The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates a workflow that ingests and processes untrusted visual content to drive downstream code generation and script execution.
Ingestion points: User-provided screenshots, images, and design references are ingested and analyzed via the ai-multimodal skill as described in SKILL.md and references/design-extraction-overview.md.
Boundary markers: The workflow mandates documenting extracted findings in docs/design-guidelines/extracted-design.md before implementation. While this intermediate step separates the vision analysis from the final code generation, it lacks automated sanitization or validation of the extracted data.
Capability inventory: The skill provides the agent with instructions to generate production-grade code (React, Vue, HTML/CSS/JS) and execute local Python scripts (scripts/gemini_batch_process.py, scripts/media_optimizer.py) which interface with external AI generation and analysis APIs.
Sanitization: There are no explicit instructions or mechanisms for sanitizing or escaping the visual content or the data extracted from it before it is interpolated into subsequent prompts or source code.

frontend-design