The Agent Skills Directory

[SAFE]: The skill is entirely instructional and navigational. It contains no executable commands, subprocess calls, or scripts that could impact the underlying system.
[PROMPT_INJECTION]: The skill proactively includes a security rule instructing the agent to treat user-provided skill definitions as data rather than instructions, which mitigates potential injection risks during skill creation or modification flows.
[DATA_EXFILTRATION]: No network-enabled tools or sensitive file path references (such as SSH keys or environment files) are present in the skill definition.
[REMOTE_CODE_EXECUTION]: The skill does not specify any external dependencies or remote script downloads. All referenced tools (like skill-maker or skill-audit) are part of the internal meta-system and are described as separate modules.

meta-guide