The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructs the agent to use the ACCOUNTADMIN role in Snowflake if a dedicated learning role is not available. This is the highest privilege level in a Snowflake account, providing full administrative control over the environment. While necessary for some setup tasks in a tutorial context, it represents a privilege escalation pattern.
Evidence: USE ROLE ACCOUNTADMIN; in references/LESSONS.md and SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from two external sources: official Snowflake documentation and sample data stored in S3. This creates a surface for indirect prompt injection if those sources were to be compromised.
Ingestion points: Documentation is fetched via web_fetch from docs.snowflake.com/en/user-guide/dynamic-tables-about (SKILL.md). Data is loaded from s3://sfquickstarts/tastybytes/ via the COPY INTO command (LESSONS.md).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when processing the documentation or sample data.
Capability inventory: The skill has extensive Snowflake environment capabilities, including creating/altering roles, databases, schemas, stages, tasks, and dynamic tables.
Sanitization: There is no evidence of sanitization or validation of the content fetched from the web or loaded from the S3 bucket before it is processed by the agent or the Snowflake environment.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch configuration and data from external Snowflake-owned infrastructure.
Evidence: web_fetch to docs.snowflake.com and COPY INTO from s3://sfquickstarts/tastybytes/. These are official vendor resources for Snowflake.

dynamic-tables-tutorial