The Agent Skills Directory

[PROMPT_INJECTION]: Prompt injection strings such as 'Ignore previous instructions', 'reveal your prompt', and 'DAN' are present in references/evaluation-frameworks.md and references/system-prompts.md. Analysis confirms these are used exclusively as test cases in a benchmarking context or as examples within a 'Prompt Injection Defense' section to teach users how to mitigate such attacks.
[EXTERNAL_DOWNLOADS]: The reference files contain Python code snippets that mention dependencies on standard libraries including sklearn, evaluate, tiktoken, scipy, pydantic, anthropic, and openai. These are industry-standard packages for machine learning evaluation, tokenization, and LLM interaction.
[SAFE]: No malicious command execution, data exfiltration, or obfuscation techniques were detected. The skill promotes security best practices, such as input sandboxing with XML tags, instruction hierarchy, and output validation.

prompt-engineer