book-illustrator

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill can download book manuscripts from user-provided URLs for analysis (Step 1 in SKILL.md).
  • [COMMAND_EXECUTION]: The skill uses the /baoyu-imagine tool to generate images and invokes subagents to parallelize prompt generation.
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it processes external manuscript data. It implements defensive measures to mitigate this risk:
  • Ingestion points: External Markdown files downloaded from URLs or read from local paths (SKILL.md).
  • Boundary markers: Explicit 'Prompt Injection Block' instructions in references/design-system.md and structured brief templates in SKILL.md are used to constrain subagent behavior.
  • Capability inventory: The skill can write files to the local file system (prompts/, imgs/), invoke subagents, and call the image generation tool.
  • Sanitization: Multiple manual and automated checklists are enforced to verify that generated prompts and images adhere strictly to the design system and do not include unauthorized elements.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 04:06 PM
Security Audit — agent-trust-hub — book-illustrator