The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill includes code templates that encourage the installation and use of highly suspicious Node.js packages: react-hook-blog and @hookblog/resolvers/zod. These packages appear to be typosquatting attempts or impersonations of the legitimate react-hook-form and @hookform/resolvers/zod libraries. Using these unverified dependencies can lead to the execution of malicious code during build or runtime.
[EXTERNAL_DOWNLOADS]: The skill advocates for downloading and integrating non-standard, impersonated packages into the project environment, posing a significant supply chain risk.
[PROMPT_INJECTION]: The skill defines a "CRITICAL RULE: UI RESEARCH HARD-GATE" using authoritative language to mandate that the agent "MUST run a Subagent" for research. This instruction is designed to override standard agent autonomy and force specific tool invocations that may be unnecessary or lead to context manipulation.
[METADATA_POISONING]: While the skill claims to provide standard frontend development guidelines, it deceptively inserts references to potentially malicious libraries, misrepresenting the safety and standard nature of the provided code samples.
[INDIRECT_PROMPT_INJECTION]: The skill provides templates that incorporate malicious dependency references. An agent utilizing these guidelines to generate code for a user would indirectly introduce these compromised packages into the user's codebase.
Ingestion points: User requests for component or feature creation in SKILL.md.
Boundary markers: None present; the skill lacks instructions to ignore or sanitize embedded instructions in output.
Capability inventory: File creation and subagent invocation across all resource files.
Sanitization: Absent; the skill lacks any mechanisms to validate or sanitize the package names it promotes.

frontend-dev-guidelines