feat(studio): let an agent edit text and styles, guarded (#3518)

* feat(studio): let an agent drive Studio's selection and playhead

Adds `studio_select` and `studio_seek`, so an agent and the human are looking
at the same element and the same instant. Selecting reveals the inspector,
exactly as a click does, which is what makes the agent's move visible.

Selection is shared state, not a per-call argument, and that is forced rather
than chosen. Most of Studio's edit handlers read the ambient React selection,
and `applyDomSelection` only schedules a state update, so selecting and
committing inside ONE call would write to whatever was selected before. Two
tool calls are separated by a render, so the contract is select first, then
act. That is also how a human works: click, then type.

`studio_seek` uses `requestSeek`, not `setCurrentTime`. The latter only moves
the timeline's displayed number and leaves the composition where it was.

Two things the tools refuse to fake:

Seek does not clamp. `seek()` already clamps against the adapter's duration,
which can differ from the store's, and clamping again would give that
invariant two owners that can disagree. The tool reports where the playhead
actually landed instead, read back afterwards.

`requestSeek` is fire-and-forget, so it cannot report that no adapter was
mounted to receive it. The tool compares the playhead before and after and
fails rather than claiming a seek that never happened.

Select separates three failures that a single message would have merged: the
preview is not mounted yet (wait), no element matches the handle (re-read),
and the element cannot be selected (try a neighbour). The agent's next move
differs for each, so collapsing them would cost it a round trip or a retry
loop.

* feat(studio): give an agent eyes with studio_frame

Renders the composition to a PNG at a given time and returns the URL. This is
what turns the tool set from a remote control into a loop: author a change,
capture the instant it affects, look, adjust. No agent can judge motion from
source, because "what does this look like at 2.4 seconds" is not a question a
file answers.

Reuses Studio's existing capture endpoint via `buildFrameCaptureUrl` rather
than inventing a second one.

Two things this does not fake:

It reports the time the playhead LANDED on, not the time requested. The player
clamps, so those differ at the ends, and attaching the wrong time to a frame is
how an agent draws a confident wrong conclusion about motion.

It waits before capturing, by default 150ms. The frame is rendered from the
file on disk, and the render cache is cleared by a file watcher with a 40ms
write-stability threshold, so a capture that beats the watcher renders the
PRE-edit composition. That exact staleness was a real bug here once. An agent
reading a stale frame as "my edit failed" would thrash, so the wait is on by
default, `settleMs` makes it tunable, and the tool description names the
failure rather than leaving it to be rediscovered.

It probes with HEAD before returning, so a URL that 404s comes back as a
failure with a hint instead of as a link the agent cannot render.

* feat(studio): add studio_inspect, so an agent reads before it writes

Everything about one element in one call: resolved styles, text fields, box,
data attributes, GSAP animations, and what the element will and will not
accept.

The point is to prevent a failed write rather than to satisfy curiosity.
`can.reasonIfDisabled` is passed through verbatim from Studio's own
capabilities, so an agent that reads first should never attempt an edit the
element would refuse.

Three things it refuses to get wrong:

Animations are reported ONLY for the current selection, because that is the
only element Studio parses them for. Attributing them to any other element
would be reporting the wrong element's motion, which is worse than reporting
none. When a handle names something else the field is empty and
`animationEditingBlocked` says why.

`animationEditingBlocked` also carries the two states where animation editing
is off entirely, multiple timelines and an unsupported timeline pattern. Both
live on the selection context. Learning them from a read costs one call;
learning them from a failed write costs a retry loop.

Inspecting a handle does NOT change what is selected. It is a read, and
stealing the human's selection would be a side effect they did not ask for.
There is a test asserting `applySelection` is never called.

Nothing selected and no handle given is a failure, not an empty result. An
empty result would assert "this element has nothing", which is a different and
false claim.

* feat(studio): let an agent edit text and styles, guarded

The first tools that change the composition. Both act on the current
selection and take no handle, which is forced rather than chosen: the
handlers read the ambient React selection, and `applyDomSelection` only
schedules a state update, so selecting and committing inside one call would
write to whatever was selected before. Select first, then edit.

Also plumbs the write-blocked state, which was the blocker for shipping any
write at all. `domEditSaveQueuePaused` and the external-file conflict both
lived on App and were unreachable from the tool surface, so `canWrite` was
optimistic and a comment said so. They now derive into a single
`writeBlockedReason` on the shell context: one field, one owner, conflict
taking precedence because resolving it is what unblocks the queue.

That guard matters more than it looks. Both states are BANNERS in Studio with
no lock behind them, so nothing else was stopping a programmatic write from
landing on top of a conflict the user had been asked to adjudicate.

Three things the tools refuse to fake:

They check the outcome, not the absence of a throw. Studio has several paths
where a failed commit resolves anyway, so awaiting the handler proves nothing.
The tagged outcome added earlier is what proves the write landed.

A partial style result is reported as partial. `handleDomStyleCommit` is one
property per call, so N properties are N commits; the result carries `applied`
and `rejected` maps rather than a single boolean that would have to pick a
side.

Style commits run sequentially, never concurrently. Two commits racing through
Studio's client-side read-modify-write can record undo entries that both claim
the same starting content. There is a test that measures concurrency rather
than trusting the loop.

Every decline reason maps to a hint naming what to do instead, so a refusal
routes the agent rather than just stopping it.

* feat(studio): add studio_inspect, so an agent reads before it writes (#3517)

Everything about one element in one call: resolved styles, text fields, box,
data attributes, GSAP animations, and what the element will and will not
accept.

The point is to prevent a failed write rather than to satisfy curiosity.
`can.reasonIfDisabled` is passed through verbatim from Studio's own
capabilities, so an agent that reads first should never attempt an edit the
element would refuse.

