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.

feat(editor)!: remove the built-in agent; plugin agent tools reach MCP clients

yhhd7ab195c5c54
diff --git a/.gitignore b/.gitignore
index 71615e0b6..ffc8b94ea 100644
--- a/.gitignore
+++ b/.gitignore
@@ -155,8 +155,8 @@ docs/local/
# Plugin API typings: written by the editor on project open, always matching the
# running version. A committed copy would go stale against a different editor.
**/.esengine/plugins/.types/
-# Conversations with the built-in agent (electron/agent/store.ts): one person's
-# working notes, holding whatever they typed and whatever the model read back.
+# Conversations an older editor's built-in agent kept: one person's working
+# notes, holding whatever they typed and whatever the model read back.
**/.esengine/agent/
# What each target weighed last build (pipeline/src/export/sizeHistory.ts) — one
# machine's measurement of one moment, which the next build compares against.
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 32a7643d5..fe57a898d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -14,6 +14,14 @@ published separately; it ships inside the editor.
## [Unreleased]
+### Removed
+
+- **The editor no longer has an agent of its own.** The Agent drawer, its panel, the status-bar and activity-bar entries, the palette's "Hand to Agent" row, the provider settings (keys, models, custom endpoints), the agent marks in the Outliner and viewport, and the History panel's folded agent runs are gone, with the `agent_send`, `agent_status`, `agent_confirm`, `agent_transcript` and `done_when` tools. Agents connect over MCP (Window → Connect AI Tools…), where the same tools have always been served. History now lists every step on its own line.
+
+### Changed
+
+- **Plugin agent tools reach MCP clients.** A tool a plugin lends with `ctx.agentTools.register` is now listed to every connected MCP client beside the editor's own, and a call runs in the editor. Its `schema` is checked first, and a tool whose `effect` is not `read` is offered only while the editor lets agents change the project. The LDtk plugin's `estella_ldtk_import` is the first.
+
## [0.97.0] - 2026-10-31
### Added
diff --git a/README.md b/README.md
index aa97a7700..44705b909 100644
--- a/README.md
+++ b/README.md
@@ -52,7 +52,7 @@ Estella is a **2D and 3D game engine** — sprites, tilemaps and UI as much as l
| **TypeScript SDK** | Type-safe API: `defineSystem`, `defineComponent`, `Query`, `Commands` |
| **Cross-Platform** | One project → web, desktop, WeChat MiniGames, playable ads, and native Android / iOS |
| **Native mobile** | A real arm64 app rendering through an embedded Dawn (Metal / Vulkan) — not a WebView |
-| **AI agents** | The editor is an MCP server with 130+ tools, and ships a built-in agent that drives the same ones |
+| **AI agents** | The editor is an MCP server with 130+ tools: connect Claude Code, Cursor or your own agent |
## Getting Started
@@ -101,14 +101,10 @@ entities from the same Create menu you use, edits component fields, enters play
takes a screenshot to *see* what it built, and exports the finished game. Sixty-five
tools, all of them the pipelines the UI itself calls.
-It also ships an agent of its own, so you don't need an external tool to start: type a
-sentence in the editor and watch every tool call it makes. Both front doors drive one
-catalog — what the built-in agent can do and what your MCP client can do are the same
-list, and both go through the editor's command surface, so creates are undoable and you
-can take the mouse back mid-turn.
+Every call goes through the editor's command surface, so creates are undoable and you
+can take the mouse back at any time. Tools a plugin adds are listed beside the editor's own.
-- [The Built-in Agent](https://estellaengine.com/docs/agents/built-in/) — set up a provider and start asking
-- [External Agents (MCP)](https://estellaengine.com/docs/agents/mcp/) — connect Claude Code, Cursor, or your own
+- [Connecting an Agent (MCP)](https://estellaengine.com/docs/agents/mcp/) — connect Claude Code, Cursor, or your own
## Documentation
@@ -122,7 +118,7 @@ Full documentation: [estellaengine.com/docs](https://estellaengine.com/docs)
- [Systems](https://estellaengine.com/docs/core-concepts/systems/)
- [Building & Exporting](https://estellaengine.com/docs/publishing/overview/)
- [Android & iOS](https://estellaengine.com/docs/publishing/android-ios/)
-- [AI Agents](https://estellaengine.com/docs/agents/built-in/)
+- [AI Agents](https://estellaengine.com/docs/agents/mcp/)
## Contributing
diff --git a/desktop b/desktop
index 71761c17e..01a2c6bbb 160000
--- a/desktop
+++ b/desktop
@@ -1 +1 @@
-Subproject commit 71761c17e50cd2588761a72c63459a0edc9e03a4
+Subproject commit 01a2c6bbb12f80f7c1abf6250b8c985ee717d968
diff --git a/docs/astro/astro.config.mjs b/docs/astro/astro.config.mjs
index 1411a6fd2..7a7010e26 100644
--- a/docs/astro/astro.config.mjs
+++ b/docs/astro/astro.config.mjs
@@ -69,9 +69,11 @@ const MOVED = {
'guides/services': 'publishing/ads-sharing',
'guides/profiling': 'performance/profiling',
'guides/editor-plugins': 'extending/editor-plugins',
- 'guides/agent': 'agents/built-in',
+ 'guides/agent': 'agents/mcp',
'guides/mcp': 'agents/mcp',
- 'extending/built-in-agent': 'agents/built-in',
+ 'extending/built-in-agent': 'agents/mcp',
+ // The editor no longer ships its own agent; an agent connects over MCP.
+ 'agents/built-in': 'agents/mcp',
'extending/mcp': 'agents/mcp',
};
@@ -177,13 +179,12 @@ export default defineConfig({
{
// Driving the editor, not extending it — which is why this sits beside
// the editor rather than under "Extending", where it read as a plugin
- // topic. Both pages describe one tool catalog with two front doors.
