The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses simulated user feedback ("重要：用户已经说过'这还不够完美...'" / "Important: User has already said 'This is not perfect yet...'") as a prompt engineering technique to force the agent into an iterative polishing state, which attempts to override default behavior patterns and instruction limits.
[EXTERNAL_DOWNLOADS]: Instructions explicitly state to "下载并使用任何需要的字体来实现这一目标" (Download and use any needed fonts to achieve this goal). While the included font license files refer to trusted and well-known organizations such as Google, Vercel, IBM, and Red Hat, the open-ended nature of the instruction allows for the retrieval of assets from potentially unverified sources.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: User-provided inputs regarding "subtle concepts," "小众参考" (niche references), and themes are incorporated into the "design philosophy" and the visual "DNA" of the final art in SKILL.md.
Boundary markers: Absent. The skill does not use specific delimiters or instructions to ignore commands potentially embedded within the user-provided design themes.
Capability inventory: The skill is capable of writing multiple file types (PDF, PNG, MD) to the system and potentially performing network operations to download font assets.
Sanitization: Absent. User input is used to guide the core creative logic without validation or filtering.

canvas-design