The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external Metabase API endpoints, such as question summaries, dashboard metadata, and database schemas, which are interpolated into the agent's context. \n
Ingestion points: Data enters the context through discovery commands like card, dashboard, tables, and search in scripts/lib/commands/discover.mjs. \n
Boundary markers: The skill instructions in SKILL.md define 'Context Safety Rules' to limit the amount of data loaded, though they do not include explicit 'ignore instructions' delimiters for all external data. \n
Capability inventory: The skill can write JSON and SQL files to the project directory (scripts/lib/client.mjs) and perform authenticated network requests to the user's Metabase instance (scripts/lib/client.mjs). \n
Sanitization: Data is processed as structured JSON before being summarized for the agent, reducing the risk of accidental command execution. \n- [DATA_EXPOSURE]: The skill handles sensitive Metabase API keys required for authentication. \n
Secret Management: Credentials are stored in ~/.metabase-skill/config.json. The skill implements security best practices by recommending environment variable usage and explicitly setting file permissions to 0600 (read/write by owner only) via scripts/lib/client.mjs. \n
Context Protection: SKILL.md includes explicit instructions for the agent to never print API keys or read the configuration file directly.

metabase