# redfox-community ships Agent Skills that front a paid key-gated API

> redfox-data/redfox-community is a folder-of-folders Agent Skills collection whose value depends on a key-gated API behind redfox.hk, with no license file, no releases, and a README that names five different install routes without pinning one. What it is really distributing is the catalogue, not runnable code.

**redfox-data/redfox-community** — 红狐数据（RedFoxHub） 技能合集：面向 Agent 的可复用 SKILL 集合，覆盖灵感、选题、文案创作、数据复盘等场景，持续更新。

- Repository: https://github.com/redfox-data/redfox-community
- Stars: 419 · Forks: 56
- Language: Python
- License: not declared
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/redfox-data-redfox-community

## The unit of distribution is a folder, not a package

The repository is a collection of Agent Skills, each one an independent subdirectory under skills/ built around a single SKILL.md file. That file is the whole entrypoint: YAML frontmatter plus prose covering when the skill triggers, the steps it walks through, its constraints, and examples. Optional references/, scripts/, and assets/ directories sit beside SKILL.md so a skill stays self-contained and references paths relatively. A minimal skill is therefore one markdown file with two frontmatter keys, name and description, the description meant to state the scenario and the boundary of what the skill claims to do. There is no build step, no manifest, and no registry entry. Installing is a file placement operation, which is why the same folder works whether it lands in a Cursor project .cursor/skills/ directory, a Cursor user skills directory, or whatever path another Agent Skills capable client expects. The trade is flexibility for reproducibility: nothing in that layout pins a version, so two machines can hold different behaviour from the same skill name after an upstream edit. The convention is also enforced socially rather than by a linter: the frontmatter guidance is framed as a recommendation to keep metadata clear so that search and import platforms can recognise a skill, which means the quality of the catalogue depends on contributors describing their own boundaries honestly. Nothing in the repository checks that a description matches what the skill does, or that a skill declares any dependency at all, even though the install prompt tells an agent to read SKILL.md and confirm its steps and dependencies.

## Python is the primary language, but no module is importable

Repository metadata tags the primary language as Python, yet the recorded file tree at the top level is only .github/, .gitignore, README.md, README.en.md, and skills/. No .py file, no pyproject.toml, no requirements.txt, and no setup entry appear anywhere in that listing. Whatever Python exists lives inside individual skill folders as optional scripts/, and the tree captured here does not descend into skills/ far enough to show them. So there is nothing to pip install and nothing to import; a consumer reads SKILL.md prose and copies the folder. Metadata language detection counts whatever bytes dominate a snapshot, and a large collection of markdown with a few helper scripts can tilt it to Python without producing an importable package. Read that tag as a hint about the auxiliary scripts a skill may call, not as a promise of a library surface.

## Every skill is only as useful as a key from redfox.hk

The README is explicit that all API requests need a valid API KEY, and the key is not issued by the repository. It is collected on a hosted settings page at redfox.hk/settings/api-keys, and every outbound link in the document carries the query parameter ?source=github, a referral marker on the site itself. This is the load-bearing dependency of the whole collection: the skills are front-ends for endpoints the vendor hosts, and none of that endpoint logic lives on the main branch here. The authentication section names no environment variable, no request header, and no scheme for how a key travels from the settings page into a skill at call time; it only says a valid key is required. A skill folder you copy locally is inert until a key is present somewhere the agent can reach, and that key belongs to an account on a third-party site. The collection also advertises a broader skill browser at redfox.hk/skills, so the repository and the hosted hub overlap in a way that makes it unclear which one is canonical for a given skill. What the repository does not carry is the contract that would make a key usable. The API section promises a hosted documentation page that will contain a request header description, a request parameter description, return value and data structure description, a request example, a response example, and a common status code list. That page lives on redfox.hk/apis, not in this repository, so the one document that would tell a skill author how to authenticate, what a failure looks like, and which status codes to branch on is the one thing a fork does not get. A team that vendors a skill folder inherits a dependency on a secret whose retrieval and transmission path is documented only on a page the team may not be able to reach later.

