The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill requires adding an untrusted external MCP server endpoint (https://rube.app/mcp).
Evidence: The 'Setup' section instructs users to add this URL as an MCP server endpoint.
Risk: This endpoint is not from a trusted source and the service logic is opaque, creating a dependency on an unverified third party.
PROMPT_INJECTION (HIGH): High-risk Indirect Prompt Injection vulnerability surface (Category 8).
Ingestion points: Data retrieved from Timely (project names, task descriptions, notes) via tool outputs.
Capability inventory: RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH provide the ability to modify data or execute operations in Timely.
Boundary markers: None identified in the prompts; the agent is simply told to use discovered schemas.
Sanitization: No sanitization or validation of the data retrieved from Timely is mentioned before it is used to influence further tool calls.
Risk: An attacker could place malicious instructions in a Timely task description which, when read by the agent, could trigger unauthorized actions (e.g., deleting projects or exfiltrating data).
PROMPT_INJECTION (MEDIUM): The skill instructs the agent to follow 'recommended execution plans' returned by the remote RUBE_SEARCH_TOOLS call. This allows the remote service to influence the agent's behavioral logic dynamically.

timely-automation