The Agent Skills Directory

[PROMPT_INJECTION] (SAFE): The instructions are focused on providing a framework for creating documentation and configuration files. There are no patterns attempting to bypass agent safety filters or override system instructions.
[DATA_EXFILTRATION] (SAFE): No sensitive file paths (e.g., credentials, SSH keys) are accessed. While WebFetch and WebSearch are permitted tools, the skill instructions do not direct the agent to send data to external or untrusted domains.
[REMOTE_CODE_EXECUTION] (SAFE): The skill does not download or execute remote scripts. It uses a static local template for generating new skills.
[COMMAND_EXECUTION] (SAFE): Although Bash is listed in allowed-tools, it is intended for the agent to perform standard development tasks (like creating directories or files). The skill itself does not contain any suspicious or hardcoded command-line instructions.
[INDIRECT_PROMPT_INJECTION] (LOW): As a 'meta-skill' that creates other skills, it technically facilitates the injection of new instructions into the agent's environment. However, this is the core intended purpose of the tool, and it includes explicit 'Content Guidelines' and 'Checklists' to ensure generated skills are structured and restricted to local references.

skill-creator