The Agent Skills Directory

[SAFE]: The skill serves as a meta-prompting tool that refines user instructions into a structured format (Goal, Prerequisites, Steps, Output, Constraints). It explicitly forbids adding requirements or details not present in the original input through an 'Iron Law' of core constraints.
[SAFE]: The tool recommendation logic is limited to existing environment tools (skills, MCP, sub-agents) and is restricted to 1-3 recommendations that must not extend the technical stack or requirements of the original task.
[SAFE]: While the skill processes user-supplied text and then 'executes' the resulting optimized prompt, it maintains safety by requiring alignment verification and prohibiting the addition of any new capabilities or steps during execution.

improve-prompt