+ // topic.
label: 'AI Agents',
translations: { 'zh-CN': 'AI 智能体' },
collapsed: true,
items: [
- { label: 'The Built-in Agent', translations: { 'zh-CN': '内置智能体' }, slug: 'agents/built-in' },
- { label: 'External Agents (MCP)', translations: { 'zh-CN': '外部智能体 (MCP)' }, slug: 'agents/mcp' },
+ { label: 'Connecting an Agent (MCP)', translations: { 'zh-CN': '连接智能体 (MCP)' }, slug: 'agents/mcp' },
],
},
{
diff --git a/docs/astro/src/assets/guides/en/agent-preview.png b/docs/astro/src/assets/guides/en/agent-preview.png
deleted file mode 100644
index 2ffca6f53..000000000
Binary files a/docs/astro/src/assets/guides/en/agent-preview.png and /dev/null differ
diff --git a/docs/astro/src/assets/guides/en/agent-settings.png b/docs/astro/src/assets/guides/en/agent-settings.png
deleted file mode 100644
index 13e33bfff..000000000
Binary files a/docs/astro/src/assets/guides/en/agent-settings.png and /dev/null differ
diff --git a/docs/astro/src/assets/guides/en/agent-turn.png b/docs/astro/src/assets/guides/en/agent-turn.png
deleted file mode 100644
index d39666fa7..000000000
Binary files a/docs/astro/src/assets/guides/en/agent-turn.png and /dev/null differ
diff --git a/docs/astro/src/assets/guides/zh/agent-preview.png b/docs/astro/src/assets/guides/zh/agent-preview.png
deleted file mode 100644
index 5452f2e2a..000000000
Binary files a/docs/astro/src/assets/guides/zh/agent-preview.png and /dev/null differ
diff --git a/docs/astro/src/assets/guides/zh/agent-settings.png b/docs/astro/src/assets/guides/zh/agent-settings.png
deleted file mode 100644
index 1b108dc5b..000000000
Binary files a/docs/astro/src/assets/guides/zh/agent-settings.png and /dev/null differ
diff --git a/docs/astro/src/assets/guides/zh/agent-turn.png b/docs/astro/src/assets/guides/zh/agent-turn.png
deleted file mode 100644
index ff6c2c597..000000000
Binary files a/docs/astro/src/assets/guides/zh/agent-turn.png and /dev/null differ
diff --git a/docs/astro/src/content/docs/agents/built-in.mdx b/docs/astro/src/content/docs/agents/built-in.mdx
deleted file mode 100644
index 769d3dc3b..000000000
--- a/docs/astro/src/content/docs/agents/built-in.mdx
+++ /dev/null
@@ -1,170 +0,0 @@
----
-title: The Built-in Agent
-description: Ask the editor to do something in plain language. The agent reads your scene and edits it through the same tools the editor's own UI uses — with an Undo checkpoint around every turn and a confirmation before anything Undo cannot reach.
----
-
-import { Aside, Steps } from '@astrojs/starlight/components';
-
-The editor ships with an agent of its own. You type a sentence; it reads the scene,
-calls the same tools the editor's UI calls, and shows you every one of them as it
-goes. Nothing about it is a separate copy of the editor: it drives the exact tool
-catalog an [external MCP client](/docs/agents/mcp/) drives, so what it can do and
-what the editor can do are one list.
-
-<Aside type="tip">
-It is the same window you were already working in. Its edits land in your scene,
-your Undo takes them back, and you can take the mouse at any moment.
-</Aside>
-
-## Give it a model
-
-The editor ships no key and calls nothing on its own. Open **Settings ▸ Editor ▸ Tools and AI ▸ AI Agents**
-and configure one provider — that is the whole setup.
-
-![Settings ▸ Editor ▸ Tools and AI ▸ AI Agents — the table of your own providers: a name, a protocol, an address, model names and a key each](@/assets/guides/en/agent-settings.png)
-
-<Steps>
-
-1. Paste an **API key** for Anthropic, DeepSeek or OpenAI. Each provider keeps its
- own, so switching back to one you used last week costs nothing.
-
-2. Some providers also want their **model names** typed in. That is deliberate: a
- list shipped in the editor would be right until the vendor's next release, and a
- name that no longer exists is not refused — the endpoint quietly serves something
- smaller instead, for the rest of the session.
-
-3. Or add your own under **Your providers** (open the page's **Advanced** fold) — any endpoint at all, as many as you
- have. Each row is a provider like the shipped ones: a name, which **protocol**
- it speaks, its address, its model names, and its own key. Expand a row to say
- what that endpoint can do.
-
-4. Close Settings. The model picker sits next to the box you type in — it is a
- per-message choice, not a preference buried in a dialog.
-
-</Steps>
-
-<Aside type="tip" title="Which protocol">
-**Chat Completions** is OpenAI's, and it is what almost everything else implements
-too — the other vendors, the aggregators, and every local runner (Ollama, vLLM, LM
-Studio). **Messages** is Anthropic's; a gateway that emulates it works here as well,
-minus the extensions that are Anthropic's own (extended thinking, prompt caching,
-automatic fallbacks). The two formats differ in every message they send, so a wrong
-pick is refused outright rather than quietly degraded.
-</Aside>
-
-<Aside type="tip" title="Where the key lives">
-In the system keychain. It is never written to the project, never written to the
-settings file, and never handed back to the editor window — only to the client that
-has to send it.
-</Aside>
-
-## What an endpoint can do
-
-The shipped providers already say. For one of your own the editor has no way to
-find out, and guessing is expensive in both directions — so a row's expander is
-where you tell it. What is left unsaid is worked around, not discovered.
-
-- **Accepts images** — the agent takes screenshots to check its own work. Where a
- picture cannot be sent it asks for the same frame as a coarse colour grid in
- text instead, and says so in the transcript. Off by default: claiming sight an
- endpoint lacks costs a whole turn to a refused request, while withholding it
- costs detail the agent can mostly do without.
-- **Context window** — how far a conversation may grow before its oldest runs are
- folded away. Blank assumes a conservative window, which folds a conversation
- that still had room; too generous fails the turn outright, after the wait.
-- **Takes a reasoning-depth argument** — whether the request may name the depth
- set above. An endpoint that has never heard of the argument refuses the whole
- call rather than ignoring it, so turn this off for one that does.
-
-## Ask it something
-
-Open the command palette (`Ctrl/Cmd+Shift+P`) and run **Agent: Open**. Or skip that:
-type your question straight into the palette. Anything that reads as a sentence
-rather than a command offers **Hand to Agent**, and `Enter` sends it.
-
-![A finished agent run: the question, two tool calls, and the answer](@/assets/guides/en/agent-turn.png)
-
-A run reads top to bottom:
-
-- **The question**, with the model that answered it, how long it took, and the
- tokens it cost.
