The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from local project configuration files and CLI outputs to generate environment plans.
Ingestion points: The skill reads pyproject.toml, uv.lock, and the JSON output from various databricks CLI commands.
Boundary markers: Uses structured --output json and a distinct dry-run phase to separate plan generation from application.
Capability inventory: Writing and backing up pyproject.toml, creating .venv directories, and executing shell commands via databricks and uv (SKILL.md, references/examples.md).
Sanitization: Instructions require the agent to display the plan, unified diffs, and warnings to the user for explicit approval before any mutations occur.
[COMMAND_EXECUTION]: The skill executes shell commands for environment provisioning and diagnostic tasks. This is restricted to standard Databricks CLI and uv operations, with instructions specifically forbidding the use of remote installers or silent background installations.
[DATA_EXFILTRATION]: The troubleshooting guidelines provide a specific protocol for external defect reporting, mandating the redaction of sensitive identifiers such as authentication tokens, workspace hosts, cluster names, and usernames to prevent accidental data exposure.

databricks-setup-local