Merge pull request #15 from pingdotgg/images

Julius Marminge committed 9 months ago

Commit 1dca7012 parents 44f3b3c ea16f6f14 files · +770 −54Browse files

Unified · Split

14 files

apps/server/src/codexAppServerManager.test.ts +117 −1
apps/server/src/codexAppServerManager.ts +28 −8
apps/web/src/components/ChatView.tsx +361 −24
apps/web/src/components/DiffPanel.tsx +1 −1
apps/web/src/components/Sidebar.tsx +3 −3
apps/web/src/historyBootstrap.test.ts +28 −0
apps/web/src/historyBootstrap.ts +29 −3
apps/web/src/persistenceSchema.test.ts +28 −1
apps/web/src/persistenceSchema.ts +42 −5
apps/web/src/store.ts +11 −1
apps/web/src/types.ts +12 −0
packages/contracts/src/model.ts +8 −1
packages/contracts/src/provider.test.ts +59 −0
packages/contracts/src/provider.ts +43 −6

apps/server/src/codexAppServerManager.test.ts+117−1

@@ -1,11 +1,42 @@
1		−import { describe, expect, it } from "vitest";
	1	+import { describe, expect, it, vi } from "vitest";
2	2
3	3	import {
	4	+ CodexAppServerManager,
4	5	classifyCodexStderrLine,
5	6	isRecoverableThreadResumeError,
6	7	normalizeCodexModelSlug,
7	8	} from "./codexAppServerManager";
8	9
	10	+function createSendTurnHarness() {
	11	+ const manager = new CodexAppServerManager();
	12	+ const context = {
	13	+ session: {
	14	+ sessionId: "sess_1",
	15	+ provider: "codex",
	16	+ status: "ready",
	17	+ threadId: "thread_1",
	18	+ createdAt: "2026-02-10T00:00:00.000Z",
	19	+ updatedAt: "2026-02-10T00:00:00.000Z",
	20	+ },
	21	+ };
	22	+
	23	+ const requireSession = vi
	24	+ .spyOn(manager as unknown as { requireSession: (sessionId: string) => unknown }, "requireSession")
	25	+ .mockReturnValue(context);
	26	+ const sendRequest = vi
	27	+ .spyOn(manager as unknown as { sendRequest: (...args: unknown[]) => Promise<unknown> }, "sendRequest")
	28	+ .mockResolvedValue({
	29	+ turn: {
	30	+ id: "turn_1",
	31	+ },
	32	+ });
	33	+ const updateSession = vi
	34	+ .spyOn(manager as unknown as { updateSession: (...args: unknown[]) => void }, "updateSession")
	35	+ .mockImplementation(() => {});
	36	+
	37	+ return { manager, context, requireSession, sendRequest, updateSession };
	38	+}
	39	+
9	40	describe("classifyCodexStderrLine", () => {
10	41	it("ignores empty lines", () => {
11	42	expect(classifyCodexStderrLine(" ")).toBeNull();
@@ -75,3 +106,88 @@ describe("isRecoverableThreadResumeError", () => { · 63 unmodified lines
75	106	).toBe(false);
76	107	});
77	108	});
	109	+
	110	+describe("sendTurn", () => {
	111	+ it("sends text and image user input items to turn/start", async () => {
	112	+ const { manager, context, requireSession, sendRequest, updateSession } = createSendTurnHarness();
	113	+
	114	+ const result = await manager.sendTurn({
	115	+ sessionId: "sess_1",
	116	+ input: "Inspect this image",
	117	+ attachments: [
	118	+ {
	119	+ type: "image",
	120	+ name: "error.png",
	121	+ mimeType: "image/png",
	122	+ sizeBytes: 1_024,
	123	+ dataUrl: "data:image/png;base64,AAAA",
	124	+ },
	125	+ ],
	126	+ model: "gpt-5.3",
	127	+ effort: "high",
	128	+ });
	129	+
	130	+ expect(result).toEqual({
	131	+ threadId: "thread_1",
	132	+ turnId: "turn_1",
	133	+ });
	134	+ expect(requireSession).toHaveBeenCalledWith("sess_1");
	135	+ expect(sendRequest).toHaveBeenCalledWith(context, "turn/start", {
	136	+ threadId: "thread_1",
	137	+ input: [
	138	+ {
	139	+ type: "text",
	140	+ text: "Inspect this image",
	141	+ text_elements: [],
	142	+ },
	143	+ {
	144	+ type: "image",
	145	+ url: "data:image/png;base64,AAAA",
	146	+ },
	147	+ ],
	148	+ model: "gpt-5.3-codex",
	149	+ effort: "high",
	150	+ });
	151	+ expect(updateSession).toHaveBeenCalledWith(context, {
	152	+ status: "running",
	153	+ activeTurnId: "turn_1",
	154	+ });
	155	+ });
	156	+
	157	+ it("supports image-only turns", async () => {
	158	+ const { manager, context, sendRequest } = createSendTurnHarness();
	159	+
	160	+ await manager.sendTurn({
	161	+ sessionId: "sess_1",
	162	+ attachments: [
	163	+ {
	164	+ type: "image",
	165	+ name: "diagram.png",
	166	+ mimeType: "image/png",
	167	+ sizeBytes: 256,
	168	+ dataUrl: "data:image/png;base64,BBBB",
	169	+ },
	170	+ ],
	171	+ });
	172	+
	173	+ expect(sendRequest).toHaveBeenCalledWith(context, "turn/start", {
	174	+ threadId: "thread_1",
	175	+ input: [
	176	+ {
	177	+ type: "image",
	178	+ url: "data:image/png;base64,BBBB",
	179	+ },
	180	+ ],
	181	+ });
	182	+ });
	183	+
	184	+ it("rejects empty turn input", async () => {
	185	+ const { manager } = createSendTurnHarness();
	186	+
	187	+ await expect(
	188	+ manager.sendTurn({
	189	+ sessionId: "sess_1",
	190	+ }),
	191	+ ).rejects.toThrow("Turn input must include text or attachments.");
	192	+ });
	193	+});

