The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses a 'fabricated feedback' technique by stating "The user ALREADY said 'It isn't perfect enough. It must be pristine...'" in the instructions to force the agent into a refinement loop regardless of actual user input. This attempts to override the agent's natural response to user feedback.
[EXTERNAL_DOWNLOADS]: The instructions explicitly tell the agent to "Download and use whatever fonts are needed", which encourages fetching unverified resources from the internet without specifying trusted sources.
[COMMAND_EXECUTION]: The skill's primary purpose involves generating visual artifacts (.pdf and .png), and the instructions "Go back to the code and refine/polish further" imply that the agent is expected to write and execute scripts to perform image processing or document generation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input to derive a 'design philosophy' which then governs the visual generation process. This creates a surface where malicious instructions in user input could influence the generated manifesto and subsequent visual output.
Ingestion points: User input foundation in SKILL.md.
Boundary markers: None provided to separate user input from the philosophy generation prompt.
Capability inventory: File-write (PDF/PNG), potential shell/code execution to generate art mentioned in SKILL.md.
Sanitization: None detected.

canvas-design