Three things it refuses to get wrong:

Animations are reported ONLY for the current selection, because that is the
only element Studio parses them for. Attributing them to any other element
would be reporting the wrong element's motion, which is worse than reporting
none. When a handle names something else the field is empty and
`animationEditingBlocked` says why.

`animationEditingBlocked` also carries the two states where animation editing
is off entirely, multiple timelines and an unsupported timeline pattern. Both
live on the selection context. Learning them from a read costs one call;
learning them from a failed write costs a retry loop.

Inspecting a handle does NOT change what is selected. It is a read, and
stealing the human's selection would be a side effect they did not ask for.
There is a test asserting `applySelection` is never called.

Nothing selected and no handle given is a failure, not an empty result. An
empty result would assert "this element has nothing", which is a different and
false claim.

* feat(studio): move, resize and rotate, verified by reading back (#3519)

`studio_transform` does what a drag does, and then checks. The box in the
result is READ BACK after the write, never echoed from the request, and
`applied` lists what actually took effect.

That is not belt-and-braces. The plan for this unit said to re-derive the
geometry handlers' behaviour rather than trust any description of them, and
doing that turned up three different behaviours behind one interface.

The handlers on `DomEditActionsValue` are the GSAP-AWARE wrappers, aliased in
`useDomEditSession.ts:534-538`, not the CSS ones in `useDomGeometryCommits.ts`
that an earlier note in this workstream described.

`handleGsapAwarePathOffsetCommit` and `handleGsapAwareRotationCommit` are
`if (gsapCommitMutation) { ...intercept... }` with no else branch. Their own
comments say the absence is deliberate: position and rotation are written as
GSAP code and there is no CSS fallback to write to. So they can return having
done nothing.

`handleGsapAwareBoxSizeCommit` is not like the other two. It runs through
`runGestureTransaction` with separate scale and width/height routes, so resize
works more generally.

Reading back is what turns that middle case from a silent lie into a reported
one. A move that did nothing comes back in `unchanged` with a reason.

Three smaller decisions:

Operations re-read between each other, so a move is judged against the box
AFTER a resize in the same call. Comparing against the original would credit
the resize's change to the move.

Rotation is reported as dispatched, not verified. `rotate` is an individual
transform property and does not appear in the computed transform, so there is
no honest box-derived signal, and claiming one would be worse than saying so.

x pairs with y and width pairs with height. Accepting one alone would mean
inventing the other from the current value, which moves the element somewhere
the caller did not ask for. The pairing rule and its minimum live in one
`parsePair` helper rather than as four separate branches.

