logos-router-reasoning

Warn

Audited by Gen Agent Trust Hub on Jul 2, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user to clone a repository from an unverified GitHub account (rak7777/mythic-mcp-proxy) instead of an official or trusted vendor source.\n- [REMOTE_CODE_EXECUTION]: The installation process involves running pip install on unverified third-party code followed by executing the library via python -m logos, which facilitates the execution of unverified scripts on the host system.\n- [PROMPT_INJECTION]: The skill processes untrusted external data such as PDF files and multi-language user prompts, creating an attack surface for indirect prompt injection.\n
  • Ingestion points: Data enters the agent context through router.query and router.create_workflow(\"analyze_research_paper.pdf\").\n
  • Boundary markers: No delimiters or instructions to ignore embedded commands are documented for the reasoning mesh.\n
  • Capability inventory: The skill interacts with local inference engines (VLLM, Ollama) and manages multi-step reasoning workflows.\n
  • Sanitization: There is no evidence of sanitization or validation for content extracted from processed documents.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 2, 2026, 03:09 AM
Security Audit — agent-trust-hub — logos-router-reasoning