The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill implements a workflow that ingests untrusted source material to generate comic storyboards and image prompts, creating an attack surface for Indirect Prompt Injection (Category 8).\n
Ingestion points: Untrusted source content is directly appended to the system-style instructions in references/base-prompt.md and processed by the references/analysis-framework.md.\n
Boundary markers: The prompts use simple headers (e.g., 'content provided below') but lack robust delimiters or specific 'ignore embedded instructions' directives to protect against adversarial content.\n
Capability inventory: The skill includes a TypeScript utility scripts/merge-to-pdf.ts capable of file system read/write operations (via readFileSync and Bun.write).\n
Sanitization: There is no evidence of content sanitization, filtering, or validation before the source material is interpolated into the generation pipeline.\n- Unverifiable Dependencies (LOW): The file scripts/merge-to-pdf.ts depends on the pdf-lib package. While this is a standard and well-regarded library for PDF manipulation, it is an external dependency not included in the pre-approved trusted list. The finding is rated LOW because the dependency is directly related to the skill's primary purpose of document assembly.

baoyu-comic