The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides instructions for the agent to automate the extraction of sensitive session cookies from a running Google Chrome instance. This is achieved by connecting to the Chrome DevTools Protocol (CDP) on port 9400. Extracted credentials are stored locally in the ~/.nlm/env file. While functional for the tool's purpose, this involves handling high-value authentication data from the user's browser environment.
[INDIRECT_PROMPT_INJECTION]: The skill allows the agent to process content from untrusted external sources, creating a pathway for indirect prompt injection attacks.
Ingestion points: Data is brought into the agent's context and the NotebookLM environment through commands like nlm source add <url>, nlm source add <youtube-url>, and nlm research, which performs web-based data retrieval.
Boundary markers: The skill does not provide clear delimiters or instructions to the agent to distinguish between safe operational instructions and potentially malicious content embedded within the ingested sources.
Capability inventory: The agent can execute shell commands via the nlm tool to generate chat responses, summaries, audio/video overviews, and other artifacts based on the provided external data.
Sanitization: There is no mention of sanitizing or validating the retrieved external content to prevent embedded instructions from overriding the agent's intended behavior.

nlm