The Agent Skills Directory

Persistence Mechanisms (MEDIUM): The script scripts/init-playwright.sh and the documentation in references/wsl2-configuration.md modify the user's ~/.bashrc file to append environment variable definitions (CHROME_BIN) and update the PATH. While intended for configuration, automated modification of shell profiles is a high-risk behavior, downgraded here due to the skill's setup purpose.
Privilege Escalation (MEDIUM): Documentation in references/wsl2-configuration.md includes commands utilizing sudo for installing Google Chrome and various system libraries required for browser automation on Linux/WSL2.
Indirect Prompt Injection (LOW): The skill enables browser navigation to external URLs via Playwright, creating a surface where malicious website content could potentially influence the agent's behavior. Evidence: 1. Ingestion points: page.goto() calls in SKILL.md and scripts/init-playwright.sh. 2. Boundary markers: No delimiters or instructions to ignore embedded content are provided. 3. Capability inventory: Subprocess calls (npm, npx, wget, dpkg, apt-get), file system operations (cat, echo, mkdir), and network access (wget). 4. Sanitization: No sanitization of website content is implemented.
External Downloads (SAFE): The skill downloads the Google Chrome Debian package from dl.google.com. As Google is a trusted organization, this download is considered safe.
Hardcoded Credentials (SAFE): Example code in references/best-practices.md contains dummy credentials ('test@example.com', 'password'). These are clearly illustrative and do not represent a security risk.

playwright-testing