The Agent Skills Directory

Local File Management and Configuration: The skill manages local state files (e.g., plugin-install-suppressions.json and automation.toml) and semantic layer artifacts within designated user directories (e.g., $CODEX_HOME/skills/). This is standard behavior for a skill designed to create and persist data context locally.
Controlled Command Execution: The skill includes Python scripts and test files that utilize subprocess.run to execute internal helper scripts. These executions are limited to the skill's own components and are used for preflight checks and state management, representing a managed approach to local task execution.
Automation and External Integration: The skill sets up 'heartbeat' and 'cron' automations to poll data sources like Databricks, BigQuery, Snowflake, and GitHub. These integrations follow standard patterns for data analytics tools and include explicit guidelines to avoid handling sensitive literals or credentials during these operations.
Data Handling Practices: The instructions include clear directives to avoid copying sensitive data, credentials, or row-level examples into generated skill files, prioritizing the storage of semantic metadata and source links instead.

create-data-context