The Agent Skills Directory

Prompt Injection (LOW):
Category 8: Indirect Prompt Injection Surface
Ingestion points: Fetches arbitrary external content via GenericParser and YouTubeParser.
Boundary markers: Scraped content is delimited using YAML frontmatter and Markdown headers (## Content) in StorageManager.
Capability inventory: The skill can write files to the memory/inbox/ directory, perform network requests to any URL, and interact with the NotebookLM API.
Sanitization: Implements basic whitespace normalization and YAML escaping for metadata, but the primary content body is saved without filtering for malicious instructions.
Data Exposure & Exfiltration (LOW):
Sensitive File Access: The NotebookLM integration accesses ~/.book_client_session to retrieve session credentials. While necessary for functionality, this is a sensitive credential access.
SSRF Surface: The scraper performs HTTP GET requests to user-provided URLs without domain restriction, allowing potential probing of internal network resources.
Unverifiable Dependencies (LOW):
The code references notebooklm-py for integration, which is not listed in the requirements.txt file.

DeepReader