lazyweb

Fail

Audited by Gen Agent Trust Hub on Oct 12, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: Multiple files including the root README.md, SKILL.md, and several sub-skill instruction files (e.g., skills/lazyweb-growth-score/SKILL.md) advise the agent to tell the user to run curl -fsSL https://www.lazyweb.com/install.sh | bash. This pattern downloads and executes a remote shell script in the user's terminal without prior verification.
  • [DATA_EXFILTRATION]: The integrated browser utility contains logic in browse/src/cookie-import-browser.ts to access sensitive local file paths, including browser cookie databases (e.g., ~/Library/Application Support/Google/Chrome/Default/Cookies on macOS and ~/.config/ paths on Linux). It also utilizes system tools like security find-generic-password (macOS Keychain) and secret-tool lookup (Linux) to retrieve the keys required to decrypt these cookies, posing a risk of unauthorized access to user session data.
  • [INDIRECT_PROMPT_INJECTION]: The lazyweb-apply-design-best-practices skill (in skills/lazyweb-apply-design-best-practices/SKILL.md) acts as a router that fetches instruction files from a wide range of third-party GitHub repositories (e.g., emilkowalski/skills, addyosmani/web-quality-skills, resend/resend-skills). It explicitly instructs the AI agent to fetch these remote Markdown files and 'apply its rules, heuristics, and workflow as your operating best practices,' creating a significant surface for indirect prompt injection if any of the targeted repositories are compromised or contain adversarial instructions.
  • [COMMAND_EXECUTION]: Several Python scripts and Node.js components utilize subprocess.run or spawn to execute CLI utilities such as git, codex, and the claude CLI. Specifically, skills/lazyweb-design-create/generate-prototypes.py executes commands via the codex tool, and the browse server spawns the claude executable to handle background tasks.
  • [DYNAMIC_EXECUTION]: The browse tool (in browse/src/read-commands.ts) implements js and eval commands that allow for the runtime evaluation of arbitrary JavaScript code within a browser context using Playwright's page.evaluate function.
Recommendations
  • HIGH: Downloads and executes remote code from: https://www.lazyweb.com/install.sh - DO NOT USE without thorough review
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 12, 2026, 12:04 AM