In this article
Data Science and Engineering Coach
| Field | Value |
|---|---|
| Kind | agent |
| Source | .github/agents/data-science-engineering/data-science-engineering-coach.agent.md |
| Invocation | Selected from the chat agent picker as Data Science and Engineering Coach |
| Interactive | Yes |
What it does
Coach a persistent data science and data engineering workstream through explicit jobs, durable state, routed skill authority, and safe customer-artifact writes.
When to use it
Use Data Science and Engineering Coach when a data scientist or data engineer needs one persistent engagement context across cataloging, feasibility, pipelines, analysis, experiments, tests, and observability. It is especially useful when work will pause, detour into another job, or resume in a later session without losing artifact and gate context.
Select a specialist agent directly when you need only one isolated output, such as a notebook or dashboard, and do not need workstream state, transitions, or durable customer-artifact safety gates.
How to use it
- Select Data Science and Engineering Coach from the agent picker.
- Provide a kebab-case project slug. The coach creates or resumes the project-scoped session state.
- Choose a job from the offered registry. The coach never selects one silently.
- Confirm any proposed transition. Bounded work is paused with its phase and gates, while completed episodic work remains in invocation history.
- Review the scan result before any durable customer-artifact write.
- At completion, choose whether to resume paused work, enrich continuous catalog context, select another job, or close.
Example usage
Start a data workstream for
retail-demand-forecasting. I need to assess feasibility first, but I may need a data model diagram once we understand the sources.
The coach initializes or resumes state, displays the data-science disclaimer
when required, asks you to confirm feasibility, and records that bounded job.
If you later request the diagram, it proposes pausing feasibility, confirms the
transition to model-diagram, runs that episodic job, then offers to resume the
saved feasibility phase rather than advancing automatically.