Data Platform
Owns repo-local data-platform guardrails. These references are self-contained for local safety and workflow rules; broad provider skill packs are reference-only unless deliberately promoted into the active runtime bundle.
1. Scope
- BigQuery cost-aware query patterns, GoogleSQL guardrails, table design, and slot usage checks.
- Databricks Queries API, VARIANT/JSON patterns, dashboard API notes, dbt integration, and Jupyter kernel execution.
- Local dbt target setup and Databricks SQL dialect caveats.
- Restricted BigQuery/dbt safety workflows that keep local runs out of production schemas.
- Cloud authentication workflows that must run through a user-authenticated pane instead of the agent pane.
Out of scope:
- Agent harness runtime, Home Manager agent config, hooks, postman routing, or
installed agent outputs; use
dotfiles. - GitHub issue, PR, review, or public-surface mechanics; use
collaboration. - Diagram authoring or export workflows; use
diagramming. - Generic Bash, Python, Nix, Markdown, or implementation-loop work; use
programming.
2. Workflow
- Inspect the relevant files, current repo conventions, and
git status. - Select the focused reference below before changing files or running data commands.
- For any command that can write cloud or warehouse state, verify the target, schema, and approval requirements before execution.
- Keep repo-local safety constraints in focused references and keep the owner skill as the primary trigger surface.
- Run the fastest focused check during iteration, then the nearest repo validation surface before reporting success.