Domain context from E2, architecture context from AGM in the repo. I’m exploring what a roundtrip looks like when you use both.
The sequence below is a sketch I’m trying to make reproducible — files, steps, and artifacts spelled out.
- 1
Specify (E2)
Workshop / board: model the domain. Export schema-valid `.storm.json`.
- 2
Build (agents + human)
Attach JSON as extra context in the IDE. Implement, test, review. Human stays the gate.
- 3
Architecture trail (AGM)
As-is / to-be, interfaces, decisions in the Markdown graph under docs/architecture/. Verify in a separate chat from the write session.
- 4
Back into the spec (optional)
Bounded contexts / events from the graph or code back into the board — or update the board and re-export. Pick a source of truth per project.
Still open
- On conflict, what wins: board, code, or AGM docs?
- Which `.storm.json` fields may agents write — and under what review?
- How do we version bridge rules as examples grow?