apps/server/src/codexAppServerManager.ts+28−8

@@ -257,20 +257,40 @@ export class CodexAppServerManager extends EventEmitter<CodexAppServerManagerEve · 256 unmodified lines
257	257	throw new Error("Session is missing a thread id.");
258	258	}
259	259
	260	+ const turnInput: Array<
	261	+ \| { type: "text"; text: string; text_elements: [] }
	262	+ \| { type: "image"; url: string }
	263	+ > = [];
	264	+ if (input.input) {
	265	+ turnInput.push({
	266	+ type: "text",
	267	+ text: input.input,
	268	+ text_elements: [],
	269	+ });
	270	+ }
	271	+ for (const attachment of input.attachments ?? []) {
	272	+ if (attachment.type === "image") {
	273	+ turnInput.push({
	274	+ type: "image",
	275	+ url: attachment.dataUrl,
	276	+ });
	277	+ }
	278	+ }
	279	+ if (turnInput.length === 0) {
	280	+ throw new Error("Turn input must include text or attachments.");
	281	+ }
	282	+
260	283	const turnStartParams: {
261	284	threadId: string;
262		− input: Array<{ type: "text"; text: string; text_elements: [] }>;
	285	+ input: Array<
	286	+ \| { type: "text"; text: string; text_elements: [] }
	287	+ \| { type: "image"; url: string }
	288	+ >;
263	289	model?: string;
264	290	effort?: string;
265	291	} = {
266	292	threadId: context.session.threadId,
267		− input: [
268		− {
269		− type: "text",
270		− text: input.input,
271		− text_elements: [],
272		− },
273		− ],
	293	+ input: turnInput,
274	294	};
275	295	const normalizedModel = normalizeCodexModelSlug(input.model);
276	296	if (normalizedModel) {

