The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install external packages using the npx skills add command (e.g., belt-sh/cli, inference-sh/skills@ai-image-generation, inference-sh/skills@web-search). These resources are fetched from external repositories and are not pinned to specific versions or hashes, which introduces a supply chain risk.
[REMOTE_CODE_EXECUTION]: The skill utilizes a CLI tool (belt) to execute remote applications provided by third-party services (e.g., falai/flux-dev-lora, tavily/search-assistant). This execution pattern relies on the integrity and security of the remote infrastructure and logic provided by those vendors.
[DATA_EXFILTRATION]: User-provided inputs, such as descriptive image prompts and research queries, are transmitted to external domains including inference.sh, fal.ai, and tavily.com. While necessary for the skill's functionality, this behavior involves sending potentially sensitive project or product details to third-party services.
[INDIRECT_PROMPT_INJECTION]: The skill incorporates external search results from tools like tavily/search-assistant and exa/answer into the agent's context. Because this data originates from the public web and lacks clear boundary markers or sanitization, it presents an attack surface for indirect prompt injection.
[DYNAMIC_EXECUTION]: The infsh/html-to-image tool accepts raw HTML strings as input to generate images. This allows for the dynamic construction and remote rendering of markup, which could be exploited if the skill interpolates untrusted data into the HTML template without proper escaping.

landing-page-design