The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines analytical queries by ingesting metadata from external dbt repositories (via the gh CLI) and database column comments. If an attacker controls these sources, they could potentially influence the agent's SQL generation logic.
Ingestion points: Fetches dbt manifests and data dictionary files from remote GitHub repositories; reads column comments from system.columns via SQL.
Boundary markers: The skill requires an 'Intent' block with manual confirmation of metrics, populations, and time windows before executing any data queries.
Capability inventory: Executes shell commands via clickhousectl and curl; performs SQL queries against local and cloud ClickHouse instances.
Sanitization: Implements a client-side regex check to prevent the execution of destructive SQL tokens (e.g., DROP, TRUNCATE, ALTER) when using the Query API path.
[DYNAMIC_EXECUTION]: The chdb-sql reference documentation describes an 'AI-Assisted SQL' feature (chdb.ask) which generates and executes SQL from natural language strings. This introduces the risk of generating unintended or malicious queries if the natural language input is compromised.
[EXTERNAL_DOWNLOADS]: The skill automates the installation of CLI tools and Python packages from official ClickHouse and community sources. These are well-known technology services and are documented neutrally.
Downloads the ClickHouse CLI installer from clickhouse.com/cli and pipes it to a shell.
Recommends the installation of the chdb Python package and various plotting libraries from official registries.
[COMMAND_EXECUTION]: Multiple sub-skills utilize shell commands to manage local ClickHouse servers, perform cloud migrations, and interact with the gh CLI for documentation retrieval. These actions are transparently documented and aligned with the skill's analytical purpose.

data-analyst