apps/web/src/components/ChatView.tsx+361−24

@@ -1,9 +1,14 @@
1	1	import {
2	2	PROVIDER_SEND_TURN_MAX_INPUT_CHARS,
	3	+ PROVIDER_SEND_TURN_MAX_ATTACHMENTS,
	4	+ PROVIDER_SEND_TURN_MAX_IMAGE_BYTES,
	5	+ type ProviderSendTurnAttachmentInput,
3	6	type ProviderApprovalDecision,
4	7	type ProviderEvent,
5	8	} from "@t3tools/contracts";
6	9	import {
	10	+ type ClipboardEvent,
	11	+ type DragEvent,
7	12	type FormEvent,
8	13	Fragment,
9	14	type KeyboardEvent,
@@ -33,6 +38,7 @@ import { · 23 unmodified lines
33	38	formatTimestamp,
34	39	} from "../session-logic";
35	40	import { useStore } from "../store";
	41	+import type { ChatImageAttachment } from "../types";
36	42	import BranchToolbar from "./BranchToolbar";
37	43	import GitActionsControl from "./GitActionsControl";
38	44	import { isTerminalToggleShortcut } from "../terminal-shortcuts";
@@ -57,6 +63,9 @@ function editorLabel(editor: (typeof EDITORS)[number]): string { · 18 unmodified lines
57	63
58	64	const LAST_EDITOR_KEY = "t3code:last-editor";
59	65	const MAX_VISIBLE_WORK_LOG_ENTRIES = 6;
	66	+const IMAGE_SIZE_LIMIT_LABEL = `${Math.round(PROVIDER_SEND_TURN_MAX_IMAGE_BYTES / (1024 * 1024))}MB`;
	67	+const IMAGE_ONLY_BOOTSTRAP_PROMPT =
	68	+ "[User attached one or more images without additional text. Respond using the conversation context and the attached image(s).]";
60	69
61	70	function workToneClass(tone: "thinking" \| "tool" \| "info" \| "error"): string {
62	71	if (tone === "error") return "text-rose-300/50 dark:text-rose-300/50";
@@ -80,6 +89,16 @@ interface EnsuredSessionInfo { · 17 unmodified lines
80	89	continuityState: SessionContinuityState;
81	90	}
82	91
	92	+interface ComposerImageAttachment extends Omit<ChatImageAttachment, "previewUrl"> {
	93	+ previewUrl: string;
	94	+ file: File;
	95	+}
	96	+
	97	+interface ExpandedImagePreview {
	98	+ src: string;
	99	+ name: string;
	100	+}
	101	+
83	102	function asRecord(value: unknown): Record<string, unknown> \| undefined {
84	103	if (!value \|\| typeof value !== "object") return undefined;
85	104	return value as Record<string, unknown>;
@@ -131,10 +150,30 @@ function derivePendingApprovals(events: ProviderEvent[]): PendingApprovalCard[] · 45 unmodified lines
131	150	return Array.from(pending.values());
132	151	}
133	152
	153	+function readFileAsDataUrl(file: File): Promise<string> {
	154	+ return new Promise((resolve, reject) => {
	155	+ const reader = new FileReader();
	156	+ reader.addEventListener("load", () => {
	157	+ if (typeof reader.result === "string") {
	158	+ resolve(reader.result);
	159	+ return;
	160	+ }
	161	+ reject(new Error("Could not read image data."));
	162	+ });
	163	+ reader.addEventListener("error", () => {
	164	+ reject(reader.error ?? new Error("Failed to read image."));
	165	+ });
	166	+ reader.readAsDataURL(file);
	167	+ });
	168	+}
	169	+
134	170	export default function ChatView() {
135	171	const { state, dispatch } = useStore();
136	172	const api = useNativeApi();
137	173	const [prompt, setPrompt] = useState("");
	174	+ const [composerImages, setComposerImages] = useState<ComposerImageAttachment[]>([]);
	175	+ const [isDragOverComposer, setIsDragOverComposer] = useState(false);
	176	+ const [expandedImage, setExpandedImage] = useState<ExpandedImagePreview \| null>(null);
138	177	const [isSending, setIsSending] = useState(false);
139	178	const [isConnecting, setIsConnecting] = useState(false);
140	179	const [isModelMenuOpen, setIsModelMenuOpen] = useState(false);
@@ -152,6 +191,8 @@ export default function ChatView() { · 11 unmodified lines
152	191	const [terminalFocusRequestId, setTerminalFocusRequestId] = useState(0);
153	192	const messagesEndRef = useRef<HTMLDivElement>(null);
154	193	const textareaRef = useRef<HTMLTextAreaElement>(null);
	194	+ const composerImagesRef = useRef<ComposerImageAttachment[]>([]);
	195	+ const dragDepthRef = useRef(0);
155	196	const modelMenuRef = useRef<HTMLDivElement>(null);
156	197	const editorMenuRef = useRef<HTMLDivElement>(null);
157	198	const terminalOpenByThreadRef = useRef<Record<string, boolean>>({});
@@ -264,7 +305,15 @@ export default function ChatView() { · 106 unmodified lines
264	305	(activeThread.messages.length > 0 \|\|
265	306	(activeThread.session !== null && activeThread.session.status !== "closed")),
266	307	);
267		− const terminalShortcutHint = navigator.platform.includes("Mac") ? "\u2318J" : "Ctrl+J";
	308	+ const terminalShortcutHint = navigator.platform.includes("Mac")
	309	+ ? "\u2318J"
	310	+ : "Ctrl+J";
	311	+ const revokePreviewUrls = useCallback((images: Array<{ previewUrl?: string }>) => {
	312	+ for (const image of images) {
	313	+ if (!image.previewUrl) continue;
	314	+ URL.revokeObjectURL(image.previewUrl);
	315	+ }
	316	+ }, []);
268	317	const focusComposer = useCallback(() => {
269	318	const textarea = textareaRef.current;
270	319	if (!textarea) return;
@@ -319,6 +368,44 @@ export default function ChatView() { · 48 unmodified lines
319	368	setExpandedWorkGroups({});
320	369	}, [activeThread?.id]);
321	370
	371	+ useEffect(() => {
	372	+ composerImagesRef.current = composerImages;
	373	+ }, [composerImages]);
	374	+
	375	+ useEffect(() => {
	376	+ setComposerImages((existing) => {
	377	+ revokePreviewUrls(existing);
	378	+ return [];
	379	+ });
	380	+ dragDepthRef.current = 0;
	381	+ setIsDragOverComposer(false);
	382	+ setExpandedImage(null);
	383	+ }, [activeThread?.id, revokePreviewUrls]);
	384	+
	385	+ useEffect(() => {
	386	+ return () => {
	387	+ revokePreviewUrls(composerImagesRef.current);
	388	+ };
