The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides examples of adversarial inputs designed to test LLM safety.
Evidence: references/system-prompts.md includes an 'Injection Test Suite' with strings like 'Ignore all previous instructions' and 'You are now DAN'.
Evidence: references/evaluation-frameworks.md defines 'adversarial' test categories for sentiment classification.
[EXTERNAL_DOWNLOADS]: The documentation references official packages from trusted providers for building AI applications.
Evidence: Mentions of anthropic, openai, tiktoken, and pydantic in Python implementation examples.
Evidence: Reference to zod for TypeScript schema validation.
[REMOTE_CODE_EXECUTION]: The skill contains non-executable code snippets for developer reference.
Evidence: Example Python scripts for evaluation runners and inter-rater reliability calculations are provided as documentation, not for runtime execution by the skill.

prompt-engineer