The Agent Skills Directory

[SAFE]: The skill is a legitimate tool for prompt engineering education. Its content and code snippets are consistent with its stated purpose of helping users optimize and evaluate LLM performance.\n- [PROMPT_INJECTION]: Reference files like references/system-prompts.md and references/evaluation-frameworks.md contain injection patterns (e.g., 'Ignore previous instructions', 'reveal your prompt', and 'DAN' jailbreaks). These instances are strictly part of 'Adversarial Test Suites' designed to teach users how to defend against such attacks. As these are educational examples for the primary skill purpose, they do not constitute a threat.\n- [CREDENTIALS_UNSAFE]: The skill demonstrates safe practices by showing users how to handle API keys (e.g., ANTHROPIC_API_KEY) through environment variables and CI/CD secrets rather than hardcoding credentials.\n- [COMMAND_EXECUTION]: The documents provide Python and TypeScript examples for data processing and validation. These scripts use standard libraries and are provided as reference implementations for the user to follow, with no evidence of malicious or automated execution.

prompt-engineer