Turn your experience into a Skill

Once a type of task is proven to work and your team will run it repeatedly, capture the steps as a Skill instead of re-describing them in every chat.

Option 1: Do it manually in the Lovrabet workspace

Step1. Create a Skill

Runtime: Go to Chat → Skill Library → + New Skill, fill in the name and skillCode, and write the Markdown steps.

Workspace: Go to App Management → AI Asset Center → Skills → New Skill, fill in the skillCode (English snake_case recommended) and the body, then save.

Step2. Capture from a successful session (optional)

1. Complete a multi-step flow you're happy with in a chat.
2. If the Save as Skill prompt appears → confirm.
3. Go back to the Skills list to flesh out and revise the Markdown (system-generated drafts usually need a human to refine the metric definitions and prohibitions).

Step3. Roll it out to the team

• Add the skillCode to the related Knowledges entries or internal how-to docs;
• In chats, reference it uniformly as use skillCode=xxx so everyone's wording stays consistent.

Option 2: Have an Agent verify via the Rabetbase CLI (for developers, optional)

When developers maintain the dataset fields a Skill depends on, they can hand verification to an external Agent. This is not the everyday path for business users.

Action	CLI the Agent can call
Inspect dataset structure	dataset detail
Trial-run a data query	data filter

Example prompt

请用 rabetbase skill 检查 skillCode=weekly_report 依赖的数据集 order 字段是否齐全，并给出 Skill 正文修改建议。

FAQ

Should write operations go into a Skill?

Yes — and they must explicitly require a confirmation card or human sign-off, so the Agent never writes back to production data on its own.

Any naming advice for skillCodes?

Keep them globally unique and self-explanatory. Bump a version on breaking changes (e.g., _v2), and keep the old Skill read-only for a while.