Files
feuerwerk/tests/editor/agent-edit.test.ts
T
adminandClaude Opus 5.5 1943539a7a
feuerwerk / gate (push) Successful in 5m7s
feuerwerk / suite (push) Successful in 18m44s
Let the agent play the game and read its state through MCP (#429)
An agent could build a game object but not see whether it works. It
had to ask the user to press the keys. Now it can play the game in the
editor itself, send keys and mouse input and read what happened.

- play_input sends steps into the running game: hold or tap a key,
  keep it down or let it go, click, press, release or move the mouse,
  turn the wheel, wait. Holding and waiting count game seconds.
- A mouse target is a point in world units or an object by id or
  name. A click on an object reports whether it landed on it.
- The answer names the game time used, the keys still held, the mouse
  and the new console lines.
- One tool instead of press_keys and pointer from the plan. A step
  list can then mix both, like holding Shift during a click.
- get_runtime reads the running game: objects with id, type and
  position, the own variables of their behaviors, the Variables
  assets, the camera, keys, mouse and the console. Named objects come
  with all their properties.
- A hidden editor tab gets no frames. play_input says so at once. If
  game time stands still for 25 s it stops, names the step and lets
  go of every key it held.
- get_view and play say when the scene has no active camera. The
  first agent test missed that and only saw it on a screenshot.
- The relay instructions tell the agent to test what it built.

Found in the live test on a real project:

- A click on an object by its name reported hit false. The hit test
  looked the target up by id only.
- get_runtime listed the built-in switches of a behavior (enabled,
  runInEditor) as its variables.

Tests:
- agent-play (new), agent-edit, agent-view, relay green.
- 5 counter-checks red: hold counts one frame, key never released,
  click ignores the object, hidden tab not refused, no variable
  filter.
- e2e 71: 21/21. The agent holds ArrowLeft 0.5 s (x -120 to -270)
  and clicks the object (hit true).
- Gate: CHECK EXIT 0, 319 files, 4944 tests.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-18 15:01:32 +02:00

522 lines
22 KiB
TypeScript
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
import { describe, expect, it } from 'vitest';
import { type Command, CommandError } from '../../src/core/commands/Command';
import { findGraphAsset } from '../../src/core/model/Project';
import { argProblem } from '../../src/editor/agent/argCheck';
import { runTool, type ToolHost, type ToolResult } from '../../src/editor/agent/agentTools';
import type { AgentView } from '../../src/editor/agent/agentView';
import { COMMANDS, type StepTarget } from '../../src/editor/agent/commandCatalog';
import { NODE_STEP } from '../../src/editor/agent/editPlan';
import { COLUMN_GAP, type GraphLint } from '../../src/editor/graph/arrange';
import { NODE_WIDTH } from '../../src/editor/graph/nodeBox';
import TOOLS from '../../src/editor/agent/tools.json';
import { makeBehavior, makeSpriteScene, type SceneFixture } from '../helpers';
/** What an agent changes and reads through the tools (#429, stage 2). */
function hostFor(fixture: SceneFixture, deleteAnswer: string | null = null, follows = false) {
const asked: string[][] = [];
const shown: StepTarget[] = [];
const host: ToolHost = {
view: () => ({ play: 'editing' }) as AgentView,
screenshot: () => Promise.resolve({ data: '', width: 0, height: 0 }),
project: () => fixture.project,
execute: (command: Command) => {
try {
fixture.stack.execute(command);
return null;
} catch (error) {
if (error instanceof CommandError) return error.message;
throw error;
}
},
confirmDeletes: (ids) => {
asked.push([...ids]);
return Promise.resolve(deleteAnswer);
},
open: () => null,
select: () => null,
play: () => null,
