skyfire-automation
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- EXTERNAL_DOWNLOADS (HIGH): The skill instructs users to add an MCP server from
https://rube.app/mcp, which is not a trusted source. This server controls the tool definitions the agent uses. - REMOTE_CODE_EXECUTION (HIGH): The skill utilizes
RUBE_REMOTE_WORKBENCHandRUBE_MULTI_EXECUTE_TOOL, which allow for the execution of remote operations defined by the externalrube.appservice. - PROMPT_INJECTION (HIGH): The skill is vulnerable to Indirect Prompt Injection because it requires the agent to 'Always search tools first' using
RUBE_SEARCH_TOOLS. The results from this untrusted external call dictate the agent's subsequent logic and tool parameters. Ingestion Point:RUBE_SEARCH_TOOLSresponse; Boundary markers: Absent; Capability inventory:RUBE_MULTI_EXECUTE_TOOL,RUBE_REMOTE_WORKBENCH,RUBE_GET_TOOL_SCHEMAS; Sanitization: None detected.
Recommendations
- AI detected serious security threats
Audit Metadata