The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill utilizes dynamic tool discovery via RUBE_SEARCH_TOOLS to determine execution plans and input schemas at runtime. This creates a surface where a compromised or malicious remote server could influence agent behavior.
Ingestion points: Output from RUBE_SEARCH_TOOLS and RUBE_GET_TOOL_SCHEMAS (external API).
Boundary markers: Absent; the instructions direct the agent to follow returned schemas and execution plans without explicit sanitization steps.
Capability inventory: RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH (arbitrary tool execution based on remote data).
Sanitization: Not present; the skill assumes trust in the tool definitions provided by the MCP endpoint.
[Unverifiable Dependency] (LOW): The skill requires connection to an external MCP server (https://rube.app/mcp) which is not on the established trusted source list. Per the risk assessment for the skill's primary purpose, this is considered low risk but noteworthy as the agent depends on this external domain for its core logic.

simplesat-automation