undo: () => {
const label = fixture.stack.undoLabel;
if (label !== null) fixture.stack.undo();
return label;
},
redo: () => {
const label = fixture.stack.redoLabel;
if (label !== null) fixture.stack.redo();
return label;
},
follows: () => follows,
show: (focus) => shown.push(focus),
};
return { host, asked, shown };
}
const text = (result: ToolResult) =>
result.content.map((item) => (item.type === 'text' ? item.text : '')).join('\n');
/** An object that walks left while the left arrow is held: behavior, placement, three nodes, three edges. */
function walkLeft(fixture: SceneFixture) {
return [
{ command: 'create_asset', args: { type: 'behavior', name: 'Walk' }, as: 'walk' },
{
command: 'place_asset',
args: { in: fixture.sceneId, asset: fixture.spriteId, x: 100, y: 50 },
as: 'player',
},
{
command: 'place_asset',
args: { in: fixture.sceneId, asset: '$walk', x: 100, y: 50, parent: '$player' },
},
{
command: 'add_node',
args: { asset: '$walk', node: 'OnKeyHeld', params: { key: 'ArrowLeft' } },
as: 'key',
},
{ command: 'add_node', args: { asset: '$walk', node: 'Move', params: { dx: -200 } }, as: 'move' },
{ command: 'add_node', args: { asset: '$walk', node: 'Owner' }, as: 'owner' },
{
command: 'connect',
args: { asset: '$walk', from: { node: '$key', port: 'exec' }, to: { node: '$move', port: 'exec' } },
},
{
command: 'connect',
args: { asset: '$walk', from: { node: '$key', port: 'dt' }, to: { node: '$move', port: 'dt' } },
},
{
command: 'connect',
args: {
asset: '$walk',
from: { node: '$owner', port: 'object' },
to: { node: '$move', port: 'object' },
},
},
];
}
describe('run_commands', () => {
it('builds an object with a behavior in one call, as one undo step that takes all of it back', async () => {
const fixture = makeSpriteScene('Hero');
const { host } = hostFor(fixture);
const before = JSON.stringify(fixture.project);
const result = await runTool('run_commands', { steps: walkLeft(fixture), label: 'walk left' }, host);
expect(result.isError).toBeFalsy();
expect(fixture.stack.undoLabel).toBe('Agent: walk left');
const behavior = fixture.project.assets.find((asset) => asset.name === 'Walk')!;
const graph = findGraphAsset(fixture.project, behavior.id)!.graph;
expect(graph.nodes.map((node) => node.type)).toEqual(['OnKeyHeld', 'Move', 'Owner']);
expect(graph.nodes[0]!.params.key).toBe('ArrowLeft');
expect(graph.nodes[1]!.params.dx).toBe(-200);
expect(graph.edges.map((edge) => `${edge.from.port}>${edge.to.port}`)).toEqual([
'exec>exec',
'dt>dt',
'object>object',
]);
const scene = fixture.sceneOf();
const [player, walk] = scene.objects;
expect(walk).toMatchObject({ assetId: behavior.id, parentId: player!.id });
expect(text(result)).toContain(`Step 1 made {"id":"${behavior.id}"}`);
const after = JSON.stringify(fixture.project);
expect(text(await runTool('undo', {}, host))).toBe('Undone: Agent: walk left.');
expect(JSON.stringify(fixture.project)).toBe(before);
await runTool('redo', {}, host);
expect(JSON.stringify(fixture.project)).toBe(after);
});
it('arranges the nodes it gave no place right of those the graph has, which stay put', async () => {
const fixture = makeSpriteScene();
const behavior = makeBehavior(fixture, 'Jump');
const { host } = hostFor(fixture);
await runTool(
'run_commands',
{ steps: [{ command: 'add_node', args: { asset: behavior.id, node: 'OnStart', x: 40, y: 80 } }] },
host,
);
await runTool(
'run_commands',
{
steps: [
{ command: 'add_node', args: { asset: behavior.id, node: 'Owner' }, as: 'owner' },
{ command: 'add_node', args: { asset: behavior.id, node: 'Move' }, as: 'move' },
{
command: 'connect',
args: {
asset: behavior.id,
from: { node: '$owner', port: 'object' },
to: { node: '$move', port: 'object' },
},
},
],
},
host,
);
const [start, owner, move] = behavior.graph.nodes;
expect([start!.x, start!.y]).toEqual([40, 80]);
// The new part keeps the place it waited at, right of the start, and reads left to right.
