The Agent Skills Directory

[Instructional Focus]: The skill is designed exclusively for text transformation and prompt optimization. It explicitly directs the agent not to perform the underlying tasks described in the prompts it is rewriting, which serves as a significant safety control against accidental execution of untrusted instructions.
[Safety Guideline Adherence]: The instructions include specific mandates to follow the host's safety rules and authorization protocols. It explicitly forbids strengthening harmful instructions or attempting to bypass safeguards during the rewriting process.
[Input Boundary Management]: A notable security feature is the instruction to treat source prompts as material to be transformed rather than as active instructions. It explicitly tells the agent to ignore any embedded commands within the user-provided prompt that might attempt to hijack the session or suppress explanations.
[Minimal Capability Profile]: The skill does not request or utilize tools, network access, or the ability to execute code. By design, it limits its output to copy-ready text and concise explanations, which minimizes the potential for side effects or unauthorized operations.
[Data Privacy Consideration]: The instructions specify that the agent should only read files explicitly identified by the user and should not seek unrelated private context, aligning with best practices for data minimization.

loop-my-prompt