From process diagram to machine-readable intent
While reviewing the export format, I realized the diagram can serve a larger purpose than process visualization. I believe the first safety net of agentic alignment is to remove ambiguity from instructions. If an agent is given an INTENT.md contract then agents can act with more clarity around vulnerable or critical systems where more deterministic outcomes are required.
In this project, each node, port (node input and output point), and relationship have theirs own markdown and metadata. The result is a structured graph where every part can explain what should happen, why it should happen, what evidence is required, and where authority stops.
This opens up a possible use case as an agent-readable intent contract - INTENT.md
During the diagram buildout, a model could inspect the complete process to find ambiguous or missing rules, conflicting conditions, unsafe capability paths, and downstream consequences. The current export already provides stable IDs, explicit relationships, decision paths, and human-review boundaries. The next step would be to add typed permissions, prohibitions, invariants, data contracts, and authoritative branch conditions. This goes beyond instructions or skills by outlining a process end-to-end so an agent understands what's in scope and how to get from point A to point B in a safe and repeatable pattern.
Log in or sign up for Devpost to join the conversation.