expect(owner!.x).toBe(40 + NODE_STEP);
expect(move!.x).toBe(owner!.x + NODE_WIDTH + COLUMN_GAP);
expect(fixture.stack.undoLabel).toBe('Agent: add_node, add_node, connect');
const lint = JSON.parse(text(await runTool('lint_graph', { asset: behavior.id }, host))) as GraphLint;
expect(lint).toMatchObject({
nodes: 3,
edges: 1,
overlaps: 0,
crossings: 0,
throughNodes: 0,
backwards: 0,
});
});
it('leaves a node given a place where it was put, and arranges the rest around it', async () => {
const fixture = makeSpriteScene();
const behavior = makeBehavior(fixture, 'Jump');
const { host } = hostFor(fixture);
await runTool(
'run_commands',
{
steps: [
{ command: 'add_node', args: { asset: behavior.id, node: 'OnStart', x: 0, y: 0 }, as: 'start' },
{ command: 'add_node', args: { asset: behavior.id, node: 'Move' }, as: 'move' },
{
command: 'connect',
args: {
asset: behavior.id,
from: { node: '$start', port: 'exec' },
to: { node: '$move', port: 'exec' },
},
},
],
},
host,
);
const [start, move] = behavior.graph.nodes;
expect([start!.x, start!.y]).toEqual([0, 0]);
expect(move!.x).toBeGreaterThanOrEqual(NODE_WIDTH);
});
it('arrange_graph arranges what it is given as one step of the edit, and undo puts it back', async () => {
const fixture = makeSpriteScene();
const behavior = makeBehavior(fixture, 'Jump');
const { host } = hostFor(fixture);
const steps = [
{ command: 'add_node', args: { asset: behavior.id, node: 'OnStart', x: 500, y: 0 }, as: 'start' },
{ command: 'add_node', args: { asset: behavior.id, node: 'Move', x: 0, y: 0 }, as: 'move' },
{
command: 'connect',
args: {
asset: behavior.id,
from: { node: '$start', port: 'exec' },
to: { node: '$move', port: 'exec' },
},
},
];
await runTool('run_commands', { steps }, host);
const tangled = JSON.stringify(behavior.graph.nodes);
const before = JSON.parse(text(await runTool('lint_graph', { asset: behavior.id }, host))) as GraphLint;
expect(before.backwards).toBe(1);
const result = await runTool(
'run_commands',
{ steps: [{ command: 'arrange_graph', args: { asset: behavior.id } }], label: 'tidy' },
host,
);
expect(result.isError).toBeFalsy();
const [start, move] = behavior.graph.nodes;
expect(move!.x).toBe(start!.x + NODE_WIDTH + COLUMN_GAP);
const after = JSON.parse(text(await runTool('lint_graph', { asset: behavior.id }, host))) as GraphLint;
expect(after.backwards).toBe(0);
await runTool('undo', {}, host);
expect(JSON.stringify(behavior.graph.nodes)).toBe(tangled);
const unknown = await runTool(
'run_commands',
{ steps: [{ command: 'arrange_graph', args: { asset: behavior.id, nodes: ['nope'] } }] },
host,
);
expect(text(unknown)).toContain('Step 1 (arrange_graph): Jump has no node nope');
});
it('changes nothing when a step fails, and names the step and why', async () => {
const fixture = makeSpriteScene();
const { host } = hostFor(fixture);
const before = JSON.stringify(fixture.project);
const steps = walkLeft(fixture);
steps[7] = {
command: 'connect',
args: { asset: '$walk', from: { node: '$key', port: 'dt' }, to: { node: '$move', port: 'exec' } },
};
const result = await runTool('run_commands', { steps }, host);
expect(result.isError).toBe(true);
expect(text(result)).toBe('Nothing was changed. Step 8 (connect): Cannot connect number to exec');
expect(JSON.stringify(fixture.project)).toBe(before);
expect(fixture.stack.canUndo).toBe(true); // the fixture's own two assets, nothing of the agent's
expect(fixture.stack.undoLabel).not.toMatch(/^Agent/);
});
it('checks every step’s arguments against its command, and refuses an unknown command or node', async () => {
