The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The README.md recommends installing the skill using a command that pipes a remote shell script from GitHub directly into bash. This pattern allows for the execution of unverified remote code on the local machine.
Evidence: curl -fsSL https://raw.githubusercontent.com/sergeyramas/3d-website-builder-skill/main/install.sh | bash in README.md.
[COMMAND_EXECUTION]: The provided installation script (install.sh) uses shell commands to create local directories and perform git operations to retrieve the skill's source code.
Evidence: git clone --depth 1 https://github.com/sergeyramas/${SLUG}-skill "$DEST" in install.sh.
[EXTERNAL_DOWNLOADS]: The skill retrieves its own components and resources from the author's external GitHub repositories.
Evidence: Resource fetching from github.com/sergeyramas/.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted user input from the brand interview and interpolates it directly into the source code of the generated website without sanitization.
Ingestion points: Responses to questions about brand name, colors, and call-to-action (CTA) text in SKILL.md.
Boundary markers: Absent. There are no delimiters or instructions provided to the agent to treat user input as data rather than potential instructions.
Capability inventory: The agent leverages the Write tool to generate index.html, style.css, and scroll.js files.
Sanitization: Absent. The instructions do not specify any validation or escaping of user-provided strings before they are written to the file system.

3d-website-builder