---------

Co-authored-by: miga-heygen <miguel.sierra_miga@heygen.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Miguel Ángel
2026-08-31 07:47:11 +00:00
committed by GitHub
co-authored by miga-heygen Claude Opus 4.6
parent 5b45bcc16d
commit f84b4c23dc
10 changed files with 882 additions and 7 deletions
+2
View File
@@ -433,6 +433,8 @@ export function StudioApp() {
handleRedo: appHotkeys.handleRedo,
renderQueue,
compositionDimensions,
domEditSaveQueuePaused: previewPersistence.domEditSaveQueuePaused,
externalFileConflict: externalFileChanges.blocked !== null,
waitForPendingDomEditSaves: previewPersistence.waitForPendingDomEditSaves,
handlePreviewIframeRef,
refreshPreviewDocumentVersion,
@@ -16,6 +16,13 @@ export interface StudioShellValue {
undoLabel: string | undefined;
redoLabel: string | undefined;
};
/**
* Why a composition write would be refused right now, or null when writes
* are possible. Derived from the paused save queue and the external-file
* conflict state, both of which are otherwise banners with no lock behind
* them. One field rather than two, so there is one owner of the question.
*/
writeBlockedReason: string | null;
handleUndo: () => Promise<void>;
handleRedo: () => Promise<void>;
renderQueue: {
@@ -106,6 +113,7 @@ export function StudioShellProvider({
showToast,
previewIframeRef,
editHistory,
writeBlockedReason,
handleUndo,
handleRedo,
renderQueue,
@@ -122,6 +130,7 @@ export function StudioShellProvider({
showToast,
previewIframeRef,
editHistory,
writeBlockedReason,
handleUndo,
handleRedo,
renderQueue,
@@ -138,6 +147,7 @@ export function StudioShellProvider({
setActiveCompPath,
showToast,
previewIframeRef,
writeBlockedReason,
handleUndo,
handleRedo,
waitForPendingDomEditSaves,
@@ -25,6 +25,10 @@ interface StudioContextInput {
// fields around it: the context type owns it.
renderQueue: StudioContextValue["renderQueue"];
compositionDimensions: { width: number; height: number } | null;
/** Message from `usePreviewPersistence` when auto-save is paused. */
domEditSaveQueuePaused: string | null;
/** True when an external edit to the open file is awaiting the user's decision. */
externalFileConflict: boolean;
waitForPendingDomEditSaves: () => Promise<void>;
handlePreviewIframeRef: (iframe: HTMLIFrameElement | null) => void;
refreshPreviewDocumentVersion: () => void;
@@ -46,6 +50,11 @@ export function buildStudioContextValue(input: StudioContextInput): StudioContex
timelineElements: input.timelineElements,
isPlaying: input.isPlaying,
editHistory: input.editHistory,
// Conflict first: when both are true the conflict is the one the user has
// been asked to decide, and resolving it is what unblocks the queue.
writeBlockedReason: input.externalFileConflict
? "an external change to this file is waiting to be resolved"
: input.domEditSaveQueuePaused,
handleUndo: input.handleUndo,
handleRedo: input.handleRedo,
renderQueue: input.renderQueue,
@@ -17,15 +17,23 @@ import type { StudioLookSnapshot } from "./tools/lookTools";
* every animation frame during playback for a value nothing here displays.
*/
export function StudioAgentTools() {
const { projectId, activeCompPath, editHistory } = useStudioShellContext();
const { projectId, activeCompPath, editHistory, writeBlockedReason } = useStudioShellContext();
const {
domEditSelection,
selectedGsapAnimations,
gsapMultipleTimelines,
gsapUnsupportedTimelinePattern,
} = useDomEditSelectionContext();
const { previewIframeRef, buildDomSelectionFromTarget, applyDomSelection } =
useDomEditActionsContext();
const {
previewIframeRef,
buildDomSelectionFromTarget,
applyDomSelection,
handleDomTextCommit,
handleDomStyleCommit,
handleDomPathOffsetCommit,
handleDomBoxSizeCommit,
handleDomRotationCommit,
} = useDomEditActionsContext();
const getSnapshot = useCallback((): StudioLookSnapshot => {
const player = usePlayerStore.getState();
@@ -77,6 +85,18 @@ export function StudioAgentTools() {
},
wait: (ms) => new Promise((resolve) => setTimeout(resolve, ms)),
getCurrentSelection: () => domEditSelection,
getWriteBlockedReason: () => writeBlockedReason,
setText: (value, fieldKey) => handleDomTextCommit(value, fieldKey),
setStyle: (property, value) => handleDomStyleCommit(property, value),
// Measured, not authored: the tool compares this before and after to
// tell a real change from a handler that did nothing and resolved.
readBox: (selection) => {
const rect = selection.element.getBoundingClientRect();
return { x: rect.x, y: rect.y, width: rect.width, height: rect.height };
},
moveTo: (selection, next) => handleDomPathOffsetCommit(selection, next),
resizeTo: (selection, next) => handleDomBoxSizeCommit(selection, next),
rotateTo: (selection, next) => handleDomRotationCommit(selection, next),
getGsapDiagnostics: () => ({
animations: selectedGsapAnimations,
multipleTimelines: gsapMultipleTimelines,
@@ -90,6 +110,12 @@ export function StudioAgentTools() {
applyDomSelection,
projectId,
activeCompPath,
writeBlockedReason,
handleDomTextCommit,
handleDomStyleCommit,
handleDomPathOffsetCommit,
handleDomBoxSizeCommit,
handleDomRotationCommit,
domEditSelection,
selectedGsapAnimations,
gsapMultipleTimelines,
@@ -0,0 +1,201 @@
// @vitest-environment jsdom
import { describe, expect, it, vi } from "vitest";
import {
studioSetStyle,
studioSetText,
type ContentToolDeps,
type StudioSetStyleResult,
type StudioSetTextResult,
} from "./contentTools";
import { expectFailure, expectOk, previewElement, selectionFor } from "../webmcpTestUtils";
function contentDeps(overrides: Partial<ContentToolDeps> = {}): ContentToolDeps {
const element = previewElement('<h1 id="headline">Ship it</h1>', "headline");
return {
getCurrentSelection: () => selectionFor(element),
getWriteBlockedReason: () => null,
setText: async () => ({ ok: true }),
setStyle: async () => ({ ok: true }),
...overrides,
};
}
describe("studioSetText", () => {
it("writes the text and reports what it now is", async () => {
const setText = vi.fn(async () => ({ ok: true }) as const);