-- **A row per tool call** — `get_scene_tree`, `get_event_bindings` — with its
- arguments and how much came back. This is the part worth reading: it is the
- difference between an answer about your scene and an answer about scenes in
- general.
-- **The answer.**
-
-In the composer, `@` references an entity by name so you do not have to describe
-which one you mean. You can drop or paste an **image** — a mock-up, a screenshot of
-the bug — and it is sent along with the question, where the endpoint accepts images.
-While it works, the status bar carries a live segment with a **Stop** button, so a
-long run is visible and interruptible with the drawer closed.
-
-## What it may do without asking
-
-The tiers are drawn where **going back stops working**, which is what makes them worth
-knowing:
-
-| Tier | Examples | What happens |
-| --- | --- | --- |
-| Read | `get_scene_tree`, `get_inspector`, `capture_viewport` | Runs. Never asks. |
-| Undoable | create, delete, `set_field`, re-parent, paint tiles | Runs. The whole turn is bracketed by one checkpoint, so a single **Undo** takes all of it back. |
-| Project files | `write_project_file`, `save_scene`, creating a script, scene or asset | Runs. The editor keeps each file's before-image, so the turn's **Revert** in History takes the files back with the scene. |
-| Irreversible | creating a new project, exporting a build, `run_editor_command`, probing the running game | **Always asks first.** Nothing can take it back afterwards. |
-
-So the approval you give is not per edit — it is the Undo button, and it applies to
-the turn as a whole. When a run finishes having changed something, the transcript
-leaves a bar behind offering **Undo** or **Keep**, and it tells you if you have
-edited since (because then Undo would take your own work back too).
-
-## Before a large edit
-
-One undoable tool still stops to ask, and not for safety: `apply_scene_ops` authors
-a whole subtree in a single call, and a hundred-node edit is worth seeing before it
-lands rather than reading back afterwards.
-
-![The preview of a bulk scene edit: six changes listed, each strikeable, with Apply all / Allow for this run / Skip](@/assets/guides/en/agent-preview.png)
-
-Every line is one change. Click one to **strike it out** — striking a new entity also
-strikes whatever refers to it — then **Apply** the rest. **Allow for this run** stops
-the asking until the turn ends, which is what keeps a long build from becoming a
-hundred prompts; the next run asks again.
-
-Declining is not a dead end: the agent is told it was skipped and works around it.
-
-## The conversation is kept with the project
-
-A finished run is saved alongside the project — the transcript and the model's memory
-of it. **Earlier conversations** lists them; pick one and it carries on where it
-stopped. The next question in an open conversation still has everything the previous
-one established.
-
-Two things are worth knowing about a long one:
-
-- **The context gauge** in the drawer's header says how full the model's window is.
- Past a threshold the earliest runs are *folded out of its memory*: what you asked is
- kept word for word, the tool calls and their results are not. The transcript still
- shows them — the model can no longer read them. When that happens the drawer says
- so, on the line where it happened.
-- **A conversation belongs to the model that answered it.** Switching models ends it
- and starts a new one. What it changed in the scene stays.
-
-You can also **ask a run again** in different words: it discards that run and
-everything after it, then re-asks. Useful when a turn went the wrong way three
-messages ago and you would rather not talk it back.
-
-<Aside type="caution">
-It can be confidently wrong, like any model — most often about state it read several
-runs ago. When an answer describes your scene rather than reasoning about it, ask it
-to re-read rather than trusting the summary. That is also why the tool rows are on
-screen: they show what it actually looked at.
-</Aside>
-
-## See also
-
-- [AI Agents (MCP)](/docs/agents/mcp/) — the same tool catalog, driven from Claude
- Code, Cursor, or your own agent instead of from inside the editor.
-- [Event Binding](/docs/scripting/events/) — what the agent reaches for when you ask
- for behaviour without code.
-- [Prefabs](/docs/world/prefabs/) — in Prefab Mode the agent edits the prefab, not
- the scene. Check the document it is working on before asking for a big change.
diff --git a/docs/astro/src/content/docs/agents/mcp.mdx b/docs/astro/src/content/docs/agents/mcp.mdx
index afe2ba189..7b4de7401 100644
--- a/docs/astro/src/content/docs/agents/mcp.mdx
+++ b/docs/astro/src/content/docs/agents/mcp.mdx
@@ -127,7 +127,7 @@ reconnects automatically rather than stranding the agent.
The everyday tools (about fifty: the scene, Play, opening and saving, screenshots, logs) are listed from the start; the rest are in
groups the agent turns on with `enable_tools` when it needs them: `assets`, `prefab`, `authoring`, `view`, `play`, `render`,
-`project`, `code`, `device` and `agent`. A full list handed over before the first call costs every session thousands of tokens and
+`project`, `code` and `device`. A full list handed over before the first call costs every session thousands of tokens and
makes the right tool harder to find; any tool can still be called by name without being listed. Add `--all-tools` for a client
that cannot follow a change of the tool list. The table below names tools from every group.
diff --git a/docs/astro/src/content/docs/editor/history-scatter-health.mdx b/docs/astro/src/content/docs/editor/history-scatter-health.mdx
index 29296e2a8..22c9ca276 100644
--- a/docs/astro/src/content/docs/editor/history-scatter-health.mdx
+++ b/docs/astro/src/content/docs/editor/history-scatter-health.mdx
@@ -12,16 +12,8 @@ Command Palette for any of them.
marking where you are. It is the visible form of Undo / Redo (`Ctrl/⌘ + Z`,
`Ctrl/⌘ + Shift + Z`): click an earlier row to step back to that state.
-A run of the [built-in Agent](/docs/agents/built-in/) folds into **one row**, named
-after the prompt (**Agent · …**). That row has a **Revert** button which takes the
-**whole run** back — the scene edits *and* the project files it wrote. Because files
-are involved, going back past a run asks first and says how many runs and files it
-will undo. This one cannot be redone. Anything too large to hold a copy of is listed
-and left as it is.
-
-After a run, History can offer **Keep this check**: a verification the run passed —
-the scene has nothing flagged, or the scripts compile — that you can pin to the
-project. Every later run is measured against it, and no run can weaken or remove it.
+Edits an [agent connected over MCP](/docs/agents/mcp/) makes land here like your own,
+one row per step, and step back the same way.