	389	+ }, [revokePreviewUrls]);
	390	+
	391	+ useEffect(() => {
	392	+ if (!expandedImage) {
	393	+ return;
	394	+ }
	395	+
	396	+ const onKeyDown = (event: globalThis.KeyboardEvent) => {
	397	+ if (event.key !== "Escape") {
	398	+ return;
	399	+ }
	400	+ setExpandedImage(null);
	401	+ };
	402	+
	403	+ window.addEventListener("keydown", onKeyDown);
	404	+ return () => {
	405	+ window.removeEventListener("keydown", onKeyDown);
	406	+ };
	407	+ }, [expandedImage]);
	408	+
322	409	const activeWorktreePath = activeThread?.worktreePath;
323	410
324	411	useEffect(() => {
@@ -431,6 +518,125 @@ export default function ChatView() { · 106 unmodified lines
431	518	setIsEditorMenuOpen(false);
432	519	};
433	520
	521	+ const setThreadError = (threadId: string \| null, error: string \| null) => {
	522	+ if (!threadId) return;
	523	+ dispatch({
	524	+ type: "SET_ERROR",
	525	+ threadId,
	526	+ error,
	527	+ });
	528	+ };
	529	+
	530	+ const addComposerImages = (files: File[]) => {
	531	+ if (!activeThreadId \|\| files.length === 0) return;
	532	+
	533	+ const nextImages = [...composerImagesRef.current];
	534	+ let error: string \| null = null;
	535	+ for (const file of files) {
	536	+ if (!file.type.startsWith("image/")) {
	537	+ error = `Unsupported file type for '${file.name}'. Please attach image files only.`;
	538	+ continue;
	539	+ }
	540	+ if (file.size > PROVIDER_SEND_TURN_MAX_IMAGE_BYTES) {
	541	+ error = `'${file.name}' exceeds the ${IMAGE_SIZE_LIMIT_LABEL} attachment limit.`;
	542	+ continue;
	543	+ }
	544	+ if (nextImages.length >= PROVIDER_SEND_TURN_MAX_ATTACHMENTS) {
	545	+ error = `You can attach up to ${PROVIDER_SEND_TURN_MAX_ATTACHMENTS} images per message.`;
	546	+ break;
	547	+ }
	548	+
	549	+ const previewUrl = URL.createObjectURL(file);
	550	+ nextImages.push({
	551	+ type: "image",
	552	+ id: crypto.randomUUID(),
	553	+ name: file.name \|\| "image",
	554	+ mimeType: file.type,
	555	+ sizeBytes: file.size,
	556	+ previewUrl,
	557	+ file,
	558	+ });
	559	+ }
	560	+
	561	+ setComposerImages(nextImages);
	562	+ setThreadError(activeThreadId, error);
	563	+ };
	564	+
	565	+ const removeComposerImage = (imageId: string) => {
	566	+ setComposerImages((existing) => {
	567	+ const match = existing.find((image) => image.id === imageId);
	568	+ if (match?.previewUrl) {
	569	+ URL.revokeObjectURL(match.previewUrl);
	570	+ }
	571	+ return existing.filter((image) => image.id !== imageId);
	572	+ });
	573	+ };
	574	+
	575	+ const openImageDialog = (src: string \| undefined, name: string) => {
	576	+ if (!src) {
	577	+ return;
	578	+ }
	579	+ setExpandedImage({ src, name });
	580	+ };
	581	+
	582	+ const onComposerPaste = (event: ClipboardEvent<HTMLTextAreaElement>) => {
	583	+ const files = Array.from(event.clipboardData.files);
	584	+ if (files.length === 0) {
	585	+ return;
	586	+ }
	587	+ const imageFiles = files.filter((file) => file.type.startsWith("image/"));
	588	+ if (imageFiles.length === 0) {
	589	+ return;
	590	+ }
	591	+ event.preventDefault();
	592	+ addComposerImages(imageFiles);
	593	+ };
	594	+
	595	+ const onComposerDragEnter = (event: DragEvent<HTMLDivElement>) => {
	596	+ if (!event.dataTransfer.types.includes("Files")) {
	597	+ return;
	598	+ }
	599	+ event.preventDefault();
	600	+ dragDepthRef.current += 1;
	601	+ setIsDragOverComposer(true);
	602	+ };
	603	+
	604	+ const onComposerDragOver = (event: DragEvent<HTMLDivElement>) => {
	605	+ if (!event.dataTransfer.types.includes("Files")) {
	606	+ return;
	607	+ }
	608	+ event.preventDefault();
	609	+ event.dataTransfer.dropEffect = "copy";
	610	+ setIsDragOverComposer(true);
	611	+ };
	612	+
	613	+ const onComposerDragLeave = (event: DragEvent<HTMLDivElement>) => {
	614	+ if (!event.dataTransfer.types.includes("Files")) {
	615	+ return;
	616	+ }
	617	+ event.preventDefault();
	618	+ const nextTarget = event.relatedTarget;
	619	+ if (nextTarget instanceof Node && event.currentTarget.contains(nextTarget)) {
	620	+ return;
	621	+ }
	622	+ dragDepthRef.current = Math.max(0, dragDepthRef.current - 1);
	623	+ if (dragDepthRef.current === 0) {
	624	+ setIsDragOverComposer(false);
	625	+ }
	626	+ };
	627	+
	628	+ const onComposerDrop = (event: DragEvent<HTMLDivElement>) => {
	629	+ if (!event.dataTransfer.types.includes("Files")) {
	630	+ return;
	631	+ }
	632	+ event.preventDefault();
	633	+ dragDepthRef.current = 0;
	634	+ setIsDragOverComposer(false);
	635	+ const files = Array.from(event.dataTransfer.files);
	636	+ addComposerImages(files);
	637	+ focusComposer();
	638	+ };
	639	+
434	640	const ensureSession = async (cwdOverride?: string): Promise<EnsuredSessionInfo \| null> => {
435	641	if (!api \|\| !activeThread \|\| !activeProject) return null;
436	642	if (activeThread.session && activeThread.session.status !== "closed") {
@@ -492,8 +698,9 @@ export default function ChatView() { · 55 unmodified lines
492	698	e.preventDefault();
493	699	if (!api \|\| !activeThread \|\| isSending \|\| isConnecting) return;
494	700	const trimmed = prompt.trim();
495		− if (!trimmed) return;
	701	+ if (!trimmed && composerImages.length === 0) return;
496	702	if (!activeProject) return;
	703	+ const composerImagesSnapshot = [...composerImages];
497	704
498	705	// On first message: lock in branch + create worktree if needed.
499	706	let sessionCwd: string \| undefined;
@@ -529,7 +736,12 @@ export default function ChatView() { · 29 unmodified lines
529	736
530	737	// Auto-title from first message
531	738	if (activeThread.messages.length === 0) {
