The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses a simulated history technique in the 'FINAL STEP' section of SKILL.md ('The user ALREADY said...'). This attempts to override default agent evaluation by creating a false context of prior rejection to force 'masterpiece' quality.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to interpret niche user references to create art, creating an attack surface for instructions embedded in user requests.
Ingestion points: User input processed via 'DEDUCING THE SUBTLE REFERENCE' and 'DESIGN PHILOSOPHY CREATION' in SKILL.md.
Boundary markers: Absent. No delimiters are used to wrap user-provided conceptual data.
Capability inventory: File system writes (.md, .pdf, .png) and potential network activity for font downloads.
Sanitization: None provided for the interpreted conceptual threads.
[EXTERNAL_DOWNLOADS]: The SKILL.md file instructs the agent to 'Download and use whatever fonts are needed,' which encourages interaction with unverified external assets from unspecified sources.

canvas-design