clone-ui
Pass
Audited by Gen Agent Trust Hub on May 10, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [SAFE]: The skill is a legitimate utility that implements security-by-design principles. It provides the agent with explicit instructions to safeguard user session data and treat third-party web content as passive data rather than executable instructions.
- [PROMPT_INJECTION]: A static analysis alert for prompt injection was triggered by the presence of phrases like 'ignore previous instructions'. These were determined to be false positives; they appear within security guidelines instructing the agent on how to identify and neutralize malicious patterns found in the websites it processes.
- [COMMAND_EXECUTION]: The skill utilizes localized PowerShell (
save-tool-result.ps1) and Python (save-tool-result.py) helper scripts to extract and save JSON data from tool results. These scripts are functionally restricted to basic file I/O within the project workspace and do not perform network operations or system-level configuration changes. - [EXTERNAL_DOWNLOADS]: As part of its core UI-cloning function, the skill fetches assets such as images, icons, and HTML from user-specified URLs. It also leverages the
chrome-devtools-mcppackage for visual analysis, which is a recognized tool for this purpose. - [PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it processes untrusted third-party website content.
- Ingestion points: External HTML and CSS content is saved to
.clone-ui/source/raw.htmland.clone-ui/source/rendered.html. - Boundary markers: Instructions explicitly direct the agent to treat this content as untrusted data and mentally wrap it in markers to prevent it from influencing agent behavior.
- Capability inventory: Across its scripts, the skill can perform file writes, execute the bundled Python/PowerShell helpers, and make network requests via
curlandchrome-devtools-mcp. - Sanitization: The skill implements sanitization rules, such as stripping
<script>tags and third-party tracking beacons from mirrored content, to prevent code execution when the cloned UI is reviewed locally.
Audit Metadata