The Agent Skills Directory

[SAFE]: The skill incorporates specific safety rules, such as treating all processed PDF and bibliographic data strictly as project data rather than instructions. This design directly addresses potential indirect prompt injection vectors.
[COMMAND_EXECUTION]: The skill uses the jinko-sdk to interact with the Jinkō API and perform local file operations. These actions, including uploading/downloading PDFs and creating extracts, are necessary for the skill's core functionality and are managed via the SDK's initialized client.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface as it processes untrusted external data (PDFs and metadata). To mitigate this, the instructions include explicit boundary rules for the agent. The skill's capabilities are limited to project item management and authorized file system paths, preventing high-impact exploitation.

jinko-reference