The Agent Skills Directory

[SAFE]: No security risks were detected. The skill is purely instructional and follows least-privilege principles by disabling tool invocation via frontmatter configuration.
[NO_CODE]: The skill package contains no executable scripts, binaries, or command-line instructions. It relies entirely on natural language instructions to guide agent behavior.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (draft prompts). While this represents a theoretical attack surface for indirect prompt injection, the risk is mitigated by the 'disable-model-invocation: true' setting, which prevents the agent from executing tools or taking automated actions based on the processed content.

meta-prompting