## Scatter Brush
diff --git a/docs/astro/src/content/docs/editor/settings.mdx b/docs/astro/src/content/docs/editor/settings.mdx
index 86b102142..8cadf2074 100644
--- a/docs/astro/src/content/docs/editor/settings.mdx
+++ b/docs/astro/src/content/docs/editor/settings.mdx
@@ -11,8 +11,8 @@ setting on any page.
Performance, Console, and the renderer backend.
- **Viewport** — grid, snapping, and the transform gizmos.
- **Shortcuts** — every shortcut below is rebindable here.
- - **Tools and AI** — the external script / image editors and browser, and the
- built-in [agent](/docs/agents/built-in/): keys, models and endpoints of your own.
+ - **Tools and AI** — the external script / image editors and browser, and whether
+ [AI agents](/docs/agents/mcp/) may connect over MCP and change the project.
- **Project** — saved into `project.esproject` and shared with whoever opens it:
- **Game** — **Display** (the project design resolution, which seeds new Canvases and
drives the device preview in any scene; the project-wide **screen orientation**, which
diff --git a/docs/astro/src/content/docs/extending/editor-plugins.mdx b/docs/astro/src/content/docs/extending/editor-plugins.mdx
index a2b285305..61aafbf53 100644
--- a/docs/astro/src/content/docs/extending/editor-plugins.mdx
+++ b/docs/astro/src/content/docs/extending/editor-plugins.mdx
@@ -399,9 +399,10 @@ bundle inputs. Editing an imported source, imported JSON or `plugin.json` still
### Lending the agent a tool
-The built-in agent's entire vocabulary is the tool catalog. A plugin that adds a
-capability without adding a tool has added it for the person and not for the
-agent — so if your plugin can bake occlusion, teach the agent to ask for it:
+An agent connected over [MCP](/docs/agents/mcp/) can do only what the tool list
+offers. A plugin that adds a capability without adding a tool has added it for
+the person and not for the agent — so if your plugin can bake occlusion, teach
+the agent to ask for it:
```typescript
ctx.agentTools.register({
@@ -419,20 +420,24 @@ ctx.agentTools.register({
deciding whether this is the right tool. Say what it does and when to reach for
it.
-`effect` decides whether the person is asked first, and the tiers are drawn
-where going back stops working. `read` and `undoable` run unasked — the turn's
-checkpoint is the approval. So does `journaled`, which is what to declare when
-your tool writes project files through `ctx.fs.writeProject`: those are captured
-before they land, so the turn's Revert takes them back with the scene. Reserve
-`irreversible` for what no checkpoint reaches — work outside the open project,
-or code whose effects nobody enumerated; it stops and asks.
+The tool is listed to every connected MCP client beside the editor's own, and a
+call runs your `run` in the editor. `schema` is checked first: a call with a
+missing argument, or one the schema does not declare, is refused before your
+handler sees it.
+
+`effect` decides who may call it. A `read` tool is offered to every connected
+agent. `undoable` (edits that go through the undo history), `journaled` (project
+files written through `ctx.fs.writeProject`, which are captured before they land)
+and `irreversible` (work outside the open project, or code whose effects nobody
+enumerated) are offered only while the person lets agents change the project
+(**Settings › AI Agents**), and refused otherwise.
Nothing verifies the field, and that is not a gap — your plugin already runs
with the whole editor surface in reach, so it could do the same work through a
command. The boundary is the trust prompt at install. What the field buys is
-the confirm gate behaving correctly for an honest plugin, so declare it
-honestly: a tool that calls itself `journaled` while writing outside `ctx.fs`
-claims a safety net that is not holding it.
+the permission behaving correctly for an honest plugin, so declare it honestly:
+a tool that calls itself `read` while changing the project runs for an agent