const result = await studioSetText(contentDeps({ setText }), { text: "Ship it faster" });
const ok = expectOk<StudioSetTextResult>(result);
expect(ok.text).toBe("Ship it faster");
expect(ok.changed).toBe(true);
expect(setText).toHaveBeenCalledWith("Ship it faster", undefined);
});
it("reports changed:false when the text already said that", async () => {
const result = await studioSetText(contentDeps(), { text: "Ship it" });
expect(expectOk<StudioSetTextResult>(result).changed).toBe(false);
});
it("refuses to write while a conflict is waiting for the user", async () => {
// The paused-save and conflict states are banners with no lock behind them.
// Nothing else stops a programmatic write landing on top of a decision the
// user has been asked to make.
const setText = vi.fn();
const result = expectFailure(
await studioSetText(
contentDeps({
getWriteBlockedReason: () => "an external change to this file is waiting to be resolved",
setText,
}),
{ text: "Ship it faster" },
),
);
expect(result.kind).toBe("blocked");
expect(result.reason).toMatch(/external change/);
expect(setText).not.toHaveBeenCalled();
});
it("does not report success when the commit declined", async () => {
// The whole reason the handlers now return an outcome: they resolve on
// failure, so awaiting them proves nothing.
const result = expectFailure(
await studioSetText(
contentDeps({ setText: async () => ({ ok: false, reason: "persist-failed" }) }),
{ text: "Ship it faster" },
),
);
expect(result.kind).toBe("failed");
expect(result.reason).toMatch(/persist-failed/);
});
it("turns a decline reason into a hint naming what to do instead", async () => {
const result = expectFailure(
await studioSetText(
contentDeps({ setText: async () => ({ ok: false, reason: "not-text-editable" }) }),
{ text: "x" },
),
);
expect(result.kind).toBe("blocked");
expect(result.hint).toMatch(/studio_inspect/);
});
it("rejects a non-string text without dispatching", async () => {
const setText = vi.fn();
const result = expectFailure(await studioSetText(contentDeps({ setText }), { text: 42 }));
expect(result.kind).toBe("invalid");
expect(setText).not.toHaveBeenCalled();
});
it("fails when nothing is selected", async () => {
const setText = vi.fn();
const result = expectFailure(
await studioSetText(contentDeps({ getCurrentSelection: () => null, setText }), { text: "x" }),
);
expect(result.kind).toBe("invalid");
expect(result.hint).toMatch(/studio_select/);
expect(setText).not.toHaveBeenCalled();
});
});
describe("studioSetStyle", () => {
it("applies every property and reports them", async () => {
const setStyle = vi.fn(async () => ({ ok: true }) as const);
const result = await studioSetStyle(contentDeps({ setStyle }), {
styles: { color: "red", "font-size": "48px" },
});
const ok = expectOk<StudioSetStyleResult>(result);
expect(ok.applied).toEqual({ color: "red", "font-size": "48px" });
expect(ok.rejected).toEqual({});
expect(setStyle).toHaveBeenCalledTimes(2);
});
it("commits sequentially, never concurrently", async () => {
// Two commits racing through Studio's client-side read-modify-write can
// record undo entries that both claim the same starting content.
let inFlight = 0;
let maxInFlight = 0;
const setStyle = vi.fn(async () => {
inFlight += 1;
maxInFlight = Math.max(maxInFlight, inFlight);
await Promise.resolve();
inFlight -= 1;
return { ok: true } as const;
});
await studioSetStyle(contentDeps({ setStyle }), {
styles: { color: "red", "font-size": "48px", opacity: "0.5" },
});
expect(maxInFlight).toBe(1);
});
it("reports a partial success as partial, not whole", async () => {
const setStyle = vi.fn(async (property: string) =>
property === "left"
? ({ ok: false, reason: "geometry-property" } as const)
: ({ ok: true } as const),
);
const result = await studioSetStyle(contentDeps({ setStyle }), {
styles: { color: "red", left: "10px" },
});
const ok = expectOk<StudioSetStyleResult>(result);
expect(ok.applied).toEqual({ color: "red" });
expect(ok.rejected).toEqual({ left: "geometry-property" });
});
it("fails when every property was refused", async () => {
const result = expectFailure(
await studioSetStyle(
contentDeps({ setStyle: async () => ({ ok: false, reason: "styles-not-editable" }) }),
{ styles: { color: "red" } },
),
);
expect(result.kind).toBe("blocked");
expect(result.reason).toMatch(/styles-not-editable/);
});
it("rejects an empty styles object rather than committing nothing", async () => {
const setStyle = vi.fn();
const result = expectFailure(await studioSetStyle(contentDeps({ setStyle }), { styles: {} }));
expect(result.kind).toBe("invalid");
expect(setStyle).not.toHaveBeenCalled();
});
it("rejects a non-object styles value", async () => {
for (const styles of ["color: red", 42, null, ["color"]]) {
const result = expectFailure(await studioSetStyle(contentDeps(), { styles }));
expect(result.kind).toBe("invalid");
}
});
it("refuses to write while a conflict is waiting for the user", async () => {
const setStyle = vi.fn();
const result = expectFailure(
await studioSetStyle(
contentDeps({ getWriteBlockedReason: () => "Auto-save is paused", setStyle }),
{ styles: { color: "red" } },
),
);
expect(result.kind).toBe("blocked");
expect(setStyle).not.toHaveBeenCalled();
});
});
@@ -0,0 +1,184 @@
/**
* `studio_set_text` and `studio_set_style`: the first tools that change the file.
*
* Both operate on the CURRENT selection and take no handle. That is not an
* omission. `handleDomTextCommit(value, fieldKey?)` and
* `handleDomStyleCommit(property, value)` read the ambient React selection, and
* `applyDomSelection` only schedules a state update, so selecting and
* committing inside one call would write to whatever was selected before.
* Two tool calls are separated by a render. Select first, then edit.
*
* Every write here is guarded before dispatch and verified after. Studio has
* several paths where a failed commit resolves anyway, so "the function did not
