The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's README.md and install.sh files utilize curl to fetch configuration files (SKILL.md) and design references (STYLE_PRESETS.md) from a public GitHub repository to the local machine.
[COMMAND_EXECUTION]: The skill provides install.sh and deploy.sh scripts that execute shell commands to manage directories, initialize git repositories, perform text substitution with sed, and interact with the GitHub CLI (gh) for project management.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided product descriptions and existing HTML source code to generate landing pages, creating a vulnerability surface.
Ingestion points: User prompts for product information and legacy HTML code ingested via SKILL.md for generation and improvement modes.
Boundary markers: The skill's instructions in SKILL.md lack explicit delimiters or boundary markers to distinguish instructions from untrusted external data.
Capability inventory: The skill generates static HTML files with embedded client-side JavaScript (using Anime.js) for browser-based interactivity. It does not perform persistent file modifications or server-side execution beyond code generation.
Sanitization: No explicit validation or sanitization logic is implemented in the skill instructions to filter malicious patterns in processed user data.

frontend-landings