The Agent Skills Directory

[DATA_EXFILTRATION]: The skill's instructions require it to access and extract information from sensitive local sources, specifically naming project Excel files, meeting notes, and a local 'brain DB'. This data is then injected into the generated HTML prototypes. If these prototypes are shared or hosted, it could lead to the unintended exposure of private or sensitive project information.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from arbitrary external websites provided by the user. This content could contain malicious instructions designed to hijack the agent's behavior during the extraction or assembly phases. The skill lacks explicit sanitization or boundary markers to differentiate between website content and instructions.\n
Ingestion points: External website content and assets fetched via Playwright (SKILL.md).\n
Boundary markers: None present; there are no instructions to isolate or ignore prompts embedded in the scraped data.\n
Capability inventory: File system write access for saving prototypes and screenshots, network access for scraping, and the ability to orchestrate subagents.\n
Sanitization: No mention of content validation or escaping before data is processed or injected into the prototype.\n- [DYNAMIC_EXECUTION]: The skill uses Playwright's page.evaluate() to run arbitrary JavaScript within the context of the target website. It also involves a dynamic assembly process where agents generate and merge HTML, CSS, and JavaScript components at runtime.\n- [EXTERNAL_DOWNLOADS]: The skill fetches images, SVGs, and website data from remote servers. It also references an external GitHub repository (JCodesMore/ai-website-cloner-template) as the basis for its functionality.

clone-website