dbskill: a Chinese-language AI Skills toolbox for business diagnosis and content work

What dbskill actually solves, and for whom

The README frames the problem as not knowing which tool to reach for. A founder with a pricing complaint, a creator with a topic that will not hold attention, or someone who knows the next action and keeps not taking it, all face the same first question: what should I even ask? dbskill answers that with a single entry point, /dbs, which reads the current conversation and decides whether one Skill is enough or whether a main Skill plus up to two supporting Skills should be combined. The stated audience is entrepreneurs and content creators working in Chinese. The repository is explicit that the toolbox is Chinese-first: the README opens with Chinese and links to English, Japanese, Korean and Traditional Chinese translations. The knowledge behind it is described as 4,176 knowledge atoms filtered from 16,152 public tweets, and those atoms are published in the repository rather than kept private. That matters, because it means the Skills are a layer over data you can inspect. If the Skills do not fit your workflow, the underlying atoms are still readable and usable on their own.

The orchestration model: one entry point, at most three Skills per task

The data flow the README draws is short. A real task goes in, /dbs reads the context and judges single versus combined, it produces a prompt you can send onward, the selected Skills deliver one unified result, and you then add feedback and let it re-orchestrate. Two constraints stand out. First, dbskill handles one current task at a time, so it is not a batch pipeline. Second, the combination ceiling is one main Skill plus up to two auxiliaries. That ceiling is a deliberate trade-off: it keeps results coherent, since three Skills writing into one deliverable is already hard to reconcile, but it also means a genuinely multi-domain request gets decomposed across rounds rather than answered in one pass. The README states that a single Skill is kept when it covers the task, and a main-plus-auxiliary combination is used when the task contains independent and necessary requirements. If you already know what you want, you skip the router and call the Skill directly, for example /dbs-diagnosis for a business question or /dbs-hook for a short-video opening. The router is a convenience, not a gate.

Installing dbskill with npx skills and running a first diagnosis

The recommended install path covers Claude Code, Doubao, WorkBuddy, Codex and other agents that support Skills. It is a single npx command, run in a terminal:

npx -y skills add dontbesilent2025/dbskill -g --all

The -g flag installs globally and --all pulls the whole toolbox. After it finishes, you return to your agent and type /dbs 新手入门 to start. Note that the install command pulls the Skills only. The README says the public tweet collection is a separate artefact and is not downloaded by this command.

Once installed, the first real use is the router with a concrete situation. The README gives this example, which you send as a normal message to the agent:

/dbs 我做少儿编程课，已经有 40 个付费学员，但续费率很低。
我需要判断问题出在产品、定价，还是我找错了客户。

What you should see is a short explanation of why the router picked what it picked, followed by a prompt you can send as your next message. After that round, you add new facts or feedback and type /dbs again; it re-judges whether the task now needs one Skill or a combination. If you would rather skip the router, the README lists direct calls such as /dbs-diagnosis, /dbs-content, /dbs-hook, /dbs-benchmark and /dbs-knowledge, each with a sample prompt in the same section.

The Claude Code plugin route and what the dbs namespace changes

Claude Code users have a second install path through the plugin marketplace, and it behaves differently enough to be worth separating. Two commands register the marketplace and install the toolbox:

claude plugin marketplace add dontbesilent2025/dbskill
claude plugin install dbs@dontbesilent-skills

That dbs plugin contains the 32 business Skills plus one system entry, dbs-update. The important difference is namespacing: Claude Code prefixes plugin Skills, so the main entry becomes /dbs:dbs and a specific capability becomes /dbs:dbs-diagnosis. If you copy a slash command from the README's direct-call examples while running the plugin install, it will not match the namespaced form. The marketplace also supports installing a single capability, for example claude plugin install dbs-diagnosis@dontbesilent-skills, which is the lighter option when you want one Skill rather than the whole set. Updates go through the agent instead of the shell: you tell the current agent 更新 dbskill, and the README states this syncs the official dbskill without touching the archives, reports and decision records stored in ~/.dbs/.

Where dbskill stops being the right tool

The most concrete limitation is the licence. The repository uses CC BY-NC 4.0, and the README spells out the consequences: personal use, study, research and non-commercial projects are fine; publishing derivative works requires attribution; commercial use needs separate authorisation from the author. For an engineer evaluating this as a component inside a paid product, that is a hard stop, not a footnote. A second limitation is language. The toolbox is Chinese-first. Translations of the README exist, but the Skills themselves, their prompts and the knowledge atoms are described in Chinese, and the README's own examples are Chinese. If your team does not work in Chinese, the output will not be usable without a translation layer you would have to build. Third, the video extraction Skill has a dependency the README does not resolve. Release v2.18.34 notes that short-video data and transcript extraction were added and that missing credentials trigger purchase and configuration guidance, and v2.18.35 added Xiaohongshu and WeChat Channels data queries with a 15-second wait limit per data request. The README does not document what those credentials are, how to obtain them, or what the 15-second limit does to a batch of videos. Treat /dbs-video-extract as the Skill to validate first if it is why you came. Finally, the README does not document rollback. The update path is described as syncing the official dbskill, and while it says ~/.dbs/ is preserved, there is no stated way to return to a previous Skill version.

How dbskill differs from Baoyu-skills and other Skills collections

The closest comparison in the search data is Baoyu-skills, another collection of Skills for agents. The difference is in what sits underneath. dbskill publishes its source material: 4,176 structured knowledge atoms in 知识库/原子库/atoms.jsonl, methodology documents organised per Skill in 知识库/Skill知识包, and a public tweet collection in books/ in both Markdown and PDF. That means the Skills are traceable to a corpus you can read, and the README explicitly points at atoms.jsonl for anyone who wants to build their own RAG instead of using the Skills at all. A Skills collection that is only prompts gives you no such fallback. The trade-off runs the other way too: dbskill's scope is narrow by design, covering business diagnosis, content production, decision records and folder knowledge bases, all in Chinese. A general-purpose Skills collection will cover more ground in more languages. If your work is Chinese-language business and content, the narrower and deeper option is the better fit. If it is not, the breadth wins.

Maintenance, upgrade cost and licence implications

The repository is not archived, and the last push was on 2026-09-07, which is recent. The release history shows a fast cadence: v2.18.33, v2.18.34 and v2.18.35 all landed on 2026-09-01, and the README badge shows version 2.18.40. That pace is a real cost as well as a benefit. The update path is a single instruction to the agent, which is cheap, but the version-to-version changes are behavioural: v2.18.33 removed /dbs-skill-cleaner along with its review and quarantine entry points, so anything built on that Skill stopped working at that release. Anyone scripting against specific Skill names should pin what they depend on and read the commit log before updating. On licence, CC BY-NC 4.0 is a non-commercial licence. The README states that attribution is required when publishing derivative works and that commercial use requires contacting the author. This is a description of what the project says, not legal advice; if you plan to ship anything built on these Skills, get your own reading of the LICENSE file.

Editorial conclusion

Adopt dbskill if you work in Chinese on business diagnosis, content production or folder-based knowledge management, and you already run Claude Code, Codex, Doubao or WorkBuddy. Do not adopt it if you need a permissively licensed component for a commercial product: the licence is CC BY-NC 4.0 and commercial use requires separate authorisation. Before installing, check the skills/ directory for the exact Skill names your agent exposes, and confirm whether /dbs-video-extract is something you need, since it depends on credentials the README does not document.

Frequently asked questions

Official sources

1. dontbesilent2025/dbskill on GitHub
2. Issues
3. README
4. Releases