The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation includes a command to download and execute the Ollama installation script (https://ollama.com/install.sh) by piping it directly to the shell. Ollama is a recognized service for local LLM execution.
[EXTERNAL_DOWNLOADS]: The skill requires cloning a code repository from GitHub (GiovanniPasq/agentic-rag-for-dummies) and installing several dependencies from the standard Python package registry.
[INDIRECT_PROMPT_INJECTION]: The skill implements a Retrieval-Augmented Generation (RAG) system, which creates an attack surface for indirect prompt injection.
Ingestion points: The system ingests PDF documents from a local directory, converting them to markdown for processing.
Boundary markers: The prompt templates provided, such as AGENT_PROMPT and AGGREGATION_PROMPT, interpolate retrieved document content directly into the context without using specific delimiters or instructions to ignore embedded commands.
Capability inventory: The skill includes tools for file system writes and vector database retrieval, which could be misused if the LLM follows instructions embedded within documents.
Sanitization: There is no implementation of content filtering or sanitization to prevent adversarial instructions within the retrieved documents from influencing the agent's logic.

agentic-rag-for-dummies