The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies a significant attack surface by providing tools and documentation for processing external data such as SQL queries, JSON/CSV files, and user messages. However, it also includes comprehensive governance and safety references (e.g., references/agent-governance-safety.md) that detail multi-layered defenses, such as tagger pre-filters, instruction isolation using XML tags, and mandatory user confirmation for mutating operations. This reflects a security-conscious design intended to educate and provide utility for engineering tasks.
[DYNAMIC_EXECUTION]: The script scripts/pipeline_orchestrator.py contains calls to the Python compile() function. These calls are used neutrally within validation methods to ensure that the code generated by the utility (such as Airflow DAGs or Prefect flows) contains no syntax errors before being saved. This is a common and safe pattern for development-oriented code generation tools.
[DYNAMIC_EXECUTION]: In scripts/inference_optimizer.py, the torch.load() function is used to analyze PyTorch model checkpoints. Although torch.load() can be exploited via the pickle module if used on untrusted data, in the context of this performance analysis and benchmarking tool, it is a standard and necessary operation for the user to provide their own models for optimization.

ai-engineering