const fixture = makeSpriteScene();
const behavior = makeBehavior(fixture);
const { host } = hostFor(fixture);
const one = async (step: object) => text(await runTool('run_commands', { steps: [step] }, host));
expect(await one({ command: 'rename_asset', args: { asset: fixture.spriteId } })).toBe(
'Nothing was changed. Step 1 (rename_asset): args.name is missing',
);
expect(await one({ command: 'create_asset', args: { type: 'texture', name: 'T' } })).toMatch(
/Step 1 \(create_asset\): args\.type has to be one of scene, /,
);
expect(await one({ command: 'teleport', args: {} })).toMatch(/Step 1 \(teleport\): no such command/);
expect(await one({ command: 'add_node', args: { asset: behavior.id, node: 'Fly' } })).toMatch(
/Step 1 \(add_node\): A behavior graph has no node "Fly" – look it up with catalog/,
);
expect(
await one({ command: 'rename_asset', args: { asset: fixture.spriteId, name: 'X', colour: 1 } }),
).toMatch(/args has no colour/);
expect(text(await runTool('run_commands', { steps: [] }, host))).toBe(
'arguments.steps needs at least 1 item(s)',
);
});
it('finds a node by its label too, and keeps a typed-in input value as given', async () => {
const fixture = makeSpriteScene();
const behavior = makeBehavior(fixture);
const { host } = hostFor(fixture);
await runTool(
'run_commands',
{ steps: [{ command: 'add_node', args: { asset: behavior.id, node: 'on key held' } }] },
host,
);
expect(behavior.graph.nodes.map((node) => node.type)).toEqual(['OnKeyHeld']);
});
it('asks the user before deleting an asset that was there, and does nothing on a no', async () => {
const fixture = makeSpriteScene();
const no = hostFor(fixture, 'The user said no; nothing was changed.');
const deleting = { steps: [{ command: 'delete_asset', args: { asset: fixture.spriteId } }] };
const refused = await runTool('run_commands', deleting, no.host);
expect(no.asked).toEqual([[fixture.spriteId]]);
expect(text(refused)).toBe('The user said no; nothing was changed.');
expect(fixture.project.assets.some((asset) => asset.id === fixture.spriteId)).toBe(true);
const yes = hostFor(fixture);
await runTool('run_commands', deleting, yes.host);
expect(fixture.project.assets.some((asset) => asset.id === fixture.spriteId)).toBe(false);
// What the agent made in the same call is its own to take away again.
const own = hostFor(fixture);
await runTool(
'run_commands',
{
steps: [
{ command: 'create_asset', args: { type: 'sprite', name: 'Tmp' }, as: 'tmp' },
{ command: 'delete_asset', args: { asset: '$tmp' } },
],
},
own.host,
);
expect(own.asked).toEqual([]);
});
it('refuses a name for a step that makes nothing, or a part a step did not make', async () => {
const fixture = makeSpriteScene();
const { host } = hostFor(fixture);
const rename = { command: 'rename_asset', args: { asset: fixture.spriteId, name: 'A' }, as: 'r' };
expect(text(await runTool('run_commands', { steps: [rename] }, host))).toMatch(
/Step 1 \(rename_asset\): makes nothing a later step could call "r"/,
);
const steps = [
{ command: 'create_asset', args: { type: 'sprite', name: 'S' }, as: 's' },
{ command: 'rename_asset', args: { asset: '$s.closing', name: 'T' } },
];
expect(text(await runTool('run_commands', { steps }, host))).toMatch(
/Step 2 \(rename_asset\): "\$s.closing": that step made no closing/,
);
});
});
describe('the editor following the agent', () => {
it('shows the graph a call built, with every node it touched selected and the removed ones left out', async () => {
const fixture = makeSpriteScene();
const { host, shown } = hostFor(fixture, null, true);
