The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted user content (articles, tutorials) to generate image generation prompts.
Ingestion points: User-provided text or files are saved to source.md and processed during the analysis and storyboard phases.
Boundary markers: The workflow uses structured Markdown templates (e.g., references/base-prompt.md) to isolate user-derived descriptions from technical instructions.
Capability inventory: The agent can execute local shell commands (npx, pngquant, optipng) and perform file system operations to create images and PDF documents.
Sanitization: There is no explicit sanitization or filtering of input content to prevent embedded instructions from influencing the prompt generation logic.
[COMMAND_EXECUTION]: The skill executes shell commands to process images and generate final outputs.
Evidence: SKILL.md and references/workflow.md document the use of npx -y bun to run a local PDF merging script (scripts/merge-to-pdf.ts) and a sibling image generation skill. It also utilizes system tools like pngquant, optipng, and sips for image compression.
[EXTERNAL_DOWNLOADS]: The skill uses npx, which may download Node.js packages or runtimes if they are not pre-installed in the environment.
Evidence: Execution of npx -y bun in SKILL.md and references to the pdf-lib package in scripts/merge-to-pdf.ts.

baoyu-comic