The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, which creates a potential surface for indirect prompt injection attacks.
Ingestion points: The skill explicitly triggers when a 'User pastes existing website code, a URL, a screenshot, or a design reference' (SKILL.md).
Boundary markers: The instructions do not define specific delimiters or instructions to the agent to ignore potentially malicious instructions embedded within the user-provided code or references.
Capability inventory: The skill has the capability to generate and suggest executable frontend code, including script inclusions and library integrations across various files (SKILL.md).
Sanitization: There is no mention of sanitizing or escaping the user-provided content before the agent analyzes or replicates it.
[EXTERNAL_DOWNLOADS]: The skill recommends the use of numerous well-known, industry-standard third-party libraries and services.
Evidence: The 'Human-Crafted Tooling Kit' mentions Lucide, Phosphor, Tabler, Heroicons, react-icons, Radix UI, Headless UI, MUI, Chakra UI, daisyUI, Tailwind CSS, Framer Motion, GSAP, Lenis, and Three.js. It also mentions platforms like Vercel, Google Fonts, and Webflow.
Assessment: These are established tools and services in the web development ecosystem. Their inclusion is consistent with the skill's primary purpose of producing professional, human-crafted frontend code and does not represent a malicious execution risk.
[SAFE]: The skill promotes several security and engineering best practices.
The 'Library-First Principle' encourages the use of mature, human-written, and maintained libraries for common tasks, which reduces the risk of bugs and vulnerabilities often found in custom-generated logic.
The 'Anti-AI-Tell Checklist' and 'Uniqueness Rule' are focused on design quality and do not introduce security risks.
Author 'Emmraan' provides clear, structured instructions that align with the stated purpose of the skill.

frontend-craft