The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions focus on adherence to the local knowledge base and do not contain patterns attempting to bypass safety filters or override system instructions.
[DATA_EXFILTRATION]: The skill explicitly instructs the agent to ignore sensitive user information such as phone numbers, emails, and IDs from the resume input to prevent data exposure. No unauthorized network operations or access to sensitive local paths (e.g., .ssh, .aws) were detected.
[EXTERNAL_DOWNLOADS]: External links are limited to the author's own platform (CodeFun2000) and Bilibili for educational purposes, which is consistent with the skill's mock-interview purpose.
[INDIRECT_PROMPT_INJECTION]: The skill handles user resumes, creating an indirect prompt injection surface. 1. Ingestion points: User resume text input via prompts. 2. Boundary markers: Explicit instructions to ignore privacy fields and use provided Markdown structures. 3. Capability inventory: Access is limited to reading and grepping local files within the knowledge directory. 4. Sanitization: Instructions strictly prohibit the fabrication of information not found in the resume or knowledge base.

hw-interview