bieji-codex-skills

Bieji AI Skills

This repository publishes two reusable AI assets:

WeRead Web GPT architecture

flowchart LR
    U["You or a user"] --> G["ChatGPT Custom GPT"]
    G --> A["GPT Action OpenAPI"]
    A --> V["Vercel actions bridge"]
    V --> W["Official WeRead Agent Gateway"]
    W --> D["Your WeRead data"]

The public package contains Custom GPT instructions, an OpenAPI schema, and the Vercel bridge source. It does not contain the previous local direct-call WeRead Skill.

Bilingual interface

1. Use the language links at the top of this page to switch repository documentation.
2. The deployed bridge landing page provides a persistent 中文 and English switch.
3. The Custom GPT follows the user's language automatically.
4. Enter 切换中文 or Switch to English to override the current GPT language.

Knowledge Card Maker

The Skill follows the complete SuperMemo 20 rules, emphasizes atomic retrieval and transfer, and returns a reviewable draft before any Anki write.

Repository use

Clone the repository and launch Codex from inside it. Codex discovers the Skill under .agents/skills.

git clone https://github.com/benabcben/bieji-codex-skills.git
cd bieji-codex-skills

User-level install

mkdir -p "$HOME/.agents/skills"
cp -R .agents/skills/knowledge-card-maker "$HOME/.agents/skills/"

Invoke it with:

$knowledge-card-maker Turn this material into a reviewable set of durable cards.

WeRead Web GPT setup

The files are under weread-web-gpt.

1. Deploy the bridge

Deploy weread-web-gpt/actions-bridge as a Vercel project and configure two server-side environment variables:

WEREAD_API_KEY=your-own-weread-key
ACTION_TOKEN=a-new-long-random-token

Keep both values out of Git, issues, screenshots, and chat messages.

2. Configure the Custom GPT

1. Create or edit a Custom GPT in ChatGPT.
2. Paste GPT-INSTRUCTIONS.md into Instructions.
3. Add a new Action and paste openapi.yaml.
4. Replace servers.url with your deployed HTTPS domain.
5. Configure Action authentication as an API Key with Bearer auth.
6. Enter ACTION_TOKEN. Never enter WEREAD_API_KEY in the GPT editor.
7. Test Search for The Three-Body Problem and 切换中文.

Security and sharing boundary

The bridge only allows an explicit read-only endpoint list, injects skill_version server-side, blocks identity and credential fields from model input, limits request and response sizes, uses timeouts, and sends no-store security headers.

Each user should deploy a separate bridge with their own WeRead key. A shared Custom GPT backed by one fixed owner key can expose that owner's shelf, highlights, thoughts, progress, and reading statistics to every authorized caller. Keep the GPT private unless the deployment provides genuine per-user authentication and key isolation.

Project structure

.
├── .agents/skills/knowledge-card-maker/
├── weread-web-gpt/
│   ├── GPT-INSTRUCTIONS.md
│   ├── openapi.yaml
│   └── actions-bridge/
├── docs/
├── README.md
└── README.zh-CN.md

Contributing

Open an issue for a reproducible bridge failure, unclear field semantics, or a concrete card-quality problem. Remove all credentials, reading notes, highlights, and personal identifiers before posting.

If the project helps your workflow, use the GitHub Star button to bookmark it and help others discover it.

License

MIT. See LICENSE.