The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests untrusted data from external sources (databases and files) and processes it to generate SQL, natural language summaries, and HTML visualizations.
Ingestion points: External database queries (MySQL, PostgreSQL, Snowflake, etc.) and file uploads (Excel, JSON, CSV) as described in references/data-sources.md.
Boundary markers: None explicitly defined in the provided files to separate untrusted data from agent instructions.
Capability inventory: The agent can execute SQL queries, read local files using Python (pandas), and write HTML files to the local file system ({domain}_chart.html).
Sanitization: While references/data-sources.md mentions masking privacy fields and read-only SQL constraints, there is no evidence of sanitization for natural language instructions that might be embedded in the data itself.
[DATA_EXFILTRATION]: The skill requires database credentials (Host, Port, Username, Password) as documented in references/data-sources.md. While it suggests read-only access, the agent's ability to execute network-based database queries and potentially write findings to files creates a data exposure risk if the environment is not properly scoped.
[COMMAND_EXECUTION]: The skill utilizes Python snippets (specifically pandas) for file processing. If the agent implements these snippets by executing code on the user's behalf, there is a risk of command injection or unauthorized file system access depending on the execution environment's restrictions.

data-analyst-visualization