Eight hundred and ninety-six archived packages, no licence, and a daily sync six weeks overdue

Eight hundred and ninety-six packages, no licence file, and a notice naming the rights holder

The five counts in the table add to 896: 295 skill packages, 103 connectors, 410 expert plugins, 57 in the official plugin market and 31 in the team plugin market. The top level of the repository holds a gitignore, one catalogue file, the readme and five content directories. There is no licence file, and the licence field resolves to unknown. The readme opens, before anything else, with a copyright notice stating that the content comes from the vendors' public content delivery network and marketplace packages, that copyright belongs to the original authors and to the vendor's related products, and that the repository is a study archive that should not be used for unauthorised commercial redistribution. So the position is neither open source nor silent: it is a third party's assertion of ownership over an archive of 896 packages.

The licence section sends you into subdirectories to find out what applies

The licensing section has three bullets and none of them grants anything. The first says third-party skill and expert packages may each carry their own licence, and to look in the corresponding subdirectory. That means the terms for 896 packages are resolved one at a time by whoever opens each folder, and the archive offers no aggregate statement of what those terms are or how many of them differ. The second bullet is a staleness disclaimer: the archive may differ from the current official version, and the official marketplace is authoritative. The third puts the consequences of using anything here on the person using it. Combined with the absence of a root licence, the practical effect is that a reader cannot answer whether a given package may be used, modified or shipped without visiting it.

A daily sync that the last commit does not support

The maintenance section is one sentence: an automation script checks the vendors' public marketplaces for skills, connectors, experts and plugins every day and synchronises the results. The default branch was last pushed on 2026-08-17. Today is 2026-10-02. That is a gap of forty-six days against a stated cadence of one day, and the repository is not archived. The script that would do the checking is not in the tree either. The top level has the catalogue, the readme, a gitignore and five content directories, and nothing that looks like a job, a workflow or a cron definition. So either the automation stopped, or it runs somewhere else, or the section is aspirational. Either way the catalogue should be read as a snapshot with a date on it, which the staleness bullet already implies.

Two of the five counted things are not directories, and two counts measure different objects

The structure section lists two content directories by name, one for skills and one for connectors, one for experts, and two plugin subdirectories nested inside a plugins directory. The counts table then reports five numbers against five labels, and only three of those labels correspond to a directory in the structure. The official and team plugin markets are counted without being located. And the two large numbers are not the same kind of thing: the skill row carries a qualifier in the table itself, counting only top-level packages that contain a skill document, while the expert row counts expert plugins that the structure says each contain agents, skills and avatars. So 295 and 410 cannot be compared, and anyone quoting a growth rate between them would be dividing two different units.

Ten skill categories sum exactly, and two of them hold five items between them

The skill breakdown has ten categories and its figures add to precisely the skill total, which is a good sign for the catalogue's arithmetic. The shape of it is the story. AI and agent tooling is 159, more than half the catalogue. The vendors' own messaging and enterprise products take 47, and documents and collaboration 44. Then the tail: search and research 15, design and maps 12, data and finance 6, content and marketing 5, cloud and deployment 4, development and engineering 2, and a miscellaneous bucket holding 1. So a marketplace of 295 agent skill packs contains two about writing software and one that fits nowhere else. The expert breakdown is the mirror image: fourteen categories that also sum exactly to their total, with no miscellaneous bucket and one category dedicated to the vendor taking 29, inside the repository that carries that vendor's copyright notice.

Copying back gives you templates, and connectors are keyed by id while skills are keyed by name

Three destination paths are given for putting content back into the product. Skills go into a directory keyed by name, experts into a marketplace directory also keyed by name, and the connector marketplace into a directory keyed by id. The mismatch in the key is small but it means you cannot derive one install path from another. Then a single sentence states the limit that matters: enabling a connector still requires completing an authorisation flow or a scan inside the application, because the repository holds configuration templates only. So the archive is complete for reading and partial for use, and the connector third is the part that cannot be made to work from the files on disk no matter how they are copied.

Three of the five source locations are truncated or tied to one bucket

The data sources list five locations, and they are not uniform. Skills come as a single archive from a product download domain. Plugins come from the same domain as a wildcard over several archives. Connectors come from a raw cloud object-storage URL whose hostname embeds a numeric account identifier and a region, which is a bucket path rather than a stable product endpoint. The two expert entries give truncated paths that begin with an ellipsis, so their host is not stated anywhere in the file. And the expert bundles are one archive per expert, which means refreshing a single expert is a separate download from refreshing the catalogue, and a refresh of the whole expert set is four hundred of them.

The example commands pick a messaging integration and a stock-market connector

# 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

The quick browse block is the most concrete part of the file and it is five commands long. It lists the skills directory, opens one skill's document, and that skill is named for a Lark integration, which is the vendors' enterprise messaging product and the same subject as the second-largest skill category. It then lists the agents directory inside one expert and prints that expert's plugin manifest, and the manifest lives in a hidden directory beside the agents rather than beside the expert. Finally it lists the connectors directory and opens one connector's configuration template, and that connector is named for stock market data. Nothing else in the readme is that specific, which makes the block the fastest way to understand the layout.

Editorial conclusion

This archive is worth cloning if you want to read how a large vendor's marketplace writes its skill, connector and expert packages without opening the product, because everything is text, the counts are broken down by category, and one catalogue file carries each package's purpose, its prerequisites and a link. It is not a redistribution channel: there is no licence at the root, the notice claims the copyright belongs to the original authors and the vendor, and the terms for individual packages live in subdirectories you would have to visit eight hundred times. It is also not a way to make the connectors work, since activation needs an authorisation flow inside the application. Before you rely on it, check whether the daily sync described in the maintenance section is still running, because the last commit and the stated cadence disagree by more than six weeks, and treat the catalogue as a snapshot rather than as the current state.

Frequently asked questions

Official sources

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