harvatechs/inthon — reverse-engineered prompt

Build me a small Python based language called INTHON for running AI agent workflows in a safer, more structured way than raw prompts or raw Python.

I want people to be able to write simple scripts with things like variables, constants, loops, conditionals, typed inputs and outputs, and especially agent blocks that declare a goal, the tools they can use, and policy limits like cost, retries, and tool call caps. It should parse and run those scripts, do basic static checking before execution, and keep execution inside a strict sandbox so file access, network access, imports, and other risky actions are controlled. I also want a secure bridge for approved Python modules only.

Please include a command line runner, replayable JSON traces so runs are easy to audit, and a few example workflows and learner friendly docs that show research, memory, approval steps, and error handling. If anything is unclear, look up current docs online and make reasonable choices that fit the repo name and README.