The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by ingesting untrusted data from external websites. An attacker could embed instructions in a website's DOM to influence the agent's behavior during the cloning process.
Ingestion points: The skill uses browser automation (e.g., Chrome MCP, Playwright) to navigate to target URLs and extract DOM structure, text content, and styles.
Boundary markers: None. There are no instructions to the agent to treat external website content as data rather than instructions or to ignore embedded commands.
Capability inventory: The skill has significant capabilities including executing shell commands (npm run build, npx tsc), writing files to the local project (src/components/), and executing generated Node.js scripts.
Sanitization: None. The skill directly extracts text content via element.textContent and computed styles via getComputedStyle without filtering or validation.
[COMMAND_EXECUTION]: The skill performs dynamic execution by generating a custom Node.js script (scripts/download-assets.mjs) at runtime based on asset discovery and then executing it (Phase 2). It also executes standard build and verification commands (npm run build, npx tsc --noEmit) throughout the cloning process.
[EXTERNAL_DOWNLOADS]: The skill downloads arbitrary assets, including images, videos, and fonts, from user-provided URLs and remote servers discovered during the reconnaissance phase (Phase 1 & 2).

clone-website