from LLM chat to agent to guided workflow

1 · Graphthinking Blogspot · July 5, 2026, 5:01 p.m.
Summary
I have an LLM agent written in pure python that has access to bash commands. The structure of the script that creates the agent is # system prompt identifies for the LLM which bash commands are available while True: # Get user message. (This is where the human interacts for input.) # append user input to context while True: # query LLM to process context if bash_command_in_LLM_response: # seek user confirmation to run bash command # run bash...