The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The installation process fetches code from external GitHub repositories that are not recognized as trusted sources.
install.sh clones the repository https://github.com/Bwkyd/wexin-read-mcp.git for WeChat extraction.
install.sh installs the Python package notebooklm-py directly from https://github.com/teng-lin/notebooklm-py.git.
[REMOTE_CODE_EXECUTION]: The skill is designed to run external scripts as part of its core functionality.
It executes a Python server script (server.py) from the cloned wexin-read-mcp repository to operate as an MCP server.
check_env.py utilizes __import__ for dynamic module loading during environment verification.
[COMMAND_EXECUTION]: The skill frequently invokes shell commands to perform file conversions and interact with CLI tools.
It calls markitdown for document conversion, using arguments derived from user-provided file paths.
It executes the notebooklm CLI tool for creating notebooks and uploading sources.
Shell commands in check_env.py use subprocess.run to verify tool versions and authentication status.
[DATA_EXFILTRATION]: The skill facilitates the transfer of local user data to an external AI service.
Local documents (PDF, DOCX, EPUB, images) are converted to text and uploaded to Google's NotebookLM platform.
The check_env.py and install.sh scripts access the user's local Claude configuration file (~/.claude/config.json) to verify settings.
[PROMPT_INJECTION]: The skill processes untrusted content from the web and local files, posing a risk of indirect prompt injection.
Ingestion points: The skill fetches content from WeChat articles, YouTube transcripts, and various user-provided document formats.
Boundary markers: The instructions in SKILL.md do not define clear delimiters or "ignore previous instructions" warnings for the processed content before it is uploaded to the AI service.
Capability inventory: The skill has significant capabilities including network access, file reading/writing, and shell command execution across multiple files.
Sanitization: There is no evidence of sanitization, filtering, or validation of the ingested content before it is processed by the AI.

anything-to-notebooklm