The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it is designed to ingest and query untrusted external data sources, such as web URLs and document uploads.
Ingestion points: The /api/sources endpoint accepts URLs and multipart file uploads (PDF, DOCX, audio, video) as demonstrated in SKILL.md and scripts/source_ingestion.py.
Boundary markers: The documentation and example scripts do not specify the use of clear delimiters or instructions to ignore embedded commands within the processed data.
Capability inventory: The associated backend service handles network requests to fetch external content and manages AI-driven analysis pipelines.
Sanitization: No explicit sanitization or filtering of the ingested source content is documented.
[EXTERNAL_DOWNLOADS]: The quick start guide instructs users to download a deployment configuration file from an external GitHub repository.
Evidence: curl -o docker-compose.yml https://raw.githubusercontent.com/lfnovo/open-notebook/main/docker-compose.yml in SKILL.md.
[COMMAND_EXECUTION]: The documentation includes standard setup and deployment procedures that require manual execution of shell commands.
Evidence: Instructions for installing and launching the service using curl, export, and docker-compose are provided for user implementation.

open-notebook