redfox-data/redfox-community: Agent Skills for Chinese Social Media Research

What redfox-community actually ships

The repository is a skills directory, not an application. Its top level holds README.md, README.en.md, a .github folder, a .gitignore and skills/. Everything of substance lives under skills/, where each subdirectory is one skill and each skill must contain a SKILL.md. That file is the entry point: YAML frontmatter plus a body that the README describes as covering trigger conditions, steps, constraints and examples. Optional references/, scripts/ and assets/ folders sit beside it so a skill stays self-contained and refers to its own files by relative path.

The intended user is someone running an agent client that understands the Agent Skills format, such as Cursor or Claude Code, and who wants the agent to follow a defined procedure for a content task: finding inspiration, picking a topic, drafting copy, or reviewing performance data. The README names skills such as xiaohongshu-realtime-search and seedance-video-gen, and says the collection is community maintained and updated continuously. The last push to the default branch was on 2026-09-09, so the repository is not dormant, but the README states no release cadence or versioning scheme.

How a SKILL.md drives the agent

There is no daemon and no server. The mechanism is file-driven: the agent reads SKILL.md, matches the current request against the trigger conditions in the frontmatter and body, then executes the steps the file lays out. The README's own example of the frontmatter is deliberately minimal, with a name and a one-line description of the scenario and capability boundary.

---
name: Example Skill
description: A one-line description of the scenario and capability boundary.
---

Because the skill is just text plus optional scripts, the data flow splits in two. The reasoning, sequencing and output shape come from SKILL.md. The actual platform data comes from RedFoxHub HTTP APIs, which the README lists per platform: Douyin, Xiaohongshu, WeChat public accounts, WeChat Channels, Kuaishou, Bilibili, Toutiao and TikTok. Those endpoints cover account info, work details, keyword search, account work lists, comment retrieval and video-to-transcript submission. The README says every API request requires a valid API key, and points to the RedFoxHub settings page to obtain one. So the architecture is: agent client reads local skill files, skill files tell the agent which HTTP calls to make, and RedFoxHub returns the platform data. Nothing in the repository hosts or proxies that data itself.

Installing a skill and running a first search

The README gives four installation routes. The most direct for a terminal user is the skills CLI, which the README shows as npx commands. The first initialises the CLI, the second lists skills from the repository so you can pick one, and the third installs a named skill straight from its GitHub path.

npx skills init
npx skills add redfox-data/redfox-community
npx skills add https://github.com/redfox-data/redfox-community/tree/main/skills/seedance-video-gen

After the add command, the README says you follow the prompts to place the skill either in a specific agent folder or globally. The second route is manual: copy the skill subdirectory into your client's skills directory, for example Cursor's user skills location or a project-level .cursor/skills/ folder, or point the client's add-skill entry at the subfolder. The README defers to your client's documentation for the exact path. The third route is to ask the agent itself to install the skill, and the README supplies a prompt template that tells the agent to check whether the skill is already present, read the remote SKILL.md, install it into the appropriate directory, and report the install path, the trigger and a short usage example. The fourth route is a catalogue: search SkillHub at skillhub.cn/skills by the Chinese skill name, or the RedFoxHub page on clawhub.ai.

Before any of this produces data, you need credentials. The README states that all API requests require a valid API key and directs you to the RedFoxHub settings page to create one. Note that the README does not document where the key is stored, which environment variable or config key the skills read it from, or how a skill reports an authentication failure. That gap matters: you will be reading SKILL.md files to find the expected key name before your first call succeeds.

Where this design breaks down

The dependency on RedFoxHub is the largest constraint. Every skill that returns platform data is only as useful as the API key behind it, and the README does not describe a free tier, a rate limit, a quota or what happens when a key expires. If your account cannot obtain a key, the data-oriented skills have nothing to call. Skills that are purely prompt logic, such as a topic-selection or copywriting procedure, would still run, but the README does not separate which skills need the API from which do not.

The second limitation is portability. Skills are plain folders with a SKILL.md, so they are portable in principle, but the README's own installation instructions are tied to clients that implement the Agent Skills convention. If your agent runtime has no notion of a skills directory, you are reduced to pasting SKILL.md contents into a prompt, which loses the frontmatter-driven triggering.

The third is the missing licence. The repository declares no licence, and the README does not discuss terms of use. That is a real blocker for commercial adoption: without a licence file, the default position is that no rights are granted, and the API terms are a separate question from the repository contents. Treat the absence as something to resolve before shipping anything built on these files.

Finally, the documentation is uneven. The README explains the folder convention and the API catalogue in detail, but it does not document versioning, a changelog, a rollback path, or how a skill is updated once installed. The agent-install prompt tells the agent to compare remote content and update as needed, which implies updates are manual and agent-driven rather than managed by a package manager.

How it differs from a scraping library or framework

The closest comparison is a general-purpose social media scraping library, which you import into your own code and call directly. The difference is where the logic lives. With a scraping library, your application owns the request flow, the retry logic and the parsing, and you maintain all of it. With redfox-community, the flow lives in a SKILL.md that an agent reads, and the data comes from a hosted API rather than from your own requests to the platform. You trade control over the request path for not having to maintain parsers against platforms that change their markup.

That trade has a cost. A library runs in your process, offline from any vendor, and you can read every line of what it does. Here, the skill files are readable but the data layer sits behind an API key, and the README does not describe response schemas beyond saying the API docs contain them. If your requirement is reproducibility or a full audit trail of how a number was produced, a self-hosted library is the better fit. If your requirement is an agent that can answer a question about Douyin or Xiaohongshu trends without you writing collection code, the skill bundle is the shorter path.

Maintenance, licensing and what to check before adopting

Maintenance signal is thin but positive: the last push was on 2026-09-09, and the description says the collection is updated continuously. There are no retrieved releases, so there is no versioned artifact to pin and no changelog to read. Upgrading therefore means re-pulling the skill folder and diffing SKILL.md, which the README's agent prompt describes as the update path. Budget for that manual step, and expect a skill you depend on to change without a version number telling you so.

On licensing, the repository carries no licence identifier. That is not a legal opinion, just the observation that nothing in the repository grants rights, and the API terms on RedFoxHub are a separate agreement from the repository contents. If you plan to redistribute a skill or embed it in a product, resolve the licence question first.

The practical checks are narrow. Confirm a RedFox API key is obtainable for your account, read the SKILL.md of the specific skill you want to see which API endpoints it calls, and verify that the client you use actually reads a skills directory rather than requiring pasted prompts.

Editorial conclusion

Adopt redfox-community if your agents already work in Cursor, Claude Code or a similar client and you need repeatable Douyin, Xiaohongshu, WeChat or Bilibili research steps rather than one-off prompts. Skip it if you have no RedFox API key, if you need an offline tool, or if you want a runnable service rather than prompt-and-script packages. Verify first that the repository carries a licence file, because none is declared, and confirm on the RedFoxHub settings page that a key can be issued for your account before you build a workflow around these skills.

Frequently asked questions

Official sources

1. Issues
2. README
3. redfox-data/redfox-community on GitHub