docs: a proposal for one editor per workflow hierarchy

[endrix]

docs/proposals/single-editor-hierarchy.md

A diagram whose nested workflows live in other files opens a new editor for every nested workflow you drill into. That gives a tab per level, each with its own chat, a Run button that runs the workflow on screen rather than the top, and its own layout.

The proposal: one editor per root (the file opened). Every nested workflow, in any file, becomes a view inside it:

1. In-place navigation for drill-down, the breadcrumb and an outline. It is a profile choice (nestedNavigation); the default keeps today's behaviour, so other products are unaffected.
2. Chat anchored to the root (already true). The turn context gains the trail, so the agent knows what is on screen.
3. Run from the root: ▶ Run, "Rerun from Here" and ⟲.
4. Nested views editable in their own files. A file shown inside another file's editor is never left dirty, and the root editor refreshes when any file in its trail changes.
5. Layout per instance, stored in the root's layout file, falling back to the defining file's layout so existing layouts carry over.
6. One cached hierarchy export from wfpy instead of a wfpy plan per drill-down, with nested graphs elaborated as their instances.
7. A hierarchy outline modelled on mlir-viewer's: plain data on the root args, single click to look, double click to go, a filter that keeps ancestors.

It lists what already works (the server renders any file a request names, edits go to the shown file, chat is keyed by the root, and overlays reach nested views through the trail), the five phases, and three open questions. Docs only.

https://claude.ai/code/session_015VK7fH1c4aKbexnq2QcuKU