The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses nltk.download('punkt') to fetch standard tokenizer models for sentence segmentation, which is an expected and safe operation for NLP tasks.
[PROMPT_INJECTION]: The RAG pipeline demonstrates the interpolation of retrieved context into a system prompt (e.g., f"Answer based on context:\n\n{context}"). This represents an indirect prompt injection surface as it lacks delimiters to separate untrusted data from instructions.
Ingestion points: rag_query function in SKILL.md.
Boundary markers: Absent in the provided prompt templates.
Capability inventory: LLM generation and retrieval from local vector databases.
Sanitization: Not demonstrated in basic pattern examples.
[CREDENTIALS_UNSAFE]: Uses a documentation placeholder string "your-api-key" for the Cohere client initialization, which does not constitute a secret leak.

rag-patterns