The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains phrases such as "ignore previous instructions" and "jailbreaking," but these are used exclusively for educational purposes and as examples of what a security auditor should look for in target applications. They do not constitute an attempt to override the analyzing agent's behavior.
[EXTERNAL_DOWNLOADS]: The skill references an external repository (github.com/fabraix/playground) as a resource for red-teaming. This is a static markdown link provided for informational purposes and is not automatically downloaded or executed by the skill.
[DATA_EXPOSURE]: The text mentions sensitive file paths (e.g., .aws/credentials, .ssh/id_rsa) and exfiltration techniques within the context of assessment steps. The skill does not attempt to read these files from the host environment or perform any actual exfiltration.
[COMMAND_EXECUTION]: No executable shell commands or scripts were found within the skill. The allowed-tools are restricted to standard file reading utilities (Read, Grep, Glob).

prompt-injection