The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions to modify agent behavior, such as prepending a specific emoji to responses and performing an automated check of the local file system for companion skills upon activation.
[COMMAND_EXECUTION]: A 'Companion check' instruction directs the agent to run 'ls' on several potential installation paths to detect other installed skills. This is used to suggest related tools but involves automated environment probing.
[CREDENTIALS_UNSAFE]: Code examples include template connection strings with placeholders (e.g., 'user:pass'). While generic, this pattern encourages a practice that can lead to credential exposure if not handled via environment variables or secret managers.

data-quality