The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to ingest and process metadata, such as cube definitions, segment names, and descriptions, from the RevOS MCP server using the api_read tool. This content originates from external organizational data sources and is processed to inform subsequent agent actions or displayed to the user.
Ingestion points: The agent retrieves semantic model metadata and existing segment configurations from the cubes and segments resources (SKILL.md).
Boundary markers: There are no instructions defining the use of output delimiters or specific instructions for the agent to ignore potentially malicious content embedded within these metadata strings when they are processed.
Capability inventory: The skill allows for significant state changes, including api_write (create/update) and api_delete operations, which are driven by the agent's interpretation of the ingested data (SKILL.md).
Sanitization: The skill lacks explicit directives for sanitizing or escaping retrieved strings before they are used in prompts or forwarded to other API operations.

segments