club-3090-llm-serving
Fail
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to clone a repository from an untrusted GitHub account (
noonghunna/club-3090.git). - [REMOTE_CODE_EXECUTION]: The skill executes multiple bash scripts directly after downloading them from an untrusted source, including
setup.sh,launch.sh,switch.sh,health.sh,verify.sh, andbench.sh. - [INDIRECT_PROMPT_INJECTION]: The skill contains a pattern for reading local documents into the LLM context, which represents a vulnerability surface for instructions embedded in external data.
- Ingestion points: The skill reads user-provided files like
large_codebase.txtinto the model prompt. - Boundary markers: There are no delimiters or explicit instructions to ignore embedded commands within the ingested text.
- Capability inventory: The skill environment has capabilities for shell script execution and Docker container management.
- Sanitization: There is no evidence of sanitization or escaping of the file content before it is processed by the LLM.
Recommendations
- AI detected serious security threats
Audit Metadata