Author a skill
A skill is a reusable analytical capability that Athena composes into broader workflows. Where an AI Assistant query is a one-off question, a skill is the encapsulated logic for a recurring analysis — a margin-leak detector, a coverage-risk scorer, a customer-segmentation classifier — that any user, any agent, or any scheduled workflow can invoke against your data.
This procedure walks through authoring a skill in AI Studio. Skills are the building blocks the Data Engineer persona composes into autonomous Employee-mode agents.
⚠ Procedure steps below need UI capture against the staging platform. The shape is in place; the actual click-paths and screenshots are TBD.
Before you start
- A ConverSight account with the Data Engineer role.
- A clear definition of what the skill does — input parameters, expected output, the decision the skill supports.
- A subject area the skill will operate against, already configured.
Steps
- Step one — what to click and what to see.
- Step two.
- Step three.
Verify
The skill is invokable by name from Athena ("Run my margin-leak skill on Q3"), shows up in the AI Studio skill catalog, and produces the expected output against test data. Audit trail captures every invocation.
Tips for skill design
- One skill, one job. Skills compose well when each does one thing cleanly. "Detect margin leak" is a skill; "Detect margin leak and email the finance team" is two.
- Parameterize at the right level. Hard-code the things that won't change for your org (e.g., a margin-threshold formula); parameterize the things callers want to vary (e.g., the time window).
- Log structured output. Skills that emit structured results (not just narrative) compose more easily into agents and Storyboards.
- Test before publishing. Run against historical data; verify the skill behaves the same way on known-good and known-bad inputs.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| Skill returns nothing for valid input | Subject-area scope mismatch | Verify the skill is reading the subject area you intended |
| Inconsistent results across users | RLS differences mean different users see different rows | Decide whether the skill should run with delegating-user scope or service-account scope |
| Skill works manually but agents can't invoke it | Authorization gap in the agent's role | Add the skill to the agent's allowed-skills list in Users & Roles |
Related
- Feature: AI Studio — the agent + skill build environment
- Feature: Marketplace — publish your skill to a wider audience
- Persona: Data Engineer