+the person only let look.
The name must be namespaced with your plugin id, and may contain only letters,
digits, `_` and `-` — up to 64 of them. Both halves of that matter. An
diff --git a/docs/astro/src/content/docs/index.mdx b/docs/astro/src/content/docs/index.mdx
index 4e007bbb7..07e419ed5 100644
--- a/docs/astro/src/content/docs/index.mdx
+++ b/docs/astro/src/content/docs/index.mdx
@@ -58,6 +58,6 @@ import { Card, CardGrid, LinkCard } from '@astrojs/starlight/components';
<LinkCard title="Sprites & Rendering" href="/docs/graphics/sprites/" description="Draw images — tint, size, sort, and layer them." />
<LinkCard title="Physics" href="/docs/gameplay/physics/" description="Rigid bodies, colliders, joints, and spatial queries." />
<LinkCard title="Gameplay AI" href="/docs/gameplay/ai/overview/" description="Navigation, perception, state machines, behavior trees." />
- <LinkCard title="AI Agents" href="/docs/agents/built-in/" description="Ask the editor to build it — from the built-in agent or your own MCP client." />
+ <LinkCard title="AI Agents" href="/docs/agents/mcp/" description="Connect your own AI tool over MCP and let it build in the editor." />
<LinkCard title="Building & Exporting" href="/docs/publishing/overview/" description="Ship to web, desktop, WeChat, playable ads, Android and iOS." />
</CardGrid>
diff --git a/docs/astro/src/content/docs/zh-cn/agents/built-in.mdx b/docs/astro/src/content/docs/zh-cn/agents/built-in.mdx
deleted file mode 100644
index 9123f3957..000000000
--- a/docs/astro/src/content/docs/zh-cn/agents/built-in.mdx
+++ /dev/null
@@ -1,150 +0,0 @@
----
-title: 内置 Agent
-description: 用一句话让编辑器干活。Agent 会读你的场景,并通过编辑器 UI 自己用的那套工具来改动它——每一轮外面都套着一个撤销检查点,撤销够不到的事情则一律先问过你。
----
-
-import { Aside, Steps } from '@astrojs/starlight/components';
-
-编辑器自带一个 Agent。你打一句话,它读场景、调用编辑器 UI 调用的同一批工具,
-并把每一次调用都摊在你眼前。它不是编辑器的另一份复制品:它驱动的工具目录,
-和[外部 MCP 客户端](/docs/zh-cn/agents/mcp/)驱动的是同一份——它能做什么,
-和编辑器能做什么,是一张清单。
-
-<Aside type="tip">
-它就在你原本工作的那个窗口里。它的改动落在你的场景上,你的撤销能收回它们,
-你随时可以把鼠标拿回来。
-</Aside>
-
-## 先给它一个模型
-
-编辑器不附带任何密钥,也不会自己发起任何调用。打开**设置 ▸ 编辑器 ▸ 工具与 AI ▸ AI 代理**,
-配好一个提供方,配置就算做完了。
-
-![设置 ▸ 编辑器 ▸ 工具与 AI ▸ AI 代理——你自己的提供方表:每一行有名称、协议、地址、模型名和各自的密钥](@/assets/guides/zh/agent-settings.png)
-
-<Steps>
-
-1. 填入 Anthropic、DeepSeek 或 OpenAI 的 **API 密钥**。每个提供方各存各的,
- 所以切回上周用过的那个不用重新输入。
-
-2. 有些提供方还要你自己填**模型名**。这是故意的:编辑器内置的列表只能正确到
- 厂商下次发布为止,而写了个已经不存在的名字并不会被拒绝——接口会悄悄改用一个
- 更小的模型顶上,并且一整个会话都是如此。
-
-3. 或者在**你的提供方**(展开该页的**高级**折叠)里添加自己的——任何接口都行,有几个加几个。每一行
- 都和内置的提供方一样:名称、讲哪种**协议**、地址、模型名,以及它自己的密钥。
- 展开一行还能说明这个接口**能做什么**。
-
-4. 关掉设置。模型选择器就在你打字的那个框旁边——它是**逐条消息**的选择,
- 而不是埋在对话框里的偏好设置。
-
-</Steps>
-
-<Aside type="tip" title="该选哪种协议">
-**Chat Completions** 是 OpenAI 的协议,而且几乎其他所有东西也都实现了它——
-其他厂商、聚合网关,以及所有本地运行器(Ollama、vLLM、LM Studio)。
-**Messages** 是 Anthropic 的协议;模拟它的网关在这里同样能用,
-只是少了那些属于 Anthropic 自己的扩展(扩展思考、提示缓存、自动回退)。
-两种格式发出的每一条消息都不一样,所以选错会被直接拒绝,而不是悄悄降级运行。
-</Aside>
-
-<Aside type="tip" title="密钥存在哪">
-存在系统钥匙串里。它不写进项目、不写进设置文件,也不会交回给编辑器窗口——
-只交给那个必须发送它的客户端。
-</Aside>
-
-## 接口能做什么
-
-内置的提供方已经自己写好了。你自己加的那些,编辑器没有办法探出来,而猜错的
-两个方向代价都不小——所以一行的展开项就是你告诉它的地方。没说的部分不会被
-发现,只会被绕开。
-
-- **接受图片**——Agent 会截图来检查自己干的活。发不出去图片时,它会改成把
- 同一帧要成粗颗粒的文字色块网格,并在对话里说明。默认关闭:给一个其实看不了
- 图的接口打开它,代价是整轮请求被拒;而不打开,只是少了些它多半能不依赖的细节。
-- **上下文窗口**——一段对话能长到多大,才开始把最早的几轮折叠掉。留空按一个
- 保守值算,会把本来还有余量的对话提前折叠;写得太大则是整轮请求直接失败,
- 而且是在你等完之后。
-- **接受思考深度参数**——请求里可不可以带上上面设的那个深度。没听说过这个参数
- 的接口会拒掉整次调用,而不是忽略它——遇到这种就把它关掉。
-
-## 问它点什么
-
-打开命令面板(`Ctrl/Cmd+Shift+P`)执行 **Agent:打开**。或者干脆跳过这一步:
-直接把问题打进命令面板。凡是读起来像一句话而不像命令的输入,面板都会给出
-**交给 Agent**,按 `Enter` 就发出去了。
-
-![一轮跑完的 Agent 对话:问题、两次工具调用,以及回答](@/assets/guides/zh/agent-turn.png)
-
-一轮对话从上往下读:
-
-- **问题**,旁边是回答它的模型、耗时,以及花掉的 token。
-- **每次工具调用一行**——`get_scene_tree`、`get_event_bindings`——带上参数和
- 返回的条数。这部分才是值得读的:它决定了你拿到的是**关于你这个场景**的回答,
- 还是一段泛泛而谈。
-- **回答**。
-
-在输入框里,用 `@` 可以按名字引用实体,省得你去描述指的是哪一个。你也可以
-拖入或粘贴**图片**——一张设计稿、一张 bug 截图——在接口支持图片时它会随问题
-一起发出去。它干活期间,状态栏上会有一条实时的段落,带**停止**按钮,
-所以抽屉关着也看得见进度、也随时能叫停。
-
-## 哪些事它不用问就能做
-
