The Agent Skills Directory

[SAFE]: The skill facilitates interaction with a self-hosted research tool. It correctly handles sensitive information like instance passwords and API keys through environment variables, which is a standard and safe practice.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to download a configuration file from a public repository. This is a common pattern for self-hosted software deployment.
Evidence: curl --fail --location --output docker-compose.yml https://raw.githubusercontent.com/lfnovo/open-notebook/v1.14.0/docker-compose.yml found in SKILL.md and references/configuration.md.
[DATA_EXFILTRATION]: Network operations are appropriately scoped to the user-configured backend URL and well-known research platforms.
Evidence: The skill makes requests to the user-provided OPEN_NOTEBOOK_URL and fetches metadata from arxiv.org as described in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: While the skill processes external research data (PDFs, web content) that could contain malicious instructions, this is inherent to its primary purpose as a research assistant. The client-side helpers do not implement specific sanitization, but they rely on the backend instance's processing logic.
Ingestion points: scripts/source_ingestion.py (URL, file, and text ingestion).
Boundary markers: Not implemented in the client scripts; the skill relies on the Open Notebook backend's internal prompt construction.
Capability inventory: Network requests via requests to the backend API; file writing for podcast downloads in references/examples.md.
Sanitization: Not implemented in the client-side Python helpers.

open-notebook