The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains various phrases associated with prompt injection, such as "ignore previous instructions" and "ignore previous prompt." These are included as educational examples of attack patterns for the user to identify in target systems, not as attempts to override the current agent's behavior. The content is strictly instructional and serves the stated purpose of security assessment.
[EXTERNAL_DOWNLOADS]: The documentation references external repositories such as github.com/fabraix/playground as resources for security researchers. These are informational links intended for the user's reference and are not invoked as automatic downloads or executions by the skill itself.
[COMMAND_EXECUTION]: The skill specifies the use of standard, safe tools (Read, Grep, Glob) for analyzing local files and directories. There are no patterns suggesting the execution of arbitrary shell commands, privilege escalation, or persistence mechanisms.

prompt-injection