23 Lark Skills Exist Because Developers Want Agents Inside Their Workflows, Not Alongside Them
Lark has 23 agent skills covering documents, calendar, IM, spreadsheets, approvals, and OKRs. The breadth is telling.
Most workspace tool integrations take the same shape: there's an API, there's an AI wrapper, and the AI can read and write documents through it.
Lark has 23 agent skills.
That breadth — covering Lark Documents, calendar, instant messaging, spreadsheets, approvals, OKRs, and more — is the thing worth paying attention to. Not any individual skill. The 23.
Why Lark
Lark (called Feishu in China) is the default workspace platform for a significant number of fast-moving Chinese tech companies — ByteDance built it internally before spinning it out, and it spread from there across the ecosystem of companies ByteDance influenced or competed with. If you're a developer at a company in that orbit, Lark isn't a tool you chose. It's the tool your company runs on.
The Lark Documents skill from larksuite/cli has 83.3K installs. That's one skill in a suite of 23. The cumulative install count across the suite is something larger. The developers installing these skills have a specific problem: they want their AI agents to function inside the same workspace where work actually happens, not just in a coding environment that's adjacent to it.
The Difference Between "Alongside" and "Inside"
This distinction matters more than it sounds.
An agent that is alongside your workflow can answer questions about things you tell it. An agent that is inside your workflow can act on the context that already exists there — the doc you wrote last week, the approval decision that was made yesterday, the OKR target your team committed to in Q1.
Most AI coding assistance is alongside. You open the AI tool, you describe the context, the AI helps. The context transfer is manual. You explain what exists before the agent can use it.
The Lark skills bundle changes that relationship. An agent with the Lark Documents skill can retrieve the spec you wrote in Lark without you having to copy-paste it. With the approvals skill, it can check whether a decision has been made. With the OKR skill, it can surface what the team is trying to achieve and use that to evaluate whether what you're building is aligned.
The agent becomes a participant in the workspace, not a tool you consult outside of it.
23 Skills as a Design Philosophy
The fact that larksuite built 23 skills — not one — reflects something about how they understand agent integration.
A single "Lark integration" skill would be a lowest common denominator: connect to the API, read documents, write documents. Useful enough to demonstrate the capability. Not deep enough to change how developers work.
23 skills means: we think there's a distinct enough workflow for documents that it warrants its own encoding. And for calendar, another distinct workflow. And for approvals, another. Each skill is the answer to "how should an agent behave when operating in this specific part of Lark?"
That granularity is what makes the agent actually useful in each context, rather than generically capable across all of them. A document isn't a spreadsheet isn't a calendar entry, and the conventions for working with each are different enough that a single skill would have to be vague where multiple skills can be specific.
The Pattern It Points To
The Lark suite is the clearest example in the skills ecosystem of a workspace platform treating agent integration as a first-class investment, not an afterthought.
The investment is substantial: 23 skills, each encoding the patterns for a specific part of the platform. That's a commitment to the thesis that agents should operate inside workflows, not alongside them.
The developers who installed these skills voted with their install counts for the same thesis. They didn't want an agent that could theoretically access Lark if they set it up correctly. They wanted an agent that knows how Lark works and uses it the way Lark is meant to be used.
83.3K installs on Documents alone suggests they found what they were looking for.
Part of the AI Skill Daily series — skills worth understanding, one at a time.