The Agent Skills Directory

[SAFE]: The skill demonstrates secure credential handling by using System.get_env to retrieve API keys from the environment rather than hardcoding them in configuration files.
[SAFE]: Dependencies listed in the mix.exs snippet (jido, jido_ai, req_llm) are standard for the agent's ecosystem and trace back to the vendor's own libraries.
[PROMPT_INJECTION]: The skill defines a framework for AI agents that ingest user input to perform actions. This creates a surface for indirect prompt injection, which is documented here as an inherent characteristic of the skill's primary purpose.
Ingestion points: User input provided via the ask/2 and ask_sync/3 functions in SKILL.md.
Boundary markers: The provided templates do not explicitly show boundary markers or delimiters for user-supplied strings within the LLM prompt.
Capability inventory: The agents can execute arbitrary Jido.Action modules as tools (e.g., MyApp.Actions.GetWeather).
Sanitization: The 'DO / DON'T' section provides guidance to filter sensitive data from the tool context before passing it to the LLM.

jido-ai