The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes a pipeline for extracting entities, relations, and events from unstructured external data such as text, transcripts, and PDFs. Ingesting untrusted data into the agent context for extraction tasks creates a potential surface for indirect prompt injection. Ingestion points: Unstructured text sources are processed in stages 4, 5, and 6 of the extraction pipeline as described in references/extraction.md. Boundary markers: The skill suggests using <text> and <ontology> tags in extraction prompts to isolate input data from instructions. Capability inventory: The skill provides templates for data extraction and describes multi-agent orchestration using tools like the omo:teammode controller. Sanitization: Defensive measures include requiring verbatim evidence quotes and performing domain/range validation for extracted relationship edges.
[EXTERNAL_DOWNLOADS]: The skill contains references to external repositories on GitHub for educational materials (npubird/KnowledgeGraphCourse) and a task orchestration runtime (code-yeongyu/lazycodex). These are documented as technical references and do not involve automated script execution or piped downloads.

graph-engineering