The Agent Skills Directory

[SAFE]: The skill functions as a prompt engineering assistant and does not exhibit malicious patterns. The instructions are focused on enhancing user-provided text for better LLM performance.
[COMMAND_EXECUTION]: The skill instructs the agent to use the mkdir command to manage a dedicated directory (.aix/) for its outputs. This is a legitimate operational task for the specified environment and does not involve unsafe interpolation of user input.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to generate optimized prompts. It mitigates potential risks by recommending structured XML tags for data separation and providing clear instructions on handling missing context.
Ingestion points: User-supplied draft prompts and background context provided to the skill for optimization.
Boundary markers: The skill explicitly recommends the use of XML tags (e.g., <instructions>, <context>, <input>) to delimit different sections of the prompt.
Capability inventory: File system writes and directory creation (mkdir) are mentioned as part of the AIX Pack compliance logic.
Sanitization: Recommends grounding analysis in quotes and using explicit instruction structuring to prevent the model from drifting.

prompt-optimizer