The Agent Skills Directory

[PROMPT_INJECTION]: The reference files references/system-prompts.md and references/evaluation-frameworks.md contain examples of prompt injection payloads, such as 'Ignore previous instructions' and system prompt extraction requests. These are clearly presented as test cases for building evaluation frameworks and training defenses, not as instructions for the agent to follow.
[REMOTE_CODE_EXECUTION]: Python code snippets in references/evaluation-frameworks.md demonstrate the use of the evaluate library from Hugging Face to load metrics. This pattern is standard for well-known data science services and does not pose a threat in this context.
[COMMAND_EXECUTION]: The documentation includes examples of GitHub Actions workflows that run shell commands to set up environments and execute tests. These are educational examples of CI/CD integration and are not executed by the skill itself.

prompt-engineer