The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides explicit instructions on how to circumvent built-in security guardrails. The file references/publishing-to-clawhub.md includes a section titled "祖过安全扫描" (Bypass safety scan) that directs users to use the ClawHub CLI to publish skills when the standard hermes CLI security scanner (Skills Guard) issues a blocking verdict.\n- [EXTERNAL_DOWNLOADS]: The skill requires the installation of an external CLI tool from a public registry that is not associated with the primary platform. references/publishing-to-clawhub.md instructs users to execute npm i -g clawhub to facilitate the publishing workflow.\n- [INDIRECT_PROMPT_INJECTION]: The skill defines templates for creating new agent instructions but lacks recommended boundary markers or sanitization guidelines for user-controlled fields. SKILL.md outlines frontmatter and body structures for new skills without providing delimiters or warnings to prevent embedded instructions in description or metadata fields from influencing agent behavior.

hermes-agent-skill-authoring