Fix blank-graph creation and actor execution workflow #52

New issue

Closed

opened 2026-07-26 01:00:08 +00:00 by lost-rob0t · 0 comments

lost-rob0t commented

2026-07-26 01:00:08 +00:00

(Migrated from github.com)

Problem

The graph workbench exposes a quick-add control, but a newly created document is unreviewed while the graph defaults to reviewed-only. The new node is immediately hidden, its selection is pruned, and the workflow looks broken. The empty graph also emphasizes import instead of creating the first node.

Actor execution compounds the problem:

every actor is hidden behind the custom-code enable switch, including trusted bundled actors;
buttons are disabled without explaining selection/type requirements;
running and completed states are not visible;
actor output is unreviewed and can disappear behind the same review filter.

Scope

Make the empty graph a first-class starting state with Create first node and Import actions.
Keep quick-add, selection, and review visibility synchronized so a manually created node remains visible.
Preserve manual relation creation for exactly two selected nodes.
Allow trusted bundled actors to run without enabling user-supplied actor code.
Add explicit selection bounds, dtype applicability reasons, running state, completion summary, and output limits.
Reveal and select actor-created nodes after a successful validated batch.
Add initial bundled username-candidate and WhatsMyName query actors.
Add unit and browser coverage for the blank-graph-to-actor flow.

Acceptance criteria

Starting with an empty local corpus, a user can create and see the first node without leaving /graph.
A user can create more nodes and connect exactly two selected nodes.
Newly created and actor-created unreviewed nodes do not silently disappear.
Bundled actors clearly explain why they can or cannot run.
Custom actor code remains opt-in.
Actor results use the canonical validation and undo path.
The username actors can derive bounded variants from canonical person names and prepare clickable WhatsMyName searches.

## Problem The graph workbench exposes a quick-add control, but a newly created document is unreviewed while the graph defaults to reviewed-only. The new node is immediately hidden, its selection is pruned, and the workflow looks broken. The empty graph also emphasizes import instead of creating the first node. Actor execution compounds the problem: - every actor is hidden behind the custom-code enable switch, including trusted bundled actors; - buttons are disabled without explaining selection/type requirements; - running and completed states are not visible; - actor output is unreviewed and can disappear behind the same review filter. ## Scope - Make the empty graph a first-class starting state with Create first node and Import actions. - Keep quick-add, selection, and review visibility synchronized so a manually created node remains visible. - Preserve manual relation creation for exactly two selected nodes. - Allow trusted bundled actors to run without enabling user-supplied actor code. - Add explicit selection bounds, dtype applicability reasons, running state, completion summary, and output limits. - Reveal and select actor-created nodes after a successful validated batch. - Add initial bundled username-candidate and WhatsMyName query actors. - Add unit and browser coverage for the blank-graph-to-actor flow. ## Acceptance criteria - Starting with an empty local corpus, a user can create and see the first node without leaving /graph. - A user can create more nodes and connect exactly two selected nodes. - Newly created and actor-created unreviewed nodes do not silently disappear. - Bundled actors clearly explain why they can or cannot run. - Custom actor code remains opt-in. - Actor results use the canonical validation and undo path. - The username actors can derive bounded variants from canonical person names and prepare clickable WhatsMyName searches.