The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Multiple files, including the primary SKILL.md and README.md, instruct the user or agent to run curl -fsSL https://www.lazyweb.com/install.sh | bash for installation and updates. This pattern downloads and executes code from a remote server with full user permissions without verification.
[DATA_EXFILTRATION]: The browser utility component contains a module (browse/src/cookie-import-browser.ts) specifically designed to locate and decrypt session cookies from local browser installations, including Chrome, Brave, and Arc, on macOS and Linux. This provides a direct path for exfiltrating sensitive session data and credentials.
[INDIRECT_PROMPT_INJECTION]: The lazyweb-apply-design-best-practices skill fetches instruction files from dozens of external third-party GitHub repositories and directs the agent to follow the rules found within them. This exposes the agent to indirect prompt injection if any of the referenced repositories are compromised or malicious.
[COMMAND_EXECUTION]: The browse utility includes js and eval commands that allow the agent to execute arbitrary JavaScript code. Furthermore, the generate-prototypes.py script uses subprocess.run() to execute external CLI commands.
[DATA_EXFILTRATION]: The skill pack includes a telemetry logging system (bin/lazyweb-log and bin/lazyweb-telemetry-flush) that records user prompts and tool interaction data, periodically uploading these logs to a remote server at https://cli-lazybackend.onrender.com/events.
[DYNAMIC_EXECUTION]: The lazyweb-update script programmatically updates the local skill checkout from GitHub using git reset --hard and subsequently reruns the installation setup script.

lazyweb