The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied prompts and images for generative tasks, which represents a potential surface for indirect prompt injection if the source data contains malicious instructions. 1. Ingestion points: scene_prompt, instruction, and product_file_ids defined in SKILL.md. 2. Boundary markers: The skill instructs that 'Generated media always requires review,' promoting human-in-the-loop validation. 3. Capability inventory: The skill invokes remote generation tools and accesses the AgentPMT File Manager. 4. Sanitization: No explicit sanitization logic is described in the prompt templates; reliance is placed on provider safety filters and manual review.
[SAFE]: The skill follows security best practices by advising against the inclusion of credentials in prompts and using platform-managed remote tool calls for its operations. It references installation scripts and setup skills from the vendor's repository, which is standard for this ecosystem. No malicious patterns, obfuscation, or unauthorized access attempts were detected.

product-mockup-studio