await runTool('run_commands', { steps: walkLeft(fixture) }, host);
const walk = fixture.project.assets.find((asset) => asset.name === 'Walk')!;
const nodes = findGraphAsset(fixture.project, walk.id)!.graph.nodes.map((node) => node.id);
expect(shown).toEqual([{ kind: 'nodes', asset: walk.id, ids: nodes }]);
const steps = [
{ command: 'add_node', args: { asset: walk.id, node: 'OnStart' }, as: 'start' },
{ command: 'set_node_param', args: { asset: walk.id, node: nodes[1], key: 'dx', value: -100 } },
{ command: 'remove_node', args: { asset: walk.id, node: '$start' } },
];
await runTool('run_commands', { steps }, host);
expect(shown[1]).toEqual({ kind: 'nodes', asset: walk.id, ids: [nodes[1]] });
});
it('switches to the scene of the objects a call ends on, inside the same undo step', async () => {
const fixture = makeSpriteScene();
const { host, shown } = hostFor(fixture, null, true);
const made = await runTool(
'run_commands',
{ steps: [{ command: 'create_asset', args: { type: 'scene', name: 'Cave' } }] },
host,
);
const cave = (made.structuredContent as { made: { id: string }[] }).made[0]!.id;
expect(shown).toEqual([{ kind: 'asset', asset: cave }]);
expect(fixture.project.activeSceneId).toBe(fixture.sceneId);
const place = {
command: 'place_asset',
args: { in: cave, asset: fixture.spriteId, x: 0, y: 0 },
as: 'bat',
};
const placed = await runTool('run_commands', { steps: [place], label: 'bat' }, host);
const bat = (placed.structuredContent as { made: { id: string }[] }).made[0]!.id;
expect(shown[1]).toEqual({ kind: 'objects', in: cave, ids: [bat] });
expect(fixture.project.activeSceneId).toBe(cave);
expect(text(await runTool('undo', {}, host))).toBe('Undone: Agent: bat.');
expect(fixture.project.activeSceneId).toBe(fixture.sceneId);
});
it('leaves the view and the scene alone when the user switched following off', async () => {
const fixture = makeSpriteScene();
const { host, shown } = hostFor(fixture);
const steps = [
{ command: 'create_asset', args: { type: 'scene', name: 'Cave' }, as: 'cave' },
{ command: 'place_asset', args: { in: '$cave', asset: fixture.spriteId, x: 0, y: 0 } },
];
expect((await runTool('run_commands', { steps }, host)).isError).toBeFalsy();
expect(shown).toEqual([]);
expect(fixture.project.activeSceneId).toBe(fixture.sceneId);
});
});
describe('reading', () => {
it('get_graph lists each node’s ports as they are, and the edges', async () => {
const fixture = makeSpriteScene();
const { host } = hostFor(fixture);
await runTool('run_commands', { steps: walkLeft(fixture) }, host);
const walk = fixture.project.assets.find((asset) => asset.name === 'Walk')!;
const graph = (await runTool('get_graph', { asset: walk.id }, host)).structuredContent as {
nodes: { type: string; inputs: { key: string; type: string }[] }[];
edges: unknown[];
};
expect(graph.edges).toHaveLength(3);
expect(graph.nodes.find((node) => node.type === 'Move')!.inputs.map((port) => port.key)).toEqual([
'exec',
'object',
'dx',
'dy',
'scale',
'dt',
]);
});
it('get_asset counts a graph instead of spelling it out, and names a scene’s objects', async () => {
const fixture = makeSpriteScene('Hero');
const { host } = hostFor(fixture);
await runTool('run_commands', { steps: walkLeft(fixture) }, host);
const walk = fixture.project.assets.find((asset) => asset.name === 'Walk')!;
const asset = (await runTool('get_asset', { asset: walk.id }, host)).structuredContent as Record<
string,
unknown
>;
expect(asset.graph).toEqual({ nodes: 3, edges: 3, read: 'get_graph' });
const scene = (await runTool('get_asset', { asset: fixture.sceneId }, host)).structuredContent as {
