The Agent Skills Directory

[SAFE]: The skill instructions focus on utilizing local Python scripts located within the scripts/rag-perf directory of the repository. It uses standard environment management tools like uv to handle dependencies.
[COMMAND_EXECUTION]: The skill facilitates the execution of performance benchmarks through the rag-perf CLI and aiperf subprocesses. These operations are restricted to local development environments and are driven by user-provided configuration files.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface by reading external query files (.jsonl, .csv) and generating synthetic queries via an LLM. While this data is processed and sent to a RAG server, the workflow is a documented part of its benchmarking purpose. The analysis found no evidence of these queries being used in unsafe sinks such as eval() or unsanitized shell commands.
[DATA_EXFILTRATION]: Network operations are confined to local RAG server endpoints (localhost:8081) and LLM generation endpoints (localhost:8999). No unauthorized external data exfiltration patterns were identified.

rag-perf