Skip to content

A fast 2D and 3D game engine — TypeScript SDK, C++/WebAssembly core, visual editor. Ship one project to web, desktop, WeChat MiniGames, playable ads, and native Android / iOS.

docs: changelog for the agent UX pass

esengined159188c8a74
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 1e156461a..7ca90c7f2 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -83,6 +83,20 @@ published separately; it ships inside the editor.
- **Reasoning shows how long it took**, a stopped or refused run explains itself in the
transcript rather than only in its header badge, `@` offers project assets alongside
entities, and re-ask is on the answer as well as on the run header.
+- **There is one way to be a summoned drawer.** There were two of them (the Content
+ Browser from the bottom, the agent from the right) and one implementation and a half:
+ the second wrote the scrim and the Esc key again, and got the slide and the focus
+ handling by not having them — it cut in and out instead of moving, and Tab walked out of
+ it into the panels its own scrim had just made unclickable. What differs between them is
+ an edge and a size, which is now all either one states. Both are bounded to the
+ workspace, so the menu bar and toolbar stay live and undimmed under a summoned surface,
+ the way the status bar already did.
+- **A streaming reply costs the run it belongs to, not the whole transcript.** Every token
+ re-rendered every finished run in the conversation, and each of those re-read the undo
+ stack twice in a full linear pass — so a long conversation over a long undo stack paid
+ all of that per token. Runs and tool rows memoise, the change set counts from the list
+ it already read, and a run's window in the history is found by binary search rather than
+ scanned for.
- **The agent can edit the asset editors, not just the scene.** Animation clips, timelines,
tilesets, materials, material graphs and the animator/state/behaviour graphs were
@@ -109,6 +123,20 @@ published separately; it ships inside the editor.
names it is not a smaller edit but a failed batch. Since authoring is one undo step, the
gate is neutral rather than a warning, and "Allow for this run" turns it off for the
rest of a long build.
+- **The agent can be answered from the keyboard.** A passage that stops a run takes focus
+ when it arrives — that is where the keyboard belongs while a run waits on it — and Enter
+ is its primary action, Shift+Enter the for-this-run one, both labelled on the buttons
+ rather than left to be found. The batch preview also shows the *values* it wants to
+ write, as the same `before → after` the change set uses, where it used to say "3 fields"
+ — which is exactly the part a preview exists to show. Strike-all takes a batch apart in
+ one gesture, and an emptied one disables Apply instead of offering "Apply 0", which the
+ kernel would have treated as a decline anyway.
+- **A run that was cut off says so, and can be carried on.** The turn loop is bounded so a
+ model that keeps calling tools cannot spin forever, and reaching that bound was reported
+ as an ordinary end of turn — on screen, a run cut off mid-task looked exactly like one
+ that had finished, and the sentence that would have said otherwise is the one it never
+ got to write. Both endings a run could have gone further from, that and Stop, now offer
+ to carry on from where it stopped.
### Fixed
@@ -132,6 +160,26 @@ published separately; it ships inside the editor.
the fresh one looked like a duplicate of the old — the reset travels in the event stream
now); and folding a second time ate what the first fold had saved, so the earliest thing
you asked disappeared one compaction at a time.
+- **A conversation no longer disappears on a click aimed at the model name.** A session is
+ built for one model, so switching has to end it — but that is the *cost* of the switch,
+ not what was asked for, and doing it silently read as having lost the transcript rather
+ than as having chosen to. Both it and New conversation ask first, and declining a switch
+ now leaves the pick alone too: a selection saved for "next time" while the running
+ conversation still answers on the old model is a picker that lies about what is running.
+- **Pressing Esc no longer costs you the message you were writing.** The draft lived in
+ the composer, and the drawer's composer unmounts when the drawer closes — so Esc, or a
+ click on the viewport to check something before sending, threw away what had been typed;
+ the docked panel and the drawer also disagreed about it. It is one draft now, restored
+ at the height it was, and a summoned drawer puts the caret in it.
+- **The one question most likely to scroll away was the one with nothing to say so.** The
+ "waiting on you" notice and the arrival flash were written against the irreversible gate
+ alone, so a batch preview — `apply_scene_ops` being the write the agent is *told* to
+ prefer, and the only one it asks about by default — could scroll out of the transcript
+ with nothing on screen to say the run was blocked on you.
+- **A call that failed says why without being asked.** The result cell beside a tool row
+ is clipped to a couple of dozen characters, which is enough to tell two successes apart
+ and never enough to carry a reason — so the one row in a run worth reading was folded
+ like all the rest. It opens itself now.
## [0.39.0] - 2026-08-01