mashup-mods
Warn
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONPRIVILEGE_ESCALATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references dozens of third-party GitHub repositories in references/mashup-cases.md (e.g., SkyCraft, LibertyCraft, IW4L) and instructs the agent to read their READMEs and design documents to guide the creation of new mods. These sources are not from trusted organizations and are controlled by individual users. \n- [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by directing the agent to ingest and follow instructions found in external README files, design documents, and game wikis. There are no boundary markers or sanitization protocols mentioned to prevent the agent from being influenced by malicious instructions embedded in these external sources. \n- [DYNAMIC_EXECUTION]: The instructions promote the use of 'converter scripts' to process game assets at runtime or installation and suggest techniques like 'proxy DLLs' and 'DLL injection' for host game modifications. These methods involve the execution or loading of binary content that can be difficult to verify statically. \n- [COMMAND_EXECUTION]: The 'Passthrough' pattern (Pattern 2) relies on inter-process communication (IPC) via shared memory, named pipes, and local network sockets (127.0.0.1), which could be leveraged for unauthorized data transfer between processes. \n- [PRIVILEGE_ESCALATION]: The skill advocates for host-side plugins that perform memory hooking and injection into game processes (e.g., SKSE, REFramework), which are high-privilege operations typically requiring administrative access or bypass of process protections.
Audit Metadata