner entry is the universal template, which the README renders as a fill-in-the-blank sentence: you are a [role], my goal is [goal], background is [background], output in [format/tone] at about [length], and ask me first if information is missing.

The account, payment and routing guides are the fragile half

The prompt library ages slowly. The account and payment guides age quickly. They depend on third-party services, store policies and proxy clients, and the README links out to specific commercial sites for proxy subscriptions, virtual phone numbers and gift card resale, several of which carry affiliate parameters in the URL.

That is a real editorial weakness, and the README is honest about the limits of the routing half. It states that a routing configuration can only manage routing behaviour and cannot guarantee an account will never trigger risk controls, and that the outcome still depends on node quality, IP reputation, account region, client environment and usage pattern. It also says to comply with local laws and the terms of service of the relevant platforms.

The freeroad strategy itself is described in concrete terms: LAN and domestic services go direct, ads and matched tracking requests are rejected, other foreign traffic defaults to the proxy, and AI traffic is split into four isolated groups (Claude, OpenAI/AI, Google/Gemini/Antigravity, other AI services). There is a two-level exit selection, country first and then automatic speed test or a pinned node, with the note that automatic speed testing never crosses countries. The README also warns that configuration files are not interchangeable: Clash Verge Rev and FLClash use a global extension script, while Shadowrocket uses a native .conf.

If you only want prompts, skip this half entirely. If you want the account path, treat every link as something to verify yourself rather than as a recommendation.

Where this repository is the wrong tool

If you need an English-first prompt library, the README_EN.md exists but the body content is Chinese, and the online site is presented in Chinese. English readers get the structure and the English README, not a translated prompt set.

If you need something installable, this is the wrong shape. There is no package, no release, no CLI. The only runnable piece is the build script, and its job is to regenerate documentation.

If you need a maintained library in the software sense, note the last push was on 2026-09-01, which is recent, but there are no retrieved releases, so there is no versioned artifact to pin. There is also no test suite described in the repository, so a change to web/data/prompts.js is validated by reading the generated output, not by CI.

Finally, if your concern is prompt quality measured against outcomes, the README gives no evaluation methodology. Entries are labelled by difficulty and target model, and sourced entries carry attribution, but there is no statement about how prompts were selected or checked. Treat it as a curated collection, not a benchmarked one.

Alternatives and how the approach differs

The most direct alternative is the upstream collections this project says it draws from. The README states that the plugin/MCP section is compiled and translated from official documentation and open source collections, and that the ecosystem resource document aggregates MCP servers, 100+ Claude Code subagents and 50+ skills from four well-known open source collections. If you read English and want the primary sources, going upstream gives you the original, more current material without a translation layer.

The second alternative is a generic prompt-sharing site. The difference is structural rather than editorial: those sites are databases you search, while this repository is a data file plus a build script plus a hosted page. If you want to fork and self-host a prompt library, the second model is easier to own. If you just want to find a prompt, the first is faster.

For the routing half, the alternative is writing your own Clash rules by hand. Freeroad's stated contribution is not nodes but grouping, routing, DNS and exit verification layered on top of a subscription you already have. If you already maintain clean rule sets, the added value is the four-way AI isolation and the per-service verification step, not the routing concept itself.

Maintenance cost, licence and what to check before forking

The repository is MIT licensed, which permits reuse and modification with attribution, but note that the licence covers the repository's own content. Third-party material compiled into the prompt set is attributed through source fields, and the README says non-original entries are marked. If you republish, those attributions are the part to preserve. This is a description of what the repository states, not legal advice.

Maintenance splits into two very different jobs. The prompt data file needs editorial work: adding entries, keeping model names current, checking that sourced entries still point somewhere valid. The guide documents need link work: proxy clients ship new versions, stores change payment flows, and affiliate links rot. The build script is the stable part, and because it is a single generator over one data file, a contributor who only cares about prompts never has to touch the guides.

The upgrade path is a git pull plus a rebuild. There are no releases to track and no migration steps documented, so the risk is not a breaking API change but a stale link or an outdated model reference. Before forking, confirm three things: that scripts/build.js runs on your Node version, that web/data/prompts.js is the file you would edit, and that you are comfortable with the third-party links the account and payment guides depend on, since those are the parts most likely to need replacing.

Editorial conclusion

Adopt this if you are a Chinese-speaking beginner who wants prompts and a step-by-step account and payment path in one repository, or if you maintain a prompt collection and want the scripts/build.js generation model. Do not adopt it if you need an English-first prompt library, a maintained package with releases, or a tool that runs on its own. Verify first that the online site at wikieden.github.io/Freedom-To-Chatgpt-Claude-Agent/ still loads, that web/data/prompts.js is the file you would edit rather than the generated README sections, and that the third-party links in the payment and proxy sections are ones you are willing to follow.

Frequently asked questions

Official sources

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