The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill outlines frameworks for RAG and knowledge extraction that process external, untrusted content like research papers and summaries, creating an inherent injection surface.\n
Ingestion points: Processes text from academic papers (rag-methodology-guide/SKILL.md), abstracts (knowledge-graph-construction/SKILL.md), and lab observations (graphiti-guide/SKILL.md).\n
Boundary markers: Prompt snippets use simple headers (e.g., 'EXCERPTS:') rather than robust delimiters or explicit instructions to ignore adversarial commands embedded within the documents.\n
Capability inventory: The described systems require network access for LLM interaction and databases, alongside file system access for storage and indexing.\n
Sanitization: No explicit sanitization or validation of the input research text is included in the provided code logic.\n- [EXTERNAL_DOWNLOADS]: The skill provides instructions for setting up necessary research infrastructure using third-party software and registries.\n
Python dependencies such as graphiti-core, openspg, neo4j, and spacy are downloaded from standard package registries.\n
Integration guides reference open-source Zotero plugins hosted on GitHub, including Notero (dvanoni/notero) and MarkDB-Connect (daeh/zotero-markdb-connect).\n
Core knowledge graph frameworks like Ant Group's OpenSPG are utilized for semantic reasoning and graph storage.

knowledge-graph-skills