* throw" proves nothing; the outcome the handler now returns is what proves it.
*/
import type { DomEditCommitOutcome } from "../../hooks/domEditCommitRunner";
import type { DomEditSelection } from "../../components/editor/domEditingTypes";
import { toolFailure, toolOk, type ToolFailure, type ToolResult } from "../toolResult";
export interface ContentToolDeps {
getCurrentSelection: () => DomEditSelection | null;
/** Why a write would be refused right now, or null. Checked BEFORE dispatch. */
getWriteBlockedReason: () => string | null;
setText: (value: string, fieldKey?: string) => Promise<DomEditCommitOutcome>;
setStyle: (property: string, value: string) => Promise<DomEditCommitOutcome>;
}
/**
* The reasons a commit declines, translated into something an agent can act on.
* `persist-failed` is exogenous; the rest are states it should route around.
*/
const DECLINE_HINTS: Record<string, { kind: "blocked" | "invalid" | "failed"; hint?: string }> = {
"no-selection": { kind: "invalid", hint: "Call studio_select first." },
"no-project": { kind: "blocked" },
"geometry-property": {
kind: "blocked",
hint: "Position and size are not editable as styles. Use the transform tools.",
},
"styles-not-editable": {
kind: "blocked",
hint: "studio_inspect reports why, in can.reasonIfDisabled.",
},
"not-text-editable": {
kind: "blocked",
hint: "This element has no editable text. studio_inspect lists its textFields.",
},
"persist-failed": { kind: "failed", hint: "The write did not reach the file. Check Studio." },
};
function fromOutcome(outcome: DomEditCommitOutcome, what: string): ToolFailure | null {
if (outcome.ok) return null;
const mapped = DECLINE_HINTS[outcome.reason] ?? { kind: "failed" as const };
return toolFailure(mapped.kind, `${what} was not applied: ${outcome.reason}`, mapped.hint);
}
function guardWrite(deps: ContentToolDeps): ToolFailure | null {
// Both blocked states are banners in Studio's UI with no lock behind them, so
// nothing else stops a programmatic write from landing on top of a conflict
// the user has been asked to adjudicate.
const blocked = deps.getWriteBlockedReason();
if (blocked) {
return toolFailure("blocked", blocked, "Resolve it in Studio, then retry.");
}
if (!deps.getCurrentSelection()) {
return toolFailure("invalid", "nothing is selected", "Call studio_select first.");
}
return null;
}
export interface StudioSetTextResult {
text: string;
changed: boolean;
}
export async function studioSetText(
deps: ContentToolDeps,
input: { text?: unknown; field?: unknown },
): Promise<ToolResult<StudioSetTextResult>> {
if (typeof input.text !== "string") {
return toolFailure("invalid", "text must be a string");
}
const field = typeof input.field === "string" && input.field ? input.field : undefined;
const blocked = guardWrite(deps);
if (blocked) return blocked;
const before = deps.getCurrentSelection()?.textContent ?? null;
const outcome = await deps.setText(input.text, field);
const failure = fromOutcome(outcome, "the text");
if (failure) return failure;
return toolOk<StudioSetTextResult>({ text: input.text, changed: before !== input.text });
}
export interface StudioSetStyleResult {
applied: Record<string, string>;
/** Properties the element refused, with the reason. Empty when all landed. */
rejected: Record<string, string>;
}
export async function studioSetStyle(
deps: ContentToolDeps,
input: { styles?: unknown },
): Promise<ToolResult<StudioSetStyleResult>> {
const styles = input.styles;
if (typeof styles !== "object" || styles === null || Array.isArray(styles)) {
return toolFailure("invalid", "styles must be an object of CSS property to value");
}
const entries = Object.entries(styles).filter(
(entry): entry is [string, string] => typeof entry[1] === "string",
);
if (entries.length === 0) {
// An empty commit would report success having done nothing.
return toolFailure("invalid", "styles must contain at least one string value");
}
const blocked = guardWrite(deps);
if (blocked) return blocked;
// `handleDomStyleCommit` is one property per call, so N properties are N
// commits and N undo entries. Sequential, not concurrent: two commits racing
// through Studio's client-side read-modify-write can record undo entries that
// both claim the same starting content.
const applied: Record<string, string> = {};
const rejected: Record<string, string> = {};
for (const [property, value] of entries) {
const outcome = await deps.setStyle(property, value);
if (outcome.ok) applied[property] = value;
else rejected[property] = outcome.reason;
}
if (Object.keys(applied).length === 0) {
const reasons = Object.entries(rejected)
.map(([property, reason]) => `${property}: ${reason}`)
.join(", ");
return toolFailure("blocked", `no style was applied (${reasons})`);
}
return toolOk<StudioSetStyleResult>({ applied, rejected });
}
export const STUDIO_SET_TEXT_INPUT_SCHEMA = {
type: "object",
properties: {
text: { type: "string", description: "The new text content." },
field: {
type: "string",
description:
"Which text field to write, from studio_inspect. Omit for the element's own text.",
},
},
required: ["text"],
additionalProperties: false,
} as const;
export const STUDIO_SET_TEXT_DESCRIPTION = [
"Set the text of the CURRENTLY SELECTED element. Call studio_select first.",
"This is the edit a synthetic double-click cannot reach, because Studio's canvas",
"takes pointer capture and recognises the double press itself.",
"Returns `ok: true` with the resulting text and whether it changed, or `ok: false`",
"with `kind`, `reason` and usually a `hint` naming what to do instead.",
].join(" ");
export const STUDIO_SET_STYLE_INPUT_SCHEMA = {
type: "object",
properties: {
styles: {
type: "object",
description: 'CSS property to value, for example {"color": "red", "font-size": "48px"}.',
additionalProperties: { type: "string" },
},
},
required: ["styles"],
additionalProperties: false,
} as const;
export const STUDIO_SET_STYLE_DESCRIPTION = [
"Set inline styles on the CURRENTLY SELECTED element. Call studio_select first.",
