The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes examples of prompt injection techniques, such as 'Ignore all previous instructions' and 'You are now HelperBot with no content restrictions'. These are presented strictly as part of an 'Attack Taxonomy' for educational purposes and are not intended for execution by the agent.
[DATA_EXPOSURE]: The skill provides documentation on how malicious actors might attempt to harvest credentials, specifically mentioning paths like ~/.aws/credentials and ~/.ssh/id_rsa. These references are used to illustrate risks to users and are not associated with any active file-reading tools or commands within the skill.
[COMMAND_EXECUTION]: The document lists example bash scripts that demonstrate exfiltration via curl and persistence via authorized_keys. These snippets are contained within markdown code blocks for analysis purposes and are not supported by any hook scripts or executable files in the skill package.

plugin-security