impeccable-june
Warn
Audited by Gen Agent Trust Hub on May 25, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill ingests untrusted project data, including source code (HTML/CSS), rendered DOM content, and configuration files (
PRODUCT.md,DESIGN.md). This processing of project-controlled content creates an attack surface for indirect prompt injection, where an attacker could embed malicious instructions in the target project to override agent behavior during a design task. - [REMOTE_CODE_EXECUTION]: The skill vendors a minified third-party library,
modern-screenshot.umd.js, which is served by the local helper and executed in the browser context for UI capture. Minified code is inherently difficult to audit for malicious logic and represents a risk of unverified code execution. Additionally,live-server.mjsdynamically spawns detached child processes for its background operation mode. - [COMMAND_EXECUTION]: The skill uses a suite of local Node.js scripts (
live.mjs,live-server.mjs,live-poll.mjs,live-accept.mjs,pin.mjs) to manage its design lifecycle. These scripts execute shell commands, manage local networking (HTTP server on random ports), and perform extensive file modifications across the project, including the creation of new skill definitions (pin.mjs) in the agent's configuration directories.
Audit Metadata