-分级是按**撤回在哪里失效**划的,这也是它值得你知道的原因:
-
-| 层级 | 例子 | 会发生什么 |
-| --- | --- | --- |
-| 只读 | `get_scene_tree`、`get_inspector`、`capture_viewport` | 直接跑,从不询问。 |
-| 可撤销 | 新建、删除、`set_field`、改父子、刷瓦片 | 直接跑。整一轮外面套着一个检查点,一次**撤销**就能全退。 |
-| 项目文件 | `write_project_file`、`save_scene`、新建脚本、场景或资产 | 直接跑。编辑器会保存每个文件改动前的样子,所以历史面板里这一轮的**撤回**会把文件和场景一起退回。 |
-| 不可逆 | 新建项目、导出构建、`run_editor_command`、探测运行中的游戏 | **一定先问**。事后什么都救不回来。 |
-
-所以你给出的许可并不是逐条改动的许可——它就是撤销按钮,而且作用于**整一轮**。
-一轮跑完且确实改了东西时,记录下面会留一条**撤销 / 保留**的横条;
-如果你自己之后又编辑过,它会提示你——因为那时撤销会把你自己的活也一并退回。
-
-## 大批改动落地之前
-
-有一个可撤销的工具仍然会停下来问,而且不是出于安全:`apply_scene_ops`
-一次调用就能搭出一整棵子树,上百个节点的改动值得在落地**之前**看一眼,
-而不是事后再读回来。
-
-![一批场景改动的预览:列出 6 项改动,每一行都能划掉,下面是全部应用 / 本轮都允许 / 跳过](@/assets/guides/zh/agent-preview.png)
-
-每一行就是一项改动。点一行可以**划掉**它——划掉一个新建实体时,
-引用它的行会一并划掉——然后应用剩下的。**本轮都允许**会让它在这一轮里
-不再询问,这正是让一次长时间搭建不至于变成一百次弹窗的东西;下一轮会重新问。
-
-拒绝也不是死路:Agent 会被告知这一步被跳过了,然后绕开它继续。
-
-## 对话跟着项目一起保存
-
-跑完的对话会和项目存在一起——记录,以及模型对它的记忆。**历史对话**里列着它们,
-挑一段就能从当时停下的地方接着聊。在一段打开的对话里继续提问,
-前面建立起来的东西都还在。
-
-一段长对话里有两件事值得知道:
-
-- 抽屉顶部的**上下文刻度**表示模型的窗口用掉了多少。超过阈值后,
- 最早的几轮会被*折叠出它的记忆*:你问过的话原样保留,工具调用和结果不保留。
- 记录里仍然显示着它们——只是模型读不到了。发生这件事时,抽屉会在发生的那一行说出来。
-- **一段对话属于回答它的那个模型。** 切换模型会结束当前这段并重新开始。
- 它对场景做过的改动仍然保留。
-
-你还可以**换个说法重新提问**某一轮:它会丢弃那一轮以及之后的所有内容,然后重问。
-当三条消息之前就跑偏了、而你不想再靠聊天把它拉回来时,这个很好用。
-
-<Aside type="caution">
-和任何模型一样,它可能非常自信地说错——最常见的是关于它好几轮之前读到的状态。
-当一个回答是在**描述**你的场景而不是在推理时,让它重新读一遍,别信那段转述。
-这也正是工具调用行摆在明面上的原因:它们显示的是它真正看过的东西。
-</Aside>
-
-## 参见
-
-- [AI 代理 (MCP)](/docs/zh-cn/agents/mcp/) —— 同一份工具目录,
- 只不过是从 Claude Code、Cursor 或你自己的代理去驱动,而不是从编辑器内部。
-- [事件绑定](/docs/zh-cn/scripting/events/) —— 你要求"不写代码的行为"时,Agent 会去用的东西。
-- [预制体](/docs/zh-cn/world/prefabs/) —— 在预制体模式下,Agent 编辑的是预制体而不是场景。
- 要它做大改动之前,先确认它正在编辑的是哪个文档。
diff --git a/docs/astro/src/content/docs/zh-cn/agents/mcp.mdx b/docs/astro/src/content/docs/zh-cn/agents/mcp.mdx
index 910177e1d..9477e3996 100644
--- a/docs/astro/src/content/docs/zh-cn/agents/mcp.mdx
+++ b/docs/astro/src/content/docs/zh-cn/agents/mcp.mdx
@@ -111,7 +111,7 @@ claude mcp add estella --env ESTELLA_MCP_ALLOW_WRITES=1 -- \
## 代理能做什么
-常用的工具(约五十个:场景、Play、打开和保存、截图、日志)一开始就列出;其余分在若干组里,代理需要时用 `enable_tools` 打开:`assets`、`prefab`、`authoring`、`view`、`play`、`render`、`project`、`code`、`device` 和 `agent`。第一次调用之前就把全部工具交过去,每个会话都要多花几千 token,也更难挑出合适的那个;不管列没列出,任何工具都可以直接按名字调用。客户端跟不上工具列表变化的,加 `--all-tools`。下表列出的工具来自所有组。
+常用的工具(约五十个:场景、Play、打开和保存、截图、日志)一开始就列出;其余分在若干组里,代理需要时用 `enable_tools` 打开:`assets`、`prefab`、`authoring`、`view`、`play`、`render`、`project`、`code` 和 `device`。第一次调用之前就把全部工具交过去,每个会话都要多花几千 token,也更难挑出合适的那个;不管列没列出,任何工具都可以直接按名字调用。客户端跟不上工具列表变化的,加 `--all-tools`。下表列出的工具来自所有组。
| 领域 | 工具(节选) |
| --- | --- |
diff --git a/docs/astro/src/content/docs/zh-cn/editor/history-scatter-health.mdx b/docs/astro/src/content/docs/zh-cn/editor/history-scatter-health.mdx
index 1a5dcef07..5edccd571 100644
--- a/docs/astro/src/content/docs/zh-cn/editor/history-scatter-health.mdx
+++ b/docs/astro/src/content/docs/zh-cn/editor/history-scatter-health.mdx
@@ -10,13 +10,7 @@ description: 编辑周边的几个面板——历史列表及它如何撤回 Age
**历史**按顺序列出你在当前场景里做过的每一次编辑,**当前状态**标出你现在的位置。它是撤销 / 重做
(`Ctrl/⌘ + Z`、`Ctrl/⌘ + Shift + Z`)的可视化形式:点一个更早的行就回到那个状态。
-[内置 Agent](/docs/zh-cn/agents/built-in/)的一次运行会折叠成**一行**,以提示词命名(**Agent · …**)。
-这一行有**撤回**按钮,把**整次运行**退回——场景改动*和*它写下的项目文件都会退回。
-因为牵涉文件,退回前会先确认,并说明将撤销多少次运行、多少个文件。这个操作无法重做;
-太大而没法留副本的文件会被列出,并保持原样。
-
-运行之后,历史还可能给出**保留这条检查**:这次运行通过的某项验证——场景里没有标出问题,或脚本能编译——
-你可以把它钉进项目。之后每次运行都以它为准,任何一次运行都不能削弱或移除它。
+[通过 MCP 连接的代理](/docs/zh-cn/agents/mcp/)做的编辑也和你自己的一样记在这里,每步一行,同样可以退回。
## 散布笔刷
diff --git a/docs/astro/src/content/docs/zh-cn/editor/settings.mdx b/docs/astro/src/content/docs/zh-cn/editor/settings.mdx
index d507d241a..7d65b1ac2 100644
--- a/docs/astro/src/content/docs/zh-cn/editor/settings.mdx
+++ b/docs/astro/src/content/docs/zh-cn/editor/settings.mdx
@@ -9,8 +9,8 @@ description: "编辑器偏好,以及写入 project.esproject 的项目级设置
- **常规** — 外观(**语言**、强调色、界面字体、UI 缩放)、性能、控制台,以及渲染后端。
- **视口** — 网格、吸附和变换 gizmo。
- **快捷键** — 下面每个快捷键都能在这里重绑。