532		− const title = trimmed.length > 50 ? `${trimmed.slice(0, 50)}...` : trimmed;
	739	+ const titleSeed =
	740	+ trimmed \|\|
	741	+ (composerImagesSnapshot.length > 0
	742	+ ? `Image: ${composerImagesSnapshot[0]?.name ?? "attachment"}`
	743	+ : "New thread");
	744	+ const title = titleSeed.length > 50 ? `${titleSeed.slice(0, 50)}...` : titleSeed;
533	745	dispatch({
534	746	type: "SET_THREAD_TITLE",
535	747	threadId: activeThread.id,
@@ -537,43 +749,63 @@ export default function ChatView() { · 1 unmodified line
537	749	});
538	750	}
539	751
540		− dispatch({
541		− type: "SET_ERROR",
542		− threadId: activeThread.id,
543		− error: null,
544		− });
	752	+ setThreadError(activeThread.id, null);
	753	+ const messageAttachments: ChatImageAttachment[] = composerImagesSnapshot.map((image) => ({
	754	+ type: "image",
	755	+ id: image.id,
	756	+ name: image.name,
	757	+ mimeType: image.mimeType,
	758	+ sizeBytes: image.sizeBytes,
	759	+ previewUrl: image.previewUrl,
	760	+ }));
545	761	dispatch({
546	762	type: "PUSH_USER_MESSAGE",
547	763	threadId: activeThread.id,
548	764	id: crypto.randomUUID(),
549	765	text: trimmed,
	766	+ ...(messageAttachments.length > 0 ? { attachments: messageAttachments } : {}),
550	767	});
551	768	const previousMessages = activeThread.messages;
552	769	setPrompt("");
	770	+ setComposerImages([]);
553	771
554	772	const sessionInfo = await ensureSession(sessionCwd);
555	773	if (!sessionInfo) return;
556	774
557	775	setIsSending(true);
558	776	try {
	777	+ const turnAttachments = await Promise.all(
	778	+ composerImagesSnapshot.map(async (image): Promise<ProviderSendTurnAttachmentInput> => ({
	779	+ type: "image",
	780	+ name: image.name,
	781	+ mimeType: image.mimeType,
	782	+ sizeBytes: image.sizeBytes,
	783	+ dataUrl: await readFileAsDataUrl(image.file),
	784	+ })),
	785	+ );
559	786	const shouldBootstrap =
560	787	previousMessages.length > 0 &&
561	788	(sessionInfo.continuityState === "new" \|\| sessionInfo.continuityState === "fallback_new");
	789	+ const latestPromptForBootstrap = trimmed \|\| IMAGE_ONLY_BOOTSTRAP_PROMPT;
562	790	const input = shouldBootstrap
563		− ? buildBootstrapInput(previousMessages, trimmed, PROVIDER_SEND_TURN_MAX_INPUT_CHARS).text
564		− : trimmed;
	791	+ ? buildBootstrapInput(
	792	+ previousMessages,
	793	+ latestPromptForBootstrap,
	794	+ PROVIDER_SEND_TURN_MAX_INPUT_CHARS,
	795	+ ).text
	796	+ : (trimmed \|\| undefined);
565	797	await api.providers.sendTurn({
566	798	sessionId: sessionInfo.sessionId,
567		− input,
	799	+ ...(input ? { input } : {}),
	800	+ ...(turnAttachments.length > 0 ? { attachments: turnAttachments } : {}),
568	801	model: selectedModel \|\| undefined,
569	802	effort: selectedEffort \|\| undefined,
570	803	});
571	804	} catch (err) {
572		− dispatch({
573		− type: "SET_ERROR",
574		− threadId: activeThread.id,
575		− error: err instanceof Error ? err.message : "Failed to send message.",
576		− });
	805	+ setThreadError(
	806	+ activeThread.id,
	807	+ err instanceof Error ? err.message : "Failed to send message.",
	808	+ );
577	809	} finally {
578	810	setIsSending(false);
579	811	}
@@ -631,7 +863,11 @@ export default function ChatView() { · 51 unmodified lines
631	863	if (!activeThread) {
632	864	return (
633	865	<div className="flex min-h-0 flex-1 flex-col bg-background text-muted-foreground/40">
634		− {isElectron && <div className="drag-region h-[52px] shrink-0" />}
	866	+ {isElectron && (
	867	+ <div className="drag-region flex h-[52px] shrink-0 items-center border-b border-border px-5">
	868	+ <span className="text-xs text-muted-foreground/50">No active thread</span>
	869	+ </div>
	870	+ )}
635	871	<div className="flex flex-1 items-center justify-center">
636	872	<div className="text-center">
637	873	<p className="text-sm">Select a thread or create a new one to get started.</p>
@@ -645,7 +881,7 @@ export default function ChatView() { · 7 unmodified lines
645	881	<div className="flex min-h-0 flex-1 flex-col bg-background">
646	882	{/* Top bar */}
647	883	<header
648		− className={`flex items-center justify-between border-b border-border px-5 pb-3 ${isElectron ? "drag-region pt-[28px]" : "pt-3"}`}
	884	+ className={`flex items-center justify-between border-b border-border px-5 ${isElectron ? "drag-region h-[52px]" : "py-3"}`}
649	885	>
650	886	<div className="flex items-center gap-3">
651	887	<h2 className="text-sm font-medium text-foreground">{activeThread.title}</h2>
@@ -876,13 +1112,39 @@ export default function ChatView() { · 224 unmodified lines
876	1112	}
877	1113
878	1114	if (timelineEntry.message.role === "user") {
	1115	+ const userImages = timelineEntry.message.attachments ?? [];
879	1116	return (
880	1117	<Fragment key={timelineEntry.id}>
881	1118	<div className="flex justify-end">
882	1119	<div className="max-w-[80%] rounded-2xl rounded-br-sm border border-border bg-secondary px-4 py-3">
883		− <pre className="whitespace-pre-wrap wrap-break-word font-mono text-sm leading-relaxed text-foreground">
884		− {timelineEntry.message.text}
885		− </pre>
	1120	+ {userImages.length > 0 && (
	1121	+ <div className="mb-2 grid max-w-[420px] grid-cols-2 gap-2">
	1122	+ {userImages.map((image) => (
	1123	+ <div
	1124	+ key={image.id}
	1125	+ className="overflow-hidden rounded-lg border border-border/80 bg-background/70"
	1126	+ >
	1127	+ {image.previewUrl ? (
	1128	+ <img
	1129	+ src={image.previewUrl}
	1130	+ alt={image.name}
	1131	+ className="h-full max-h-[220px] w-full cursor-zoom-in object-cover"
	1132	+ onClick={() => openImageDialog(image.previewUrl, image.name)}
