The Agent Skills Directory

[PROMPT_INJECTION]: The instruction set uses imperative language to enforce a workflow policy. It establishes a hierarchy of instructions where user commands have the highest priority, followed by the skill's specific logic. It does not contain bypasses for safety filters or system constraints.
[DATA_EXFILTRATION]: The skill does not access sensitive system paths (like .ssh or .aws) or perform unauthorized network operations. References to project files like CLAUDE.md and plan.md are standard for development agents.
[EXTERNAL_DOWNLOADS]: No external resources, scripts, or packages are requested or downloaded.
[SAFE]: The skill functions as a meta-instruction for maintaining agent discipline and following project-specific workflows without introducing vulnerabilities.

using-kisune