## The endpoint catalogue contradicts itself on identity

The multi-platform API section is the longest part of the document and the messiest. The WeChat Channels block lists 获取视频号账号作品列表 (广域库) and 视频号作品链接更新详情（实时） both pointing at the same identifier, shipinhao/OVUTOTCV, so two differently named operations resolve to one endpoint page. The Toutiao block lists 获取今日头条账号作品列表 (实时) twice under two different identifiers, jinritoutiao/28CFGF5I and jinritoutiao/R1IGR09H, for what is nominally one operation. Two separate ID families also appear for the same idea: pure-digit slugs such as douyin/20060017 and xiaohongshu/20060016 beside opaque eight-character strings, which suggests endpoints minted at different times by different tooling. A Xiaohongshu link is even missing its closing bracket, the label reading 查询小红书账号作品列表（优质库 with the parenthesis left open. None of this blocks reading the catalogue, but it means the list cannot be trusted as a machine-readable index, and you should open the endpoint page before wiring a call to any identifier.

## Two coverage pools and no rule for choosing between them

Endpoints are labelled either 优质库 or 广域库, and the same logical operation frequently exists in both, for example fetching Douyin work detail, searching Douyin accounts, listing a Douyin account's works, and the equivalents for WeChat official accounts all appear as a premium-pool and a broad-pool pair. The document never says which to prefer, how their coverage differs, or whether one consumes more quota. Coverage is also uneven across platforms. Douyin, Xiaohongshu, official accounts, and Bilibili offer premium-pool variants; WeChat Channels and Kuaishou entries are broad-pool only; Toutiao is marked 实时 rather than either pool; and TikTok entries carry no pool label at all. Treat the pool suffix as a routing hint you must verify per endpoint, not a quality ranking you can generalise from. Several endpoints are also asynchronous, submitted as a task, for example the 视频提文案-提交任务 and 链接提文案-提交任务 families on several platforms, which implies a submit-then-poll flow the repository does not spell out. Platform names are written inconsistently even where the endpoint exists: Tiktok appears once as Tiktok and once as TikTok, X(Twitter) is carried inside the URL path itself, and the pools are marked with two different bracket styles, half-width round brackets on the Douyin entries against full-width parentheses elsewhere. Two endpoints sit outside the pool system entirely and are labelled as real-time fetches by URL rather than by index, including 根据作品地址获取公众号作品（实时） and 今日头条 entries, which hints at a pull-from-address fallback distinct from the keyed search endpoints. The catalogue covers Douyin, Xiaohongshu, WeChat official accounts, WeChat Channels, Kuaishou, Bilibili, Toutiao, TikTok, and X, with the X section cut off partway through in the copy on hand.

## Five install routes and no canonical one

The document offers five ways in and blesses none. You can copy a skill folder into your client's skills directory, deferring the exact path to whatever your client documents. You can paste a natural-language prompt into an agent, which is given as a bash-fenced block that names a specific skill, xiaohongshu-realtime-search, with its GitHub tree URL and five ordered instructions: check whether the skill is already installed, read SKILL.md and the README to confirm steps and dependencies, install it into the project's existing skills path if missing, diff against the remote and report changes if present, then report the install path, how to trigger it, and a short usage example. That block also tells the agent that when the network is restricted it may try git clone or curl, which is an instruction to fetch and run code by URL. Then there is a CLI path built on npx, run from an interactive terminal and copyable as-is:

```bash
npx skills init //安装skills cli

npx skills add redfox-data/redfox-community //检索redfox仓库选择安装skill

npx skills add https://github.com/redfox-data/redfox-community/tree/main/skills/seedance-video-gen //安装具体skill
```

