workbuddyskills: a daily-synced archive of Tencent's WorkBuddy marketplace

A marketplace archive, and why that is useful

WorkBuddy, related to Tencent's CodeBuddy AI coding products, runs a marketplace of skills, connectors and expert packages, and marketplaces are hard to study: you need the app, an account, and patience to inspect one package at a time. workbuddyskills, maintained by infometa, is an unofficial archive that mirrors those public marketplace packages into a plain GitHub repository for offline study, retrieval and, in the README's phrase, comparing implementations against each other.

The scale is the point. The archive holds 295 skill packages with SKILL.md files at their top level, 103 connector configurations, 410 expert plugins combining agents, skills and avatars, plus two plugin markets with 57 official and 31 team packages. A CATALOG.md lists every item with its Chinese-language purpose, prerequisites, whether an API key or login is required, and clickable links. For anyone researching how agentic-skill ecosystems are packaged in practice, that index is the artifact.

What the categories reveal

The breakdown tells you what a Chinese enterprise AI-assistant market actually contains. The biggest skill category is AI and agent tooling at 159 packages, followed by Tencent, WeChat and WeCom integrations at 47, and documents, office and collaboration at 44. Search and research knowledge, design and UI and maps, data and finance, content and marketing, cloud and deployment make up most of the rest, with development and engineering a surprisingly thin 2 packages.

The expert plugins, 410 of them, are organized by profession: content creation 45, technical engineering 43, data intelligence 41, finance and investment 38, marketing growth 36, industry consulting 33, down to a long tail of specializations. The connector set ships mcp.json configuration templates, which makes the archive quietly valuable as a reference corpus: hundreds of real MCP connector configs, skill file layouts and expert definitions that show conventions no documentation page teaches.

Browsing without installing anything

The README's quick-browse block is the fastest way to understand the archive's texture:

# 技能
ls skills | head

# 某个技能
cat skills/lark-unified/SKILL.md

# 专家（例：全域内容分发）
ls experts/content-distribution-team/agents
cat experts/content-distribution-team/.codebuddy-plugin/plugin.json

# 连接器
ls connectors | head
cat connectors/westock-mcp/mcp.json

Every item is a plain directory of text files, so grep and any code search turn the corpus into a dataset. For skill authors, reading twenty real SKILL.md files from a production marketplace, with their prerequisite declarations and packaging conventions, is a faster education than most writing on the subject, and that is precisely the study use the archive declares.

The copyright position, stated by the archivist

The README does not pretend this is neutral ground. A prominent notice states that the content comes from WorkBuddy and CodeBuddy public CDNs and market packages, that copyright belongs to the original authors and Tencent's related products, that the repository is for learning and archiving only, and that unauthorized commercial redistribution is not permitted. The licence section repeats the caution: individual third-party skill and expert packages may carry their own licences in their subdirectories.

The repository itself publishes no licence file, which is consistent with its self-description: it is an archive of other people's work, not a licensed corpus. The data sources are listed openly, the public CDN URLs the packages are distributed from, including the skill marketplace zip, connector config archive, and expert bundle endpoints, and an automated script checks those sources daily and syncs updates. Openness about provenance is the right posture for a project whose entire existence depends on tolerance of the use case.

Putting packages back into WorkBuddy

For WorkBuddy users, the archive doubles as a manual marketplace. The README's table maps content to destination paths: skills copy into ~/.workbuddy/skills/<name>/, expert packages into the plugins marketplace experts folder, and connector marketplace items into the connectors directory under your home.

One boundary is stated where it matters: connectors still require enabling inside the app with OAuth or QR-code authentication, and the archive contains only their configuration templates, so the connectors directory is reference material rather than a working install. The archive may also lag or diverge from the official market, and the README directs users to the official marketplace as the source of truth. Treat the install-back path as convenience for study sessions, not a distribution channel.

Limits and the questions it raises

The caveats are the honest ones for an unofficial mirror. Content drifts from the official market daily, in both directions, and the README says the official market governs. Consequences of using the content rest on the user. The repository has no releases, carries no licence of its own, and its last push was on 2026-08-17, so freshness depends on the sync script's continued operation.

The deeper questions are worth naming for anyone mirroring similar ecosystems. A public CDN is public, but public access and redistribution permission are different things, and this project's own notice concedes the distinction by restricting commercial reuse. Researchers using the corpus for analysis of skill packaging conventions are on the firmest ground; anyone considering shipping these packages onward should read the subdirectory licences and, where enterprise use is involved, talk to the rights holders the notice names.

Against official markets and awesome lists

The official WorkBuddy marketplace is where these packages are meant to be consumed, with search, updates and trust handled for you; the archive's advantage is cross-package study, offline access and the ability to diff implementations, none of which the market offers. Community awesome lists in the agent ecosystem curate links and descriptions, but rarely mirror full package trees with connectors and plugin manifests side by side from a single vendor's ecosystem.

That is the archive's actual niche: a snapshot corpus of one vendor's agentic marketplace, kept current by a script, indexed by a catalog, and wrapped in an unusually clear notice about what it is and is not for. As a research instrument it is genuinely useful; as a distribution channel it declines the role itself, which is exactly the honesty that keeps such projects out of trouble.

Editorial conclusion

workbuddyskills fits researchers and skill authors who want to study how a production Chinese AI-assistant marketplace packages skills, MCP connectors and expert agents, offline and in bulk, with the explicit understanding that content belongs to its original authors and commercial redistribution is off the table. Skip it if you want a licensed corpus, a guarantee of freshness, or packages to ship in a product. Verify its usefulness in ten minutes: read CATALOG.md, open three SKILL.md files from different categories, and decide whether the conventions you find are worth mirroring into your own skill design.

Frequently asked questions

Official sources

1. infometa/workbuddyskills on GitHub
2. Issues
3. README