The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external documents (e.g., using RecursiveCharacterTextSplitter and create_documents), which creates a surface for indirect prompt injection if the processed content contains malicious instructions.
Ingestion points: Document content is processed in SKILL.md (via raw_docs) and references/chunking-strategies.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the examples.
Capability inventory: The skill utilizes file system access for document reading and performs network operations via various API clients (OpenAI, Cohere, Qdrant).
Sanitization: No content sanitization or filtering is implemented before the data is processed or summarized by an LLM.
[REMOTE_CODE_EXECUTION]: The LateChunker implementation in references/chunking-strategies.md uses trust_remote_code=True to load the jinaai/jina-embeddings-v2-base-en model from Hugging Face. This setting allows the execution of custom code provided in the model's repository during the loading process, representing a remote code execution vector from a well-known service.
[METADATA_POISONING]: The SKILL.md frontmatter identifies the author as https://github.com/Jeffallan, which does not match the platform's registered author metadata (austin-aziz).

rag-architect