	1133	+ />
	1134	+ ) : (
	1135	+ <div className="flex min-h-[72px] items-center justify-center px-2 py-3 text-center text-[11px] text-muted-foreground/70">
	1136	+ {image.name}
	1137	+ </div>
	1138	+ )}
	1139	+ </div>
	1140	+ ))}
	1141	+ </div>
	1142	+ )}
	1143	+ {timelineEntry.message.text && (
	1144	+ <pre className="whitespace-pre-wrap break-words font-mono text-sm leading-relaxed text-foreground">
	1145	+ {timelineEntry.message.text}
	1146	+ </pre>
	1147	+ )}
886	1148	<p className="mt-1.5 text-right text-[10px] text-muted-foreground/30">
887	1149	{formatTimestamp(timelineEntry.message.createdAt)}
888	1150	</p>
@@ -945,9 +1207,48 @@ export default function ChatView() { · 56 unmodified lines
945	1207	{/* Input bar */}
946	1208	<div className="px-5 pb-1 pt-2">
947	1209	<form onSubmit={onSend} className="mx-auto max-w-3xl">
948		− <div className="group rounded-[20px] border border-border bg-card transition-colors duration-200 focus-within:border-ring">
	1210	+ <div
	1211	+ className={`group rounded-[20px] border bg-card transition-colors duration-200 focus-within:border-ring ${
	1212	+ isDragOverComposer ? "border-primary/70 bg-accent/30" : "border-border"
	1213	+ }`}
	1214	+ onDragEnter={onComposerDragEnter}
	1215	+ onDragOver={onComposerDragOver}
	1216	+ onDragLeave={onComposerDragLeave}
	1217	+ onDrop={onComposerDrop}
	1218	+ >
949	1219	{/* Textarea area */}
950	1220	<div className="px-4 pt-4 pb-2">
	1221	+ {composerImages.length > 0 && (
	1222	+ <div className="mb-3 flex flex-wrap gap-2">
	1223	+ {composerImages.map((image) => (
	1224	+ <div
	1225	+ key={image.id}
	1226	+ className="relative h-16 w-16 overflow-hidden rounded-lg border border-border/80 bg-background"
	1227	+ >
	1228	+ {image.previewUrl ? (
	1229	+ <img
	1230	+ src={image.previewUrl}
	1231	+ alt={image.name}
	1232	+ className="h-full w-full cursor-zoom-in object-cover"
	1233	+ onClick={() => openImageDialog(image.previewUrl, image.name)}
	1234	+ />
	1235	+ ) : (
	1236	+ <div className="flex h-full w-full items-center justify-center px-1 text-center text-[10px] text-muted-foreground/70">
	1237	+ {image.name}
	1238	+ </div>
	1239	+ )}
	1240	+ <button
	1241	+ type="button"
	1242	+ className="absolute right-1 top-1 flex h-4 w-4 items-center justify-center rounded-full bg-background/80 text-[11px] leading-none text-foreground transition-colors duration-150 hover:bg-background"
	1243	+ onClick={() => removeComposerImage(image.id)}
	1244	+ aria-label={`Remove ${image.name}`}
	1245	+ >
	1246	+ ×
	1247	+ </button>
	1248	+ </div>
	1249	+ ))}
	1250	+ </div>
	1251	+ )}
951	1252	<textarea
952	1253	ref={textareaRef}
953	1254	className="w-full resize-none bg-transparent text-[14px] leading-relaxed text-foreground placeholder:text-muted-foreground/35 focus:outline-none"
@@ -955,8 +1256,11 @@ export default function ChatView() { · 1 unmodified line
955	1256	value={prompt}
956	1257	onChange={(e) => setPrompt(e.target.value)}
957	1258	onKeyDown={onKeyDown}
	1259	+ onPaste={onComposerPaste}
958	1260	placeholder={
959		− phase === "disconnected" ? "Ask for follow-up changes" : "Ask anything..."
	1261	+ phase === "disconnected"
	1262	+ ? "Ask for follow-up changes or attach images"
	1263	+ : "Ask anything, or attach images..."
960	1264	}
961	1265	disabled={isSending \|\| isConnecting}
962	1266	/>
@@ -1146,7 +1450,7 @@ export default function ChatView() { · 183 unmodified lines
1146	1450	<button
1147	1451	type="submit"
1148	1452	className="flex h-8 w-8 items-center justify-center rounded-full bg-primary/90 text-primary-foreground transition-all duration-150 hover:bg-primary hover:scale-105 disabled:opacity-30 disabled:hover:scale-100"
1149		− disabled={isSending \|\| isConnecting \|\| !prompt.trim()}
	1453	+ disabled={isSending \|\| isConnecting \|\| (!prompt.trim() && composerImages.length === 0)}
1150	1454	aria-label={
1151	1455	isConnecting ? "Connecting" : isSending ? "Sending" : "Send message"
1152	1456	}
@@ -1221,6 +1525,39 @@ export default function ChatView() { · 68 unmodified lines
1221	1525	}
1222	1526	/>
1223	1527	)}
	1528	+
	1529	+ {expandedImage && (
	1530	+ <div
	1531	+ className="fixed inset-0 z-50 flex items-center justify-center bg-black/75 px-4 py-6 [-webkit-app-region:no-drag]"
	1532	+ role="dialog"
	1533	+ aria-modal="true"
	1534	+ aria-label="Expanded image preview"
	1535	+ onClick={() => setExpandedImage(null)}
	1536	+ >
	1537	+ <div
	1538	+ className="relative isolate max-h-[92vh] max-w-[92vw]"
	1539	+ onClick={(event) => event.stopPropagation()}
	1540	+ >
	1541	+ <button
	1542	+ type="button"
	1543	+ className="absolute right-2 top-2 z-30 flex h-8 w-8 cursor-pointer items-center justify-center rounded-full bg-black/60 text-xl leading-none text-white transition-colors duration-150 hover:bg-black/80 focus:outline-none focus-visible:ring-2 focus-visible:ring-white/80 [-webkit-app-region:no-drag]"
	1544	+ onClick={() => setExpandedImage(null)}
	1545	+ aria-label="Close image preview"
	1546	+ >
	1547	+ <span className="pointer-events-none">×</span>
	1548	+ </button>
	1549	+ <img
	1550	+ src={expandedImage.src}
	1551	+ alt={expandedImage.name}
	1552	+ className="max-h-[86vh] max-w-[92vw] select-none rounded-lg border border-border/70 bg-background object-contain shadow-2xl"
	1553	+ draggable={false}
	1554	+ />
	1555	+ <p className="mt-2 max-w-[92vw] truncate text-center text-xs text-muted-foreground/80">
	1556	+ {expandedImage.name}
	1557	+ </p>
	1558	+ </div>
	1559	+ </div>
	1560	+ )}
1224	1561	</div>
1225	1562	);
1226	1563	}

