The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's documentation and reference files frequently instruct the agent to execute shell commands using local Python scripts, specifically scripts/gemini_batch_process.py and scripts/media_optimizer.py. These scripts are used to interface with Google's Imagen and Gemini models and to optimize media assets. Although central to the skill, the source code for these scripts is not provided in the bundle.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for the agent to install the animejs package using the Node Package Manager (npm). This is used for implementing the animation guidelines provided in the reference documentation.
[PROMPT_INJECTION]: The skill is designed with a workflow that ingests untrusted user content (such as screenshots, images, and design videos) to extract design systems and guidelines. These extracted guidelines are then used as instructions for the agent to generate code, which constitutes an indirect prompt injection attack surface.
Ingestion points: External media assets provided by users are processed through multimodal vision analysis (as described in SKILL.md and references/design-extraction-overview.md).
Boundary markers: The skill does not include instructions or markers to delimit the untrusted data or to warn the agent against following instructions embedded within the analyzed visual content.
Capability inventory: The agent is authorized to execute subprocesses (via the provided scripts), install packages, and perform file system writes to generate documentation and code.
Sanitization: There is no evidence of sanitization or validation of the data extracted from the visual assets before it is used to influence the agent's code generation.

frontend-design