Consider a unified Lark CLI skill organization similar to an umbrella skill

Motivation

Lark CLI currently exposes many domain-specific AI Agent Skills (IM, Docs, Drive, Calendar, Sheets, Base, Mail, Tasks, Approval, etc.). This is powerful, but it also makes the injected skill list very large in agent environments.

In practice, this can create a few problems:

• Higher baseline token usage before the user even starts the task.
• More cognitive overhead for the agent when selecting among many similar lark-* skills.
• More fragmented discoverability: users need to know whether to ask for lark-doc, lark-drive, lark-sheets, lark-base, lark-calendar, etc.
• Harder long-term maintenance when adding new Lark product domains, because each domain becomes another top-level skill entry.

Proposal

Consider adding, or optionally switching to, a more unified skill organization model similar to an umbrella skill:

• Provide one top-level lark-cli / lark skill as the primary entry point.
• Inside that skill, route to subdomains such as:
  • im
  • contact
  • doc
  • drive
  • wiki
  • calendar
  • vc
  • minutes
  • sheets
  • base
  • mail
  • task
  • approval
  • attendance
  • openapi
• Keep domain-specific instructions in referenced files or lazily loaded sections, instead of injecting every domain skill description into the initial agent context.
• Optionally keep the existing lark-* skills for backward compatibility, but recommend the unified entry point for users who care about context size and simpler agent routing.

Expected benefits

• Lower default prompt/token overhead in agent environments.
• Simpler skill discovery: users can think in terms of "use Lark CLI" instead of choosing from many top-level Lark skills.
• More predictable routing for agents: one Lark entry point can perform intent classification and then load the appropriate domain workflow.
• Easier extensibility as more Lark product capabilities are added.

Compatibility suggestion

This could be introduced non-disruptively:

1. Add a new unified lark-cli skill.
2. Keep existing lark-* skills unchanged for advanced/direct use.
3. Document that agent users may install/use the unified skill when they want a smaller top-level skill surface.
4. Gradually move shared guidance (auth, identity switching, permission handling, common CLI conventions) into shared references used by both the unified skill and domain skills.

Thanks for building and open-sourcing Lark CLI. The current domain-specific skills are already very useful; this proposal is mainly about improving token efficiency and agent ergonomics as the skill surface grows.

Start by reviewing the existing domain-specific AI Agent Skills and the shared guidance around auth, identity switching, permissions, and common CLI conventions. Compare their current top-level organization with the proposed unified lark-cli/lark entry point; done means a decided organization, documented compatibility behavior, and an agreed approach for lazy or referenced domain guidance.