apps/web/src/components/DiffPanel.tsx+1−1

@@ -8,7 +8,7 @@ export default function DiffPanel() { · 7 unmodified lines
8	8	<aside className="flex h-full w-[400px] shrink-0 flex-col border-l border-border bg-card">
9	9	{/* Header */}
10	10	<div
11		− className={`flex items-center justify-between border-b border-border px-4 pb-3 ${isElectron ? "drag-region pt-[28px]" : "pt-3"}`}
	11	+ className={`flex items-center justify-between border-b border-border px-4 ${isElectron ? "drag-region h-[52px]" : "py-3"}`}
12	12	>
13	13	<h3 className="text-xs font-medium text-foreground">Uncommitted changes</h3>
14	14	<button

apps/web/src/components/Sidebar.tsx+3−3

@@ -237,10 +237,10 @@ export default function Sidebar() { · 236 unmodified lines
237	237
238	238	return (
239	239	<aside className="sidebar flex h-full w-[260px] shrink-0 flex-col border-r border-border bg-card">
240		− {/* Drag region / traffic light space (Electron only) */}
241		− {isElectron && <div className="drag-region h-[52px] shrink-0" />}
242	240	{/* Branding */}
243		− <div className={`flex items-center gap-2.5 px-4 pb-4 ${isElectron ? "" : "pt-4"}`}>
	241	+ <div
	242	+ className={`flex items-center gap-2.5 px-4 ${isElectron ? "drag-region h-[52px] pl-[76px]" : "py-4"}`}
	243	+ >
244	244	<span className="flex-1 text-sm font-semibold tracking-tight text-foreground">
245	245	T3 <span className="font-normal text-muted-foreground">Code</span>
246	246	</span>

apps/web/src/historyBootstrap.test.ts+28−0

@@ -91,4 +91,32 @@ describe("buildBootstrapInput", () => { · 90 unmodified lines
91	91	expect(result.omittedCount).toBe(1);
92	92	expect(result.truncated).toBe(true);
93	93	});
	94	+
	95	+ it("captures user image attachment context in transcript blocks", () => {
	96	+ const result = buildBootstrapInput(
	97	+ [
	98	+ {
	99	+ id: "u-image",
	100	+ role: "user",
	101	+ text: "",
	102	+ attachments: [
	103	+ {
	104	+ type: "image",
	105	+ id: "img-1",
	106	+ name: "screenshot.png",
	107	+ mimeType: "image/png",
	108	+ sizeBytes: 2_048,
	109	+ },
	110	+ ],
	111	+ createdAt: "2026-02-09T00:00:00.000Z",
	112	+ streaming: false,
	113	+ },
	114	+ ],
	115	+ "What does this error mean?",
	116	+ 1_500,
	117	+ );
	118	+
	119	+ expect(result.text).toContain("Attached image");
	120	+ expect(result.text).toContain("screenshot.png");
	121	+ });
94	122	});