- - **工具与 AI** — 外部脚本编辑器 / 图片编辑器 / 浏览器,以及内置 [Agent](/docs/zh-cn/agents/built-in/):
- 你自己的密钥、模型和接口。
+ - **工具与 AI** — 外部脚本编辑器 / 图片编辑器 / 浏览器,以及是否允许 [AI 代理](/docs/zh-cn/agents/mcp/)通过 MCP
+ 连接、能否修改项目。
- **项目** — 存进 `project.esproject`,谁打开这个项目都共用:
- **游戏** — **显示**(项目设计分辨率:播种新 Canvas,驱动任意场景的设备预览;项目级的**屏幕方向**:
所有导出目标都遵循它,留空则从设计分辨率的宽高比推导;**相机适配**:主相机如何在无 UI Canvas
diff --git a/docs/astro/src/content/docs/zh-cn/extending/editor-plugins.mdx b/docs/astro/src/content/docs/zh-cn/extending/editor-plugins.mdx
index 6c901eee9..115cdbb67 100644
--- a/docs/astro/src/content/docs/zh-cn/extending/editor-plugins.mdx
+++ b/docs/astro/src/content/docs/zh-cn/extending/editor-plugins.mdx
@@ -370,7 +370,7 @@ await ctx.project.refreshAssets();
### 借一个工具给 Agent
-内置 Agent 的全部词汇就是工具目录。**插件加了能力却没加工具,等于只给人加了、没给 Agent 加**——所以如果你的插件能烘焙遮蔽,就把它教给 Agent:
+通过 [MCP](/docs/zh-cn/agents/mcp/) 连接的代理只能做工具列表里有的事。**插件加了能力却没加工具,等于只给人加了、没给 Agent 加**——所以如果你的插件能烘焙遮蔽,就把它教给 Agent:
```typescript
ctx.agentTools.register({
@@ -386,9 +386,11 @@ ctx.agentTools.register({
`description` **不是文档**——它是模型判断「该不该调这个」时唯一会读的东西。写清楚它做什么、什么时候该用。
-`effect` 决定要不要先问人,分档的界线画在**「还退得回去」到此为止**的地方。`read` 和 `undoable` 直接跑(这一轮的检查点就是许可)。`journaled` 也直接跑——当你的工具通过 `ctx.fs.writeProject` 写项目文件时就声明这一档:这些写入会在落盘前被留底,所以这一轮的「撤回」会把它们连同场景一起收回来。`irreversible` 留给检查点够不着的:动到已打开项目之外,或者跑没人枚举得清后果的代码;它会停下来问。
+这个工具会和编辑器自己的工具一起列给每个连上的 MCP 客户端,调用时在编辑器里执行你的 `run`。先按 `schema` 校验:缺参数、或带了 schema 没声明的参数,都会在交给你之前被拒绝。
-**没有任何东西会去核实这个字段**,而这不是漏洞——你的插件本来就握着整个编辑器接口,同样的事它用一个命令也能做。边界在安装时的信任提示,不在这个字段。这个字段换来的是:对一个诚实的插件,确认门能正确工作。所以请如实声明——一个自称 `journaled` 却绕开 `ctx.fs` 去写盘的工具,是在宣称一张并没有接住它的安全网。
+`effect` 决定谁能调它。`read` 对每个连上的代理都开放。`undoable`(走撤销历史的编辑)、`journaled`(通过 `ctx.fs.writeProject` 写的项目文件,落盘前会留底)和 `irreversible`(动到已打开项目之外,或跑没人枚举得清后果的代码)只在用户允许代理修改项目时(**设置 › AI 代理**)才开放,否则拒绝。
+
+**没有任何东西会去核实这个字段**,而这不是漏洞——你的插件本来就握着整个编辑器接口,同样的事它用一个命令也能做。边界在安装时的信任提示,不在这个字段。这个字段换来的是:对一个诚实的插件,权限能正确起作用。所以请如实声明——一个自称 `read` 却会改项目的工具,会在用户只允许「看」的代理手里照样执行。
名字必须以你的插件 id 开头,且**只能含字母、数字、`_` 和 `-`,最多 64 个**。两半都要紧:不带命名空间的工具可能**盖住内置工具**,而模型在调 `delete_entity` 时是相信自己知道会发生什么的;而字符集之外的名字模型端点根本不收——它拒的是**整个请求**而不是那一个工具,于是这一个工具会把每一次对话都带下水。
diff --git a/docs/astro/src/content/docs/zh-cn/index.mdx b/docs/astro/src/content/docs/zh-cn/index.mdx
index 2e6d377d6..197740690 100644
--- a/docs/astro/src/content/docs/zh-cn/index.mdx
+++ b/docs/astro/src/content/docs/zh-cn/index.mdx
@@ -53,6 +53,6 @@ import { Card, CardGrid, LinkCard } from '@astrojs/starlight/components';
<LinkCard title="精灵与渲染" href="/docs/zh-cn/graphics/sprites/" description="绘制图像——着色、尺寸、排序与分层。" />
<LinkCard title="物理" href="/docs/zh-cn/gameplay/physics/" description="刚体、碰撞体、关节与空间查询。" />
<LinkCard title="游戏 AI" href="/docs/zh-cn/gameplay/ai/overview/" description="导航、感知、状态机、行为树。" />
- <LinkCard title="AI 智能体" href="/docs/zh-cn/agents/built-in/" description="让编辑器替你搭——用内置智能体,或你自己的 MCP 客户端。" />
+ <LinkCard title="AI 智能体" href="/docs/zh-cn/agents/mcp/" description="用 MCP 连接你自己的 AI 工具,让它在编辑器里替你搭。" />
<LinkCard title="构建与导出" href="/docs/zh-cn/publishing/overview/" description="发布到 Web、桌面、微信、Playable 广告、Android 与 iOS。" />
</CardGrid>
diff --git a/editor-api/index.ts b/editor-api/index.ts
index 0fa9b7344..886eada26 100644
--- a/editor-api/index.ts
+++ b/editor-api/index.ts
@@ -400,7 +400,7 @@ export interface ContextMenuTarget {
}
/**
- * A tool the built-in agent may call.
+ * A tool a connected (MCP) agent may call.
*
* The agent's entire vocabulary is the tool catalog, so a plugin that adds a
* capability without adding a tool has added it for the person and not for the
@@ -424,15 +424,17 @@ export interface AgentToolContribution {
description: string;
/**
* JSON Schema for `input`. Defaults to "an object with no properties".
- * The kernel validates against this before your handler runs, so a
- * malformed call is refused rather than handed to you half-parsed.
+ * The MCP front validates against this before your handler runs, so a
+ * malformed call — or an argument the schema does not declare — is refused
+ * rather than handed to you half-parsed.
