The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from external sources, including web URLs, YouTube videos, and Google Drive documents, through commands like nlm source add and nlm research start. This ingestion of external data presents a surface for indirect prompt injection, where malicious instructions embedded in the source material could attempt to override the agent's behavior.
Ingestion points: nlm source add <id> --url, nlm research start <query>, and nlm source add <id> --drive <doc-id>.
Boundary markers: The skill does not explicitly define markers to delimit external content, though it uses structured CLI commands.
Capability inventory: The agent can execute bash commands, perform network operations via the CLI, and manage cloud resources (NotebookLM).
Sanitization: Not explicitly mentioned in the skill instructions.
[COMMAND_EXECUTION]: The skill's primary function is to provide instructions for the agent to execute the nlm CLI tool via bash. This allows the agent to interact with the local filesystem and network through the CLI interface provided by the tool.
[EXTERNAL_DOWNLOADS]: The skill documentation (AGENTS_SECTION.md) provides instructions to install the notebooklm-mcp-cli tool using uv tool install. While uv is a well-known package manager, this involves the installation of third-party executable code on the host system.

nlm-skill