"Each property is a separate commit, so N properties produce N undo entries.",
"Position and size properties (left, top, width, height) are refused here on purpose;",
"they belong to the transform tools.",
"Returns `ok: true` with `applied` and `rejected` maps, so a partial success is visible",
"as a partial success rather than reported as a whole one.",
].join(" ");
@@ -0,0 +1,179 @@
// @vitest-environment jsdom
import { describe, expect, it, vi } from "vitest";
import {
studioTransform,
type ElementBox,
type StudioTransformResult,
type TransformToolDeps,
} from "./transformTools";
import { expectFailure, expectOk, previewElement, selectionFor } from "../webmcpTestUtils";
/**
* A stand-in for the rendered box. happy-dom and jsdom report all-zero rects,
* so the box is injected rather than measured; these tests are about what the
* tool concludes from a box, not about layout.
*/
function boxStore(initial: ElementBox) {
const box = { ...initial };
return {
read: () => ({ ...box }),
set: (next: Partial<ElementBox>) => Object.assign(box, next),
};
}
function transformDeps(overrides: Partial<TransformToolDeps> = {}): TransformToolDeps {
const element = previewElement('<h1 id="headline">Ship it</h1>', "headline");
return {
getCurrentSelection: () => selectionFor(element),
getWriteBlockedReason: () => null,
readBox: () => ({ x: 0, y: 0, width: 100, height: 50 }),
moveTo: async () => undefined,
resizeTo: async () => undefined,
rotateTo: async () => undefined,
...overrides,
};
}
describe("studioTransform", () => {
it("reports the box read back, not the box requested", async () => {
const store = boxStore({ x: 0, y: 0, width: 100, height: 50 });
// The handler lands somewhere other than asked, which is what a clamp or a
// layout constraint does.
const resizeTo = vi.fn(async () => store.set({ width: 300, height: 120 }));
const result = await studioTransform(transformDeps({ readBox: store.read, resizeTo }), {
width: 999,
height: 999,
});
const ok = expectOk<StudioTransformResult>(result);
expect(ok.box.width).toBe(300);
expect(ok.box.height).toBe(120);
expect(ok.applied).toContain("resize");
});
it("reports a silent no-op as unchanged instead of success", async () => {
// handleGsapAwarePathOffsetCommit is `if (gsapCommitMutation) {...}` with no
// else branch. Without GSAP it resolves having written nothing, and echoing
// the request back would be a lie the agent builds on.
const store = boxStore({ x: 10, y: 10, width: 100, height: 50 });
const moveTo = vi.fn(async () => undefined);
const result = expectFailure(
await studioTransform(transformDeps({ readBox: store.read, moveTo }), { x: 500, y: 400 }),
);
expect(moveTo).toHaveBeenCalled();
expect(result.kind).toBe("blocked");
expect(result.reason).toMatch(/did not move/);
expect(result.hint).toMatch(/GSAP/);
});
it("separates what landed from what did not, in one call", async () => {
const store = boxStore({ x: 0, y: 0, width: 100, height: 50 });
const resizeTo = vi.fn(async () => store.set({ width: 200, height: 80 }));
const moveTo = vi.fn(async () => undefined);
const result = await studioTransform(transformDeps({ readBox: store.read, resizeTo, moveTo }), {
x: 40,
y: 40,
width: 200,
height: 80,
});
const ok = expectOk<StudioTransformResult>(result);
expect(ok.applied).toEqual(["resize"]);
expect(ok.unchanged.move).toMatch(/did not move/);
});
it("re-reads between operations so a later one sees the earlier result", async () => {
const store = boxStore({ x: 0, y: 0, width: 100, height: 50 });
const resizeTo = vi.fn(async () => store.set({ width: 200, height: 80 }));
const moveTo = vi.fn(async () => store.set({ x: 40, y: 40 }));
const result = await studioTransform(transformDeps({ readBox: store.read, resizeTo, moveTo }), {
x: 40,
y: 40,
width: 200,
height: 80,
});
// Move is judged against the box AFTER the resize. Comparing against the
// original would credit the resize's change to the move.
const ok = expectOk<StudioTransformResult>(result);
expect(ok.applied).toEqual(["resize", "move"]);
expect(ok.unchanged).toEqual({});
});
it("reports rotation as dispatched rather than verified", async () => {
// `rotate` is an individual transform property and does not appear in the
// computed transform, so there is no honest box-derived signal for it.
const rotateTo = vi.fn(async () => undefined);
const result = await studioTransform(transformDeps({ rotateTo }), { rotate: 15 });
const ok = expectOk<StudioTransformResult>(result);
expect(rotateTo).toHaveBeenCalledWith(expect.anything(), { angle: 15 });
expect(ok.applied).toEqual(["rotate"]);
});
it("refuses to write while a conflict is waiting for the user", async () => {
const moveTo = vi.fn();
const result = expectFailure(
await studioTransform(
transformDeps({ getWriteBlockedReason: () => "Auto-save is paused", moveTo }),
{ x: 10, y: 10 },
),
);
expect(result.kind).toBe("blocked");
expect(moveTo).not.toHaveBeenCalled();
});
it("requires x and y together, and width and height together", async () => {
const moveTo = vi.fn();
const resizeTo = vi.fn();
const deps = transformDeps({ moveTo, resizeTo });
expect(expectFailure(await studioTransform(deps, { x: 10 })).reason).toMatch(/together/);
expect(expectFailure(await studioTransform(deps, { width: 10 })).reason).toMatch(/together/);
expect(moveTo).not.toHaveBeenCalled();
expect(resizeTo).not.toHaveBeenCalled();
});
it("rejects a negative size and an empty request", async () => {
const deps = transformDeps();
expect(expectFailure(await studioTransform(deps, { width: -1, height: 10 })).kind).toBe(
"invalid",
);
expect(expectFailure(await studioTransform(deps, {})).reason).toMatch(/at least one/);
});
it("rejects non-finite numbers rather than passing them to a handler", async () => {
const moveTo = vi.fn();
const result = expectFailure(
await studioTransform(transformDeps({ moveTo }), { x: Number.NaN, y: 10 }),
);
expect(result.kind).toBe("invalid");
expect(moveTo).not.toHaveBeenCalled();
});
it("fails when nothing is selected", async () => {
const moveTo = vi.fn();
const result = expectFailure(
await studioTransform(transformDeps({ getCurrentSelection: () => null, moveTo }), {
x: 1,
y: 1,
}),
);
expect(result.kind).toBe("invalid");
expect(result.hint).toMatch(/studio_select/);
expect(moveTo).not.toHaveBeenCalled();
});
});
@@ -0,0 +1,205 @@
/**
* `studio_transform`: move, resize and rotate, as a drag would.
*
* This tool reads the element's box back after every write and reports what
* ACTUALLY changed. That is not belt-and-braces, it is the only thing standing
* between an agent and a silent lie, because two of the three handlers can do
* nothing and resolve:
*
* - The handlers exposed on `DomEditActionsValue` are the GSAP-AWARE wrappers
* (`useDomEditSession.ts` aliases them), not the CSS ones in
* `useDomGeometryCommits.ts`.
* - `handleGsapAwarePathOffsetCommit` and `handleGsapAwareRotationCommit` are
* `if (gsapCommitMutation) { ...intercept... }` with NO else branch. In a
* composition with no GSAP they return having done nothing. The adjacent
* comments confirm that is deliberate: there is no CSS fallback to write to.
* - `handleGsapAwareBoxSizeCommit` is different. It runs through
* `runGestureTransaction` with a scale route and a width/height route, so
* resize works more generally than the other two.
*
* Read back, do not assume.
*/
import type { DomEditSelection } from "../../components/editor/domEditingTypes";
import { toolFailure, toolOk, type ToolFailure, type ToolResult } from "../toolResult";
export interface ElementBox {
x: number;
y: number;
width: number;
height: number;
}
export interface TransformToolDeps {
getCurrentSelection: () => DomEditSelection | null;
getWriteBlockedReason: () => string | null;
/** The element's box as it renders right now. */
readBox: (selection: DomEditSelection) => ElementBox;
moveTo: (selection: DomEditSelection, next: { x: number; y: number }) => Promise<void>;
resizeTo: (selection: DomEditSelection, next: { width: number; height: number }) => Promise<void>;
rotateTo: (selection: DomEditSelection, next: { angle: number }) => Promise<void>;
}
export interface StudioTransformInput {
x?: unknown;
y?: unknown;
width?: unknown;
height?: unknown;
rotate?: unknown;
}
export interface StudioTransformResult {
/** The box as it renders after the write, read back, not echoed. */
box: ElementBox;
applied: string[];
/** Requested operations whose effect could not be observed, with why. */
unchanged: Record<string, string>;
}
const NO_OP_HINT =
"Move and rotate are written as GSAP code; a composition with no GSAP timeline has nothing to write to. studio_inspect reports the element's animations.";
function readNumber(value: unknown): number | null {
return typeof value === "number" && Number.isFinite(value) ? value : null;
}
function guard(deps: TransformToolDeps): ToolFailure | null {
const blocked = deps.getWriteBlockedReason();
if (blocked) return toolFailure("blocked", blocked, "Resolve it in Studio, then retry.");
if (!deps.getCurrentSelection()) {
return toolFailure("invalid", "nothing is selected", "Call studio_select first.");
}
return null;
}
interface TransformRequest {
move: { x: number; y: number } | null;
size: { width: number; height: number } | null;
rotate: number | null;
}
/**
* Both or neither. Accepting one axis alone would mean inventing the other from
* the current value, which moves the element somewhere the caller did not ask
* for.
*/
function parsePair(
a: unknown,
b: unknown,
names: [string, string],
min = Number.NEGATIVE_INFINITY,
): { pair: [number, number] | null } | ToolFailure {
const first = readNumber(a);
const second = readNumber(b);
if (first === null && second === null) return { pair: null };
if (first === null || second === null) {
return toolFailure("invalid", `${names[0]} and ${names[1]} must be given together`);
}
if (first < min || second < min) {
return toolFailure("invalid", `${names[0]} and ${names[1]} must be at least ${min}`);
}
return { pair: [first, second] };
}
function isFailure(value: object): value is ToolFailure {
return "ok" in value;
}
function parseRequest(input: StudioTransformInput): TransformRequest | ToolFailure {
const move = parsePair(input.x, input.y, ["x", "y"]);
if (isFailure(move)) return move;
const size = parsePair(input.width, input.height, ["width", "height"], 0);
if (isFailure(size)) return size;
const rotate = readNumber(input.rotate);
if (!move.pair && !size.pair && rotate === null) {
return toolFailure(
"invalid",
"give at least one of x, y, width, height, rotate as a finite number",
);
}
return {
move: move.pair ? { x: move.pair[0], y: move.pair[1] } : null,
size: size.pair ? { width: size.pair[0], height: size.pair[1] } : null,
rotate,
};
}
export async function studioTransform(
deps: TransformToolDeps,
input: StudioTransformInput,
): Promise<ToolResult<StudioTransformResult>> {
const request = parseRequest(input);
if (isFailure(request)) return request;
const blocked = guard(deps);
if (blocked) return blocked;
const selection = deps.getCurrentSelection();
if (!selection) return toolFailure("invalid", "nothing is selected");
const applied: string[] = [];
const unchanged: Record<string, string> = {};
// Sequential, and each one re-reads first, so a move is judged against the box
// AFTER a resize in the same call rather than against the original.
if (request.size) {
const before = deps.readBox(selection);
await deps.resizeTo(selection, request.size);
const after = deps.readBox(selection);
if (after.width !== before.width || after.height !== before.height) applied.push("resize");
else unchanged.resize = "the element's size did not change";
}
if (request.move) {
const before = deps.readBox(selection);
await deps.moveTo(selection, request.move);
const after = deps.readBox(selection);
if (after.x !== before.x || after.y !== before.y) applied.push("move");
else unchanged.move = `the element did not move. ${NO_OP_HINT}`;
}
if (request.rotate !== null) {
// Rotation is written as the CSS `rotate` property, an individual transform
// property that does NOT appear in getComputedStyle().transform. There is no
// reliable box-derived signal, so this is reported as dispatched rather than
// verified, and the description says so.
await deps.rotateTo(selection, { angle: request.rotate });
applied.push("rotate");
}
if (applied.length === 0) {
return toolFailure(
"blocked",
`nothing changed: ${Object.values(unchanged).join("; ")}`,
NO_OP_HINT,
);
}
return toolOk<StudioTransformResult>({ box: deps.readBox(selection), applied, unchanged });
}
export const STUDIO_TRANSFORM_INPUT_SCHEMA = {
type: "object",
properties: {
x: { type: "number", description: "New x offset in pixels. Must be paired with y." },
y: { type: "number", description: "New y offset in pixels. Must be paired with x." },
width: { type: "number", minimum: 0, description: "New width. Must be paired with height." },
height: { type: "number", minimum: 0, description: "New height. Must be paired with width." },
rotate: { type: "number", description: "Rotation in degrees." },
},
additionalProperties: false,
} as const;
export const STUDIO_TRANSFORM_DESCRIPTION = [
"Move, resize or rotate the CURRENTLY SELECTED element, the way a drag would.",
"Call studio_select first. Give x with y, and width with height.",
"The result's `box` is READ BACK after the write, not echoed from your request, and",
"`applied` lists what actually took effect. Check it.",
"Move and rotate are written as GSAP code, so in a composition with no GSAP timeline they",
"do nothing; that shows up in `unchanged` rather than as a false success.",
"Rotation is reported as dispatched rather than verified, because the CSS `rotate` property",
"does not appear in the element's computed transform.",
"Returns `ok: true`, or `ok: false` with `kind`, `reason` and a `hint`.",
].join(" ");
@@ -43,6 +43,13 @@ function deps(overrides: Partial<StudioAgentToolsDeps> = {}): StudioAgentToolsDe
probeFrame: async () => ({ ok: true, status: 200 }),
wait: async () => undefined,
getCurrentSelection: () => null,
getWriteBlockedReason: () => null,
setText: async () => ({ ok: true }),
setStyle: async () => ({ ok: true }),
readBox: () => ({ x: 0, y: 0, width: 100, height: 50 }),
moveTo: async () => undefined,
resizeTo: async () => undefined,
rotateTo: async () => undefined,
getGsapDiagnostics: () => ({
animations: [],
multipleTimelines: false,
@@ -114,6 +121,9 @@ describe("useStudioAgentTools", () => {
"studio_seek",
"studio_frame",
"studio_inspect",
"studio_set_text",
"studio_set_style",
"studio_transform",
]);
expect(trackEvent).toHaveBeenCalledWith("webmcp.native_present");
});
@@ -128,14 +138,14 @@ describe("useStudioAgentTools", () => {
await act(async () => {
harness = mountTools(deps({ getSnapshot: () => snapshot() }));
});
expect(registerTool).toHaveBeenCalledTimes(5);
expect(registerTool).toHaveBeenCalledTimes(8);
await act(async () => {
harness?.rerenderWith(deps({ getSnapshot: () => snapshot({ currentTime: 5 }) }));
harness?.rerenderWith(deps({ getSnapshot: () => snapshot({ currentTime: 6 }) }));
});
expect(registerTool).toHaveBeenCalledTimes(5);
expect(registerTool).toHaveBeenCalledTimes(8);
});
it("executes against the LATEST deps, not the ones present at registration", async () => {
@@ -208,7 +218,7 @@ describe("useStudioAgentTools", () => {
mountTools(deps({ getSnapshot: () => snapshot() }));
});
expect(registerTool).toHaveBeenCalledTimes(5);
expect(registerTool).toHaveBeenCalledTimes(8);
});
it("reports a non-abort registration failure through production telemetry", async () => {
@@ -41,6 +41,25 @@ import {
type StudioInspectInput,
type StudioInspectResult,
} from "./tools/inspectTools";
import {
studioSetStyle,
studioSetText,
STUDIO_SET_STYLE_DESCRIPTION,
STUDIO_SET_STYLE_INPUT_SCHEMA,
STUDIO_SET_TEXT_DESCRIPTION,
STUDIO_SET_TEXT_INPUT_SCHEMA,
type ContentToolDeps,
type StudioSetStyleResult,
type StudioSetTextResult,
} from "./tools/contentTools";
import {
studioTransform,
STUDIO_TRANSFORM_DESCRIPTION,
STUDIO_TRANSFORM_INPUT_SCHEMA,
type StudioTransformInput,
type StudioTransformResult,
type TransformToolDeps,
} from "./tools/transformTools";
const log = makeStudioDebugLogger("webmcp");
@@ -54,7 +73,8 @@ function reportRegistration(report: ToolRegistrationReport, native: boolean): vo
}
}
export interface StudioAgentToolsDeps extends SelectionToolDeps, FrameToolDeps, InspectToolDeps {
export interface StudioAgentToolsDeps
extends SelectionToolDeps, FrameToolDeps, InspectToolDeps, ContentToolDeps, TransformToolDeps {
/** Read Studio's current state. Called per tool invocation, never cached. */
getSnapshot: () => StudioLookSnapshot;
}
@@ -126,6 +146,35 @@ function buildStudioTools(depsRef: { readonly current: StudioAgentToolsDeps }):
studioInspect(depsRef.current, input as StudioInspectInput),
),
},
{
name: "studio_set_text",
title: "Set an element's text",
description: STUDIO_SET_TEXT_DESCRIPTION,
inputSchema: STUDIO_SET_TEXT_INPUT_SCHEMA,
annotations: { readOnlyHint: false, untrustedContentHint: true },
execute: (input): Promise<ToolResult<StudioSetTextResult>> =>
runToolBody("studio_set_text", () => studioSetText(depsRef.current, input)),
},
{
name: "studio_set_style",
title: "Set an element's styles",
description: STUDIO_SET_STYLE_DESCRIPTION,
inputSchema: STUDIO_SET_STYLE_INPUT_SCHEMA,
annotations: { readOnlyHint: false },
execute: (input): Promise<ToolResult<StudioSetStyleResult>> =>
runToolBody("studio_set_style", () => studioSetStyle(depsRef.current, input)),
},
{
name: "studio_transform",
title: "Move, resize or rotate",
description: STUDIO_TRANSFORM_DESCRIPTION,
inputSchema: STUDIO_TRANSFORM_INPUT_SCHEMA,
annotations: { readOnlyHint: false },
execute: (input): Promise<ToolResult<StudioTransformResult>> =>
runToolBody("studio_transform", () =>
studioTransform(depsRef.current, input as StudioTransformInput),
),
},
];
}