llm-serving-auto-benchmark
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a result comparison workflow that ingests data from external sources, creating a potential surface for indirect prompt injection.
- Ingestion points: The
scripts/compare_benchmark_results.pyscript reads benchmark metrics and metadata from user-supplied JSONL files (--input) that may contain data derived from external model responses or logs. - Boundary markers: There are no explicit instructions or delimiters used when the agent processes the generated markdown summary to prevent the adoption of malicious instructions embedded in the benchmark results.
- Capability inventory: The skill provides a wide range of shell command templates for Docker and Python, as well as instructions for managing environment variables (e.g.,
docker exec,docker run) which could be exploited if an agent is influenced by malicious data. - Sanitization: The comparison script parses input using standard JSON and CSV libraries but lacks specific sanitization or validation of the text content before rendering it into the summary report for agent consumption.
Audit Metadata