The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (HIGH): The script setup_environment.py automatically downloads and installs a Google Chrome binary using patchright. This dependency is a non-standard fork and is not within the trusted source scope.
[REMOTE_CODE_EXECUTION] (HIGH): The skill performs remote code execution by downloading and running a browser installer (python -m patchright install chrome) during the environment setup phase. This allows for the execution of arbitrary binaries on the host system.
[PROMPT_INJECTION] (HIGH): The script ask_question.py appends a hardcoded directive (FOLLOW_UP_REMINDER) to the output retrieved from the external website. This instruction uses high-priority markers ('EXTREMELY IMPORTANT') to force the agent into a specific behavioral loop, which is a signature of control-hijacking injections.
[COMMAND_EXECUTION] (MEDIUM): The skill relies on a centralized run.py wrapper that uses subprocess.run to dynamically execute Python scripts. This structure increases the attack surface for command injection if arguments are improperly handled.
[DATA_EXFILTRATION] (MEDIUM): The authentication architecture explicitly extracts and stores Google session cookies in a local state.json file. While documented for 'session persistence', this creates a high-value target for exfiltration of sensitive credentials.

notebooklm