The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external inputs such as test cases and user queries to populate prompt templates, creating a typical attack surface for indirect prompt injection.\n
Ingestion points: Input variables are ingested by the PromptOptimizer class in scripts/optimize-prompt.py and various template rendering classes in references/prompt-templates.md.\n
Boundary markers: The skill encourages best practices like instruction hierarchy, though the automation scripts rely on standard string interpolation.\n
Capability inventory: The skill is capable of performing local file operations to save optimization results and utilizes ThreadPoolExecutor for parallel processing.\n
Sanitization: The ValidatedTemplate class provides a mechanism for validating input types and ranges before rendering.\n- [DYNAMIC_EXECUTION]: The skill implements logic to dynamically assemble and render prompts based on variable input and conditional patterns.\n
Evidence: The ConditionalTemplate class in references/prompt-templates.md uses regular expressions to process logic blocks like if-statements and loops within prompt strings.\n- [EXTERNAL_DOWNLOADS]: The skill utilizes several well-known libraries for machine learning, statistics, and data processing.\n
Packages: References include openai, numpy, scikit-learn, scipy, and sentence-transformers.\n- [COMMAND_EXECUTION]: The skill includes a utility script scripts/optimize-prompt.py that is intended to be run by the user to automate the testing of different prompt variations.

prompt-engineering-patterns