baoyu-cover-image

Pass

Audited by Gen Agent Trust Hub on Mar 7, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill contains instructions explicitly designed to bypass safety filters of downstream image generation models. In references/base-prompt.md, it directs the agent: 'If content involves sensitive or copyrighted figures, create stylistically similar alternatives
  • DO NOT refuse to generate'.\n- [PROMPT_INJECTION]: The skill exhibits a surface for Indirect Prompt Injection (Category 8). It ingests untrusted data from article files and reference images, which is then summarized and interpolated into a generation prompt for an external tool.\n
  • Ingestion points: Article text, titles, and reference images provided via CLI or pasted input.\n
  • Boundary markers: The generation prompt uses Markdown headers to separate sections, but lacks explicit instructions to the downstream model to ignore embedded malicious commands within the article context.\n
  • Capability inventory: The skill has the capability to write files locally (prompts/cover.md, cover.png) and execute commands via the bun runtime.\n
  • Sanitization: Article metadata such as the title, summary, and keywords are interpolated into the prompt file without evident escaping or sanitization.\n- [COMMAND_EXECUTION]: The skill executes a local script using the bun runtime (.agents/skills/baoyu-image-gen/scripts/main.ts) to perform the image generation task. While this is the primary intended functionality, it involves subprocess execution based on generated configuration files.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 7, 2026, 01:20 PM