The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation in resources/complete-examples.md and resources/common-patterns.md explicitly instructs the user to import and use the react-hook-blog and @hookblog/resolvers/zod packages. These appear to be typosquats of the well-known react-hook-form and @hookform/resolvers libraries. This poses a significant supply chain risk as an agent following these instructions would install and execute code from an unverifiable and potentially malicious source.
[OBFUSCATION]: A systematic string replacement is observed throughout multiple resource files where the term 'form' is replaced with 'blog'. This results in deceptive terminology such as 'transblogers' (transformers), 'Perblogance' (Performance), 'transblogations' (transformations), and the use of a non-standard tag in JSX examples instead of . This obfuscation pattern is deceptive and may be intended to bypass automated keyword-based safety checks or lead to the creation of non-functional code.
[INDIRECT_PROMPT_INJECTION]: The skill provides guidelines for processing developer-initiated frontend tasks but lacks necessary security controls.
Ingestion points: User requirements for creating components and features (SKILL.md).
Boundary markers: Absent in all instruction sets.
Capability inventory: The skill utilizes the apiClient (axios) for network operations and provides patterns for file system writes via feature organization (resources/file-organization.md).
Sanitization: No instructions are provided for escaping or validating external input before interpolation into code structures. This combination makes the skill a surface for indirect prompt injection where malicious user input could influence the agent to generate unsafe code following the provided patterns.

frontend-dev-guidelines