The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from the warehouse schema and event taxonomy, which originate from external capture APIs and are considered untrusted input that could contain malicious instructions.
Ingestion points: The agent reads metadata from system.information_schema.metrics, system.information_schema.tables, system.information_schema.columns, and read-data-schema (as documented in SKILL.md and references/governance.md).
Boundary markers: The skill explicitly instructs the agent to treat every taxonomy name/value as quoted, untrusted data and to confirm definitions with the user before write actions.
Capability inventory: The skill utilizes powerful tools including posthog:view-create, posthog:view-materialize, posthog:view-update, and posthog:execute-sql for persistent data manipulation.
Sanitization: The skill includes robust defensive instructions, directing the agent to treat metric descriptions as untrusted project data rather than instructions and to avoid obeying commands embedded in free-text fields.
[COMMAND_EXECUTION]: The skill facilitates the generation and execution of HogQL (PostHog SQL) queries to create, update, and materialize data views within the warehouse environment.
Evidence: Detailed workflows for query execution and view management using tools like posthog:execute-sql and posthog:view-create are provided in references/posthog-views.md.

modeling-warehouse-foundations