local-llm-setup
Fail
Audited by Gen Agent Trust Hub on Aug 9, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill description states that it automates the download of
llama-serverbinaries and Large Language Model (LLM) weights (e.g.,gemma-4-12b-UD-Q4_K_XL.gguf). These downloads originate from external sources that are not verified against hashes or trusted origins within the provided file. - [COMMAND_EXECUTION]: The skill establishes persistence by installing a routing proxy as a system daemon using platform-specific service managers like
launchd,systemd, orNSSM. This modification of system initialization and the creation of persistent background processes generally require administrative privileges. - [COMMAND_EXECUTION]: Core functionality depends on the execution of multiple local Python scripts (
scripts/run_server.py,scripts/run_agent.py,scripts/enable_global_routing.py) to manage the inference stack and handle task routing. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. The
run_agent.pyscript ingests untrusted content from external files (e.g.,target.py) and passes it to an LLM without specified boundary markers, sanitization, or instructions to ignore embedded commands. This could allow an attacker to influence the agent's behavior by embedding malicious instructions in files being analyzed.
Recommendations
- AI detected serious security threats
Audit Metadata