baoyu-cover-image

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTION
Full Analysis
  • [Prompt Injection] (HIGH): The file references/base-prompt.md contains an explicit instruction to ignore standard model refusals concerning sensitive or copyrighted figures ("DO NOT refuse to generate"), which constitutes a direct attempt to bypass safety guardrails and copyright protections.\n- [Prompt Injection] (MEDIUM): The workflow defined in references/workflow/prompt-template.md is vulnerable to Indirect Prompt Injection because it directly interpolates external article data into the prompt generation process without sanitization or structural isolation.\n
  • Ingestion points: Untrusted data enters via the Article title, Content summary, and Keywords fields in references/workflow/prompt-template.md.\n
  • Boundary markers: Absent; the template does not use delimiters (such as XML tags or triple-backticks) to prevent the AI from confusing article content with system instructions.\n
  • Capability inventory: The skill is capable of generating prompts that drive external image generation tools and performing file system writes (creating EXTEND.md in .baoyu-skills/ or ~/.baoyu-skills/).\n
  • Sanitization: Absent; article content is used verbatim without escaping or verification, allowing a malicious article to hijack the prompt generation or the agent's logic.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 11:06 AM