objects: { name: string }[];
};
expect(scene.objects.map((object) => object.name)).toEqual(['Hero', 'Walk']);
expect(text(await runTool('get_asset', { asset: 'nope' }, host))).toMatch(/There is no asset nope/);
});
it('list_assets filters by type and name', async () => {
const fixture = makeSpriteScene('Hero');
makeBehavior(fixture, 'Jump');
const { host } = hostFor(fixture);
const listed = (query: Record<string, unknown>) =>
runTool('list_assets', query, host).then(
(r) => (r.structuredContent as { assets: { name: string }[] }).assets,
);
expect((await listed({})).map((asset) => asset.name)).toEqual(['Level', 'Hero', 'Jump']);
expect((await listed({ type: 'behavior' })).map((asset) => asset.name)).toEqual(['Jump']);
expect((await listed({ query: 'he' })).map((asset) => asset.name)).toEqual(['Hero']);
});
it('the node catalog of a graph answers a search in full, and lists every node without one', async () => {
const fixture = makeSpriteScene();
const behavior = makeBehavior(fixture);
const { host } = hostFor(fixture);
const nodes = async (query?: string) =>
(
(await runTool('catalog', { topic: 'nodes', asset: behavior.id, ...(query ? { query } : {}) }, host))
.structuredContent as { nodes: { id: string; inputs?: unknown[]; settings?: { key: string }[] }[] }
).nodes;
const found = await nodes('key held');
expect(found[0]).toMatchObject({ id: 'OnKeyHeld', inputs: [] });
expect(found[0]!.settings!.map((setting) => setting.key)).toContain('key');
const all = await nodes();
expect(all.some((entry) => entry.id === 'Move')).toBe(true);
expect(all[0]!.inputs).toBeUndefined();
expect(text(await runTool('catalog', { topic: 'nodes' }, host))).toMatch(/Name the graph asset/);
});
});
describe('the catalog', () => {
it('has every command once, each with arguments the check understands', async () => {
const names = COMMANDS.map((command) => command.name);
expect(new Set(names).size).toBe(names.length);
for (const command of COMMANDS) {
expect(command.args).toMatchObject({ type: 'object', additionalProperties: false });
expect(argProblem(command.args, {})).toMatch(/is missing|^null$/);
}
const fixture = makeSpriteScene();
const listed = (await runTool('catalog', { topic: 'commands' }, hostFor(fixture).host))
.structuredContent as {
commands: { name: string }[];
};
expect(listed.commands.map((command) => command.name)).toEqual(names);
});
it('the tools are the ones the relay lists, each with a schema', () => {
expect(TOOLS.map((tool) => tool.name)).toEqual([
'get_view',
'screenshot',
'list_assets',
'get_asset',
'get_graph',
'lint_graph',
'catalog',
'run_commands',
'open',
'select',
'play',
'play_input',
'get_runtime',
'undo',
'redo',
]);
for (const tool of TOOLS)
expect(tool.inputSchema).toMatchObject({ type: 'object', additionalProperties: false });
});
});
describe('argProblem', () => {
it('names the place of the first problem', () => {
const schema = {
type: 'object' as const,
properties: {
at: { type: 'array' as const, items: { type: 'integer' as const, minimum: 0 } },
kind: { type: 'string' as const, enum: ['a', 'b'] },
},
required: ['kind'],
additionalProperties: false as const,
};
expect(argProblem(schema, { kind: 'a', at: [1, 2] })).toBeNull();
expect(argProblem(schema, { kind: 'c' })).toBe('arguments.kind has to be one of a, b');
expect(argProblem(schema, { kind: 'a', at: [1, -1] })).toBe('arguments.at[1] has to be 0 or more');
expect(argProblem(schema, { kind: 'a', at: [1.5] })).toBe('arguments.at[0] has to be a whole number');
expect(argProblem(schema, [])).toBe('arguments has to be an object');
});
});