ux-audit

Warn

Audited by Gen Agent Trust Hub on Jul 2, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill executes pip3 install shot-scraper to acquire an external utility for screenshot capture without pinning a specific version. It also runs shot-scraper install which downloads necessary browser binaries.
  • [COMMAND_EXECUTION]: The skill uses the Bash tool to run several operations, including curl for remote data retrieval, shot-scraper for browser-based automation, and gh (GitHub CLI) for repository management and code pushing. It also starts a local web server in the background using python3 -m http.server.
  • [DATA_EXFILTRATION]: The skill uses the gh tool to push local project files to a remote GitHub repository. While creating a private repository, this involves sending local code to an external service. It also performs network operations to user-provided domains via curl.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8) because it fetches and processes untrusted HTML and CSS from arbitrary URLs to generate reports and code. 1. Ingestion points: Website content fetched from user-supplied URLs via curl. 2. Boundary markers: None used; the agent is not instructed to ignore instructions found within the fetched data. 3. Capability inventory: File system access (Edit, Write), network access (curl), and code deployment (gh repo create). 4. Sanitization: No validation or sanitization is performed on the extracted tokens before they are used to generate the final code.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 2, 2026, 05:13 PM
Security Audit — agent-trust-hub — ux-audit