The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands using the nao CLI, incorporating user-provided inputs such as workbook names and project titles. If the agent does not properly quote or sanitize these strings before shell invocation, it could create a surface for command injection.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and parses external data exported from Tableau, including workbook metadata, visualization shelving configurations, and CSV data. This data is used to dynamically generate SQL queries and construct prompts for the ask_nao tool. Maliciously crafted Tableau content could attempt to influence the agent's downstream actions or the logic of generated queries.
Ingestion points: Data enters the context from temporary JSON metadata and CSV files (e.g., data_path) generated by the nao migrate tableau command.
Boundary markers: The instructions require verification of JSON version and success fields and explicitly direct the agent to prioritize exported images as the authoritative source of truth for visual state, which serves as a cross-check against metadata poisoning.
Capability inventory: The agent can execute shell commands, run SQL queries via execute_sql, and create persistent charts and stories.
Sanitization: The skill directs the agent to alias SQL columns to safe snake-case identifiers and use human-readable labels to prevent malformed query keys.

migrate-tableau-workbook