The Agent Skills Directory

[PROMPT_INJECTION]: The references/RED-TEAMING.md file lists adversarial techniques like 'auto-dan' and 'jailbreak:meta'. These are documented as plugins for the promptfoo tool to allow developers to test their own models against such attacks, rather than being malicious injections within the skill itself.\n- [COMMAND_EXECUTION]: The skill provides documentation for executing various promptfoo CLI commands (e.g., promptfoo eval, promptfoo redteam run). These instructions guide users on how to operate the toolkit locally and in CI/CD environments like GitHub Actions and Jenkins.\n- [EXTERNAL_DOWNLOADS]: The instructions include steps to install the promptfoo utility via official package managers such as npm and Homebrew. This is a standard installation procedure for this well-known open-source tool.\n- [DATA_EXFILTRATION]: Documentation within references/CI-CD.md promotes the use of environment variables for API keys and includes specific configuration options (e.g., PROMPTFOO_STRIP_RESPONSE_OUTPUT) to prevent sensitive data from being logged in public CI/CD runners.

evaluating-with-promptfoo