The first command installs the cli itself, the second points at the whole repository so you can browse and pick, and the third takes a single skill's tree URL. You are then prompted to install into a chosen agent folder or globally. Note that the middle command hands the tool an entire repository of folders to enumerate, so the set of skills you can select is whatever happens to sit under skills/ at that moment, with no version in the command line. Finally two hosted registries mirror the collection: SkillHub at skillhub.cn/skills, searched by the skill's Chinese name such as 公众号爆款文章查询 or 抖音每日最具影响力账号, and ClawHub at the clawhub.ai user page for redfox-data. Because a skill is just a folder, all five routes produce the same result on disk, but only the npx route records what it installed.

## What the repo does not settle: license, releases, and count

Three gaps matter before you depend on this. First, licensing: the metadata license field is empty, and the recorded top-level tree contains no LICENSE file, so there is no stated grant for the prose, the frontmatter, or any bundled script, and nothing here establishes terms for redistribution or for a commercial agent product. Treat that as unresolved rather than assume a permissive default. Second, releases: there are no GitHub releases at all, so the collection is only reachable at whatever main happens to hold, and the documented update path is the prompt step that diffs remote against installed content. Pin a commit if you need repeatability. Third, inventory: the tree captured for this repository stops at skills/ and does not enumerate the skill subdirectories, so the size and contents of the collection cannot be verified from what is on record, and one install example names a skill, seedance-video-gen, that the shallow listing does not confirm exists. The last push on record is 2026-09-29, the repository is not archived, and it shows 419 stars and 56 forks with zero open issues.

## Conclusion

Use it if you already have a redfox.hk key and want the specific Chinese-platform endpoints its skills wrap. Do not adopt it as a dependency: the repo has no license file, no releases, no Python module to import, and a README that contradicts itself on endpoint identity, so nothing here is safely forkable or version-pinned. Before betting a build on it, confirm the license, pin a skill folder by commit, and read the SKILL.md of the one skill you actually need rather than trusting the catalogue.

## FAQ

### Does redfox-data/redfox-community require an API key to use its skills?

Yes. The README states that all API requests need a valid API KEY, and the key is obtained from a hosted settings page on redfox.hk rather than from the repository. No environment variable or request header for carrying that key is specified.

### Can I pip install redfox-community as a Python package?

No. The top-level file listing contains no pyproject.toml, no requirements.txt, and no Python module, only README files, .github/, .gitignore, and skills/. Python is the tagged primary language because of scripts bundled inside individual skill folders, and installation is a folder copy.

### What license applies to the skills in redfox-community?

Unclear, and it is not resolved by the repository itself. The license metadata field is empty and no LICENSE file appears in the recorded top-level tree, so there is no stated grant covering the skill text or any bundled script. Confirm the terms with the maintainers before redistributing.

### What is the difference between the 优质库 and 广域库 endpoints?

The document does not explain it. Both labels appear on the same logical operations, for example Douyin work detail and WeChat official account lookups exist as a premium-pool and a broad-pool pair, but no coverage, cost, or quality guidance is given. Some platforms, including WeChat Channels and Kuaishou, list broad-pool entries only.

### How do I install a single skill from redfox-community with a command?

The document gives three npx commands, starting with npx skills init, then npx skills add redfox-data/redfox-community to browse and choose, or npx skills add pointed at the skill's own tree URL, for example the seedance-video-gen path. You can also copy the skill folder into your client's skills directory directly.

### Are there versioned releases for the redfox-community skills?

No. The repository has no GitHub releases, so skills are only available from the current state of the default branch, main. The documented way to refresh an installed skill is the update step in the agent prompt, which compares remote content and reports what changed.

## Sources

- [Issues](https://github.com/redfox-data/redfox-community/issues)
- [README](https://github.com/redfox-data/redfox-community/blob/main/README.md)
- [redfox-data/redfox-community on GitHub](https://github.com/redfox-data/redfox-community)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/redfox-data-redfox-community