apps/web/src/historyBootstrap.ts+29−3

@@ -18,8 +18,34 @@ function messageRoleLabel(message: ChatMessage): "USER" \| "ASSISTANT" { · 17 unmodified lines
18	18	return message.role === "assistant" ? "ASSISTANT" : "USER";
19	19	}
20	20
	21	+function attachmentSummary(message: ChatMessage): string \| null {
	22	+ const imageAttachments = message.attachments?.filter((attachment) => attachment.type === "image");
	23	+ const count = imageAttachments?.length ?? 0;
	24	+ if (count === 0) {
	25	+ return null;
	26	+ }
	27	+
	28	+ const names = imageAttachments?.slice(0, 3).map((image) => image.name) ?? [];
	29	+ const namesSummary = names.join(", ");
	30	+ const extraCount = count - names.length;
	31	+ const extraSummary = extraCount > 0 ? ` (+${extraCount} more)` : "";
	32	+ return `[Attached image${count === 1 ? "" : "s"}: ${namesSummary}${extraSummary}]`;
	33	+}
	34	+
21	35	function buildMessageBlock(message: ChatMessage): string {
22		− return `${messageRoleLabel(message)}:\n${message.text}`;
	36	+ const text = message.text;
	37	+ const attachments = attachmentSummary(message);
	38	+
	39	+ if (text && attachments) {
	40	+ return `${messageRoleLabel(message)}:\n${text}\n${attachments}`;
	41	+ }
	42	+ if (text) {
	43	+ return `${messageRoleLabel(message)}:\n${text}`;
	44	+ }
	45	+ if (attachments) {
	46	+ return `${messageRoleLabel(message)}:\n${attachments}`;
	47	+ }
	48	+ return `${messageRoleLabel(message)}:\n(empty message)`;
23	49	}
24	50
25	51	function finalizeWithPrompt(
@@ -68,7 +94,7 @@ export function buildBootstrapInput( · 42 unmodified lines
68	94	let includedNewestFirst: string[] = [];
69	95	for (const block of newestFirstBlocks) {
70	96	const nextNewestFirst = [...includedNewestFirst, block];
71		− const nextChronological = [...nextNewestFirst].reverse();
	97	+ const nextChronological = nextNewestFirst.toReversed();
72	98	const omittedCount = newestFirstBlocks.length - nextChronological.length;
73	99	const transcriptBody =
74	100	omittedCount > 0
@@ -80,7 +106,7 @@ export function buildBootstrapInput( · 5 unmodified lines
80	106	includedNewestFirst = nextNewestFirst;
81	107	}
82	108
83		− let includedChronological = [...includedNewestFirst].reverse();
	109	+ let includedChronological = includedNewestFirst.toReversed();
84	110	while (true) {
85	111	const omittedCount = newestFirstBlocks.length - includedChronological.length;
86	112	const transcriptBody =

apps/web/src/persistenceSchema.test.ts+28−1

@@ -200,6 +200,16 @@ describe("toPersistedState", () => { · 199 unmodified lines
200	200	id: "m-1",
201	201	role: "user",
202	202	text: "Hi",
	203	+ attachments: [
	204	+ {
	205	+ type: "image",
	206	+ id: "img-1",
	207	+ name: "diagram.png",
	208	+ mimeType: "image/png",
	209	+ sizeBytes: 4_096,
	210	+ previewUrl: "blob:preview-1",
	211	+ },
	212	+ ],
203	213	createdAt: "2026-02-08T10:00:00.000Z",
204	214	streaming: false,
205	215	},
@@ -236,7 +246,24 @@ describe("toPersistedState", () => { · 30 unmodified lines
236	246	model: "gpt-5.3-codex",
237	247	terminalOpen: true,
238	248	terminalHeight: 320,
239		− messages: thread.messages,
	249	+ messages: [
	250	+ {
	251	+ id: "m-1",
	252	+ role: "user",
	253	+ text: "Hi",
	254	+ attachments: [
	255	+ {
	256	+ type: "image",
	257	+ id: "img-1",
	258	+ name: "diagram.png",
	259	+ mimeType: "image/png",
	260	+ sizeBytes: 4_096,
	261	+ },
	262	+ ],
	263	+ createdAt: "2026-02-08T10:00:00.000Z",
	264	+ streaming: false,
	265	+ },
	266	+ ],
240	267	createdAt: thread.createdAt,
241	268	branch: null,
242	269	worktreePath: null,

apps/web/src/persistenceSchema.ts+42−5

@@ -23,6 +23,17 @@ const persistedMessageSchema = z.object({ · 22 unmodified lines
23	23	id: z.string().min(1),
24	24	role: z.enum(["user", "assistant"]),
25	25	text: z.string(),
	26	+ attachments: z
	27	+ .array(
	28	+ z.object({
	29	+ type: z.literal("image"),
	30	+ id: z.string().min(1),
	31	+ name: z.string().min(1),
	32	+ mimeType: z.string().min(1),
	33	+ sizeBytes: z.number().int().min(1),
	34	+ }),
	35	+ )
	36	+ .optional(),
26	37	createdAt: z.string().min(1),
27	38	streaming: z.boolean(),
28	39	});
@@ -126,10 +137,19 @@ function hydrateThread( · 97 unmodified lines
126	137	terminalOpen: thread.terminalOpen ?? false,
127	138	terminalHeight: thread.terminalHeight ?? DEFAULT_THREAD_TERMINAL_HEIGHT,
128	139	session: null,
129		− messages: thread.messages.map((message) => ({
130		− ...message,
131		− streaming: false,
132		− })),
	140	+ messages: thread.messages.map((message) => {
	141	+ const hydratedAttachments = message.attachments?.map((attachment) => ({ ...attachment }));
	142	+ return {
	143	+ id: message.id,
	144	+ role: message.role,
	145	+ text: message.text,
	146	+ ...(hydratedAttachments && hydratedAttachments.length > 0
	147	+ ? { attachments: hydratedAttachments }
	148	+ : {}),
	149	+ createdAt: message.createdAt,
	150	+ streaming: false,
	151	+ };
	152	+ }),
133	153	events: [],
134	154	error: null,
135	155	createdAt: thread.createdAt,
@@ -189,7 +209,24 @@ export function toPersistedState( · 53 unmodified lines
189	209	model: thread.model,
190	210	terminