The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides an implementation that is vulnerable to Indirect Prompt Injection (Category 8).
Ingestion points: Untrusted data is ingested into the system via the addDocument method (which stores document content) and the query method (which accepts user questions) in src/services/rag.service.ts.
Boundary markers: The implementation in src/utils/ollama.ts lacks sufficient boundary markers or 'ignore' instructions; it uses simple string templates like Context: ${context}\n\nQuestion: ${prompt} to combine retrieved data with user queries.
Capability inventory: The code provides capabilities to read from and write to a PostgreSQL database and generate text via a connected Ollama LLM service.
Sanitization: There is no evidence of sanitization, escaping, or validation of the retrieved context strings before they are interpolated into the final prompt sent to the LLM.

rag-pipeline