The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches Quickstart markdown content and file listings from official GitHub repositories under the Snowflake-Labs organization (e.g., sfquickstarts and sfguide-*). These downloads are targeted at trusted vendor-owned infrastructure.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform git clone on companion repositories when a user opts to download tutorial files locally.
[DYNAMIC_EXECUTION]: The skill extracts code blocks (SQL, Python, YAML) from the fetched markdown files and executes them using snowflake_sql_execute (in Builder mode) or writes them to the local file system for user execution (in Learner mode).
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect injection surface because it fetches external content and treats it as a source of truth for both instructions and executable code.
Ingestion points: Fetches raw markdown and companion repository files from GitHub based on user-provided or slug-derived URLs (SKILL.md, Step 2).
Boundary markers: No explicit delimiters are mentioned to prevent the LLM from following malicious instructions that might be embedded in the fetched markdown.
Capability inventory: The agent has access to snowflake_sql_execute, Bash, and Write tools, which could be abused if the fetched content were compromised.
Sanitization: While the skill performs diagnostics for environment readiness and checks for deprecated syntax, it does not explicitly sanitize the fetched markdown for adversarial prompt injection patterns.

quickstart-guide