The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill utilizes the auth login --chrome command to programmatically extract Google session cookies from the local Chrome browser storage. These cookies are sensitive authentication tokens that grant the tool full access to the user's NotebookLM environment.
[EXTERNAL_DOWNLOADS]: The installation instructions fetch external code using npx -y @mvanhorn/printing-press-library and go install github.com/mvanhorn/printing-press-library. These resources originate from the vendor's GitHub repository and are executed on the local system.
[COMMAND_EXECUTION]: The skill drives the execution of the notebooklm-pp-cli binary to perform file operations, database management (SQLite), and network requests.
[DATA_EXFILTRATION]: Extracted authentication cookies and notebook data (sources and chat history) are transmitted to Google's RPC endpoints via the CLI. While functional for the tool, this behavior moves sensitive browser data to external cloud services.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection:
Ingestion points: The source add <url> command allows adding content from any external web page to a notebook, and the chat ask tool processes these materials.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between its core logic and the untrusted content fetched from sources.
Capability inventory: The skill allows the agent to execute shell commands, write to local databases, and perform network requests.
Sanitization: No filtering or validation of the ingested URL content is described, allowing potentially malicious instructions in source documents to be processed by the agent.

pp-notebooklm