prompt-engineering-patterns
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill presents an inherent surface for indirect prompt injection as its core functionality involves interpolating user-provided data into prompts.\n
- Ingestion points: Data is ingested through variables in templates within
references/prompt-templates.mdand via test cases inscripts/optimize-prompt.py.\n - Boundary markers: Prompts include structural delimiters (e.g., 'Text: {text}') to help distinguish between instructions and data.\n
- Capability inventory: The skill's capabilities are limited to string processing and performing parallel API calls to an LLM provider via
ThreadPoolExecutorinscripts/optimize-prompt.py; no sensitive file access or shell execution is present.\n - Sanitization: There is no explicit sanitization of interpolated data, which is standard for prompt engineering tools where inputs are expected to be diverse.\n- [EXTERNAL_DOWNLOADS]: The skill documentation and code snippets reference industry-standard libraries and services such as the OpenAI API and common Python packages (
numpy,scikit-learn,sentence-transformers,scipy). These are documented as necessary dependencies for the skill's utility functions.
Audit Metadata