*/
schema?: unknown;
/**
- * What calling it costs, which decides whether the person is asked first.
- * The turn's checkpoint covers `read`, `undoable` and `journaled`, so all
- * three run unasked; `journaled` is for writes through `ctx.fs.writeProject`,
- * which are captured before they land. `irreversible` stops to ask.
+ * What calling it costs, which decides who may call it. A `read` tool is
+ * offered to every connected agent; anything else is offered only while the
+ * person lets agents change the project (Settings > AI Agents), and refused
+ * otherwise. `journaled` is for writes through `ctx.fs.writeProject`, which
+ * are captured before they land.
*
* Declare it honestly — nothing verifies it. A tool calling itself
* `journaled` while writing outside `ctx.fs` claims a net not holding it.
@@ -658,7 +660,7 @@ export interface PluginContext {
readonly contextMenus: {
register(item: ContextMenuContribution): Disposable;
};
- /** Lend the built-in agent a tool. See {@link AgentToolContribution}. */
+ /** Lend connected (MCP) agents a tool. See {@link AgentToolContribution}. */
readonly agentTools: {
register(tool: AgentToolContribution): Disposable;
};
diff --git a/plugins/ldtk/README.md b/plugins/ldtk/README.md
index 20f33d041..45d76da4e 100644
--- a/plugins/ldtk/README.md
+++ b/plugins/ldtk/README.md
@@ -33,7 +33,7 @@ larger one.
Nothing here reaches into the engine: it is an editor plugin using
`ctx.assets.registerImporter`, `ctx.fs.readProject` / `writeProject` (which is why
-it declares the `fs:project` capability), and one agent tool so the built-in agent
+it declares the `fs:project` capability), and one agent tool so an agent connected over MCP
can run the import too. The conversion itself is a pure function over JSON —
`src/convert.ts`, tested on its own.
diff --git a/tools/check-dirty-source.mjs b/tools/check-dirty-source.mjs
index 32891a447..51e5ca40e 100644
--- a/tools/check-dirty-source.mjs
+++ b/tools/check-dirty-source.mjs
@@ -32,7 +32,7 @@ import { dirname, join, resolve, relative } from 'node:path';
const ROOT = resolve(dirname(fileURLToPath(import.meta.url)), '..');
-/** The automation surface: what MCP and the built-in agent drive the editor through. */
+/** The automation surface: what MCP agents drive the editor through. */
const SURFACE = join(ROOT, 'desktop', 'src', 'main.tsx');
/** Where the aggregate is defined, and where the rule is written down. */
const REGISTRY = join(ROOT, 'desktop', 'src', 'document', 'DirtyRegistry.ts');
diff --git a/tools/check-mutator-parity.mjs b/tools/check-mutator-parity.mjs
index 3955bf6d1..390d7779d 100644
--- a/tools/check-mutator-parity.mjs
+++ b/tools/check-mutator-parity.mjs
@@ -4,8 +4,8 @@
* @file check-mutator-parity.mjs — automation may not reach a project setting a
* person cannot.
*
- * The editor has a built-in agent, and every project mutator is exposed to it
- * through one facade. That is the right shape until a mutator has ONLY that
+ * Connected agents drive the editor through MCP, and every project mutator is
+ * exposed to them through one facade. That is the right shape until a mutator has ONLY that
* caller: `setAudio` writes bus volumes, effects and duck rules, is reachable
* from the automation surface, and has no button, row or panel anywhere — so an
* agent can author the project mixer and the person who owns the project cannot.
diff --git a/tools/check-project-settings.mjs b/tools/check-project-settings.mjs
index 28af24036..b653cac00 100644
--- a/tools/check-project-settings.mjs
+++ b/tools/check-project-settings.mjs
@@ -368,7 +368,7 @@ const CONSUMPTION = {
* row looks like every other one, so silence still reads as "this works". */
disclaimedBy: 'set.project.spine.version.desc',
/**
- * The readers that are not the settings UI. The built-in agent's tool
+ * The readers that are not the settings UI. An agent tool's
* description promised "the build bundles that runtime", which matches no
* BUILD_PROMISE pattern — enumerating the ways to lie does not work. So the
* rule is the row's: wherever it is described, it says what it does not do.
diff --git a/tools/check-theme.mjs b/tools/check-theme.mjs
index 964e1461f..a5150e74a 100644
--- a/tools/check-theme.mjs
+++ b/tools/check-theme.mjs
@@ -157,18 +157,6 @@ const IDENTITIES = [
names: "the viewport's selection, gizmos and axes",
places: NAMES_THE_SAME_THING,
},
- {
- token: /^--nebula(?:-soft)?$/,
- names: 'the built-in agent',
- places: [
- ['.ag-', 'the agent drawer'],
- ['.md-', "the agent's rendered replies"],
- ['.row.agent-', 'Outliner rows the agent touched'],
- ['.act-agent', "the agent's activity-bar button"],
- ['.viewport__agentpeek', 'what the agent is pointing at in the viewport'],
- ['@keyframes ag-', "the agent's own animations"],
- ],
- },
{
token: /^--info-warm(?:-soft|-line)?$/,
names: 'a prefab',
@@ -531,7 +519,6 @@ const ICON_BUTTONS_KEPT = {
'tab-popout': "dock tab chrome, styled with the dock's tabs",
'tab-x': "dock tab chrome, styled with the dock's tabs",
'dv-collapse': "dock group chrome, styled with the dock's tabs",
- 'ag-send': "the composer's send: the drawer's accent-filled primary action",
'awp-play': "an audio preview's play: the accent-outlined primary control",
'transport__play': "the transport's play: the accent-outlined primary control",
'ts-zone': 'a cell of the tile being edited; the cell is the content, not an icon',
@@ -547,7 +534,6 @@ const ICON_BUTTONS_KEPT = {
'prop-reset': 'resets a changed property, in the modified colour, only when there is a change',
'ctrl-chip-x': "a chip's remove, in the chip's own colour, shown on hover",
'proj-card__remove': 'a remove badge placed over a project card',
- 'ag-att-x': 'a remove badge placed over an attachment thumbnail',
'lc-field__browse': "the browse end of a path field, part of the field's well",
'build__scene-start': 'marks the start scene; its state is its fill',
'ovbtn': "a viewport overlay menu trigger; the rest of its family carry text",