The Agent Skills Directory

[COMMAND_EXECUTION]: Utilizes curl to perform API operations against Lytics endpoints for managing audience segments, including listing, creating, and deleting resources.
[EXTERNAL_DOWNLOADS]: Communicates with api.lytics.io to retrieve audience composition data, field schema information, and ML model summaries. These interactions are directed at official vendor infrastructure.
[CREDENTIALS_UNSAFE]: Accesses authentication tokens from environment variables (LYTICS_API_TOKEN) and a local configuration file located at ~/.lytics/accounts.toml. These are documented as the standard methods for managing account access for the vendor's CLI tools.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it translates user-provided natural language into FilterQL queries and segment metadata.
Ingestion points: User descriptions of audiences and business goals ingested in build.md and advise.md.
Boundary markers: Employs a mandatory 'Confirmation Gate' (references/confirmation-gate.md) which requires the user to review and explicitly approve the generated FilterQL and JSON payload before any write operations are executed.
Capability inventory: Performs network operations (GET, POST, PUT, DELETE) via shell commands and reads local account configuration files.
Sanitization: Includes explicit instructions to URL-encode all user-supplied values used in API paths or query parameters.

lytics-audiences