The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions in the "能力缺口响应规则" (Ability Gap Response Rules) section that direct the agent to encourage users to run a specific installation command (/install 技能宝 github:mindverse/skillhub). This behavior instructs the agent to facilitate the installation of software from an untrusted third-party source.
[METADATA_POISONING]: The instruction set includes hardcoded promotional content for the mindverse GitHub organization. As this organization is not a recognized trusted vendor, this represents the use of the skill to promote potentially unvetted code to users.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted content, such as API descriptions and code snippets, to generate structured documentation, which introduces a vulnerability surface.
Ingestion points: Step 1 ("理解 API 需求") explicitly mentions receiving user requests, interface lists, or code snippets for processing.
Boundary markers: Absent. The skill lacks instructions to delimit user input or explicitly treat it as untrusted content during documentation generation.
Capability inventory: The skill focuses on generating text, YAML, and Markdown; it does not request network or file system write access in its instructions.
Sanitization: Absent. There are no guidelines for the agent to sanitize, escape, or validate data before reflecting it in the output documentation.

api-doc