uniteperformance-web
Warn
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches content, configuration files, and tool definitions from https://new.uniteperformance.no and https://dev.uniteperformance.no.
- [DYNAMIC_EXECUTION]: Instructs the agent to initialize a Model Context Protocol (MCP) endpoint at /mcp on remote sites to dynamically discover (tools/list) and execute (tools/call) functionality at runtime.
- [DYNAMIC_EXECUTION]: Defines agent capabilities based on external OpenAPI specifications (/openapi.json) and tool inventories (/api/public/tools) retrieved from remote sources.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from multiple web endpoints which could influence agent behavior.
- [INDIRECT_PROMPT_INJECTION]: Ingestion points: Reads from subpaths like /llms.txt, /pricing.md, and tool responses on uniteperformance.no.
- [INDIRECT_PROMPT_INJECTION]: Boundary markers: No explicit delimiters are defined to separate external data from system instructions.
- [INDIRECT_PROMPT_INJECTION]: Capability inventory: Uses MCP tools for remote content retrieval and search.
- [INDIRECT_PROMPT_INJECTION]: Sanitization: The skill relies on natural language instructions for citation and link verification rather than technical sanitization.
- [COMMAND_EXECUTION]: Instructions to call tools via tools/call represent the execution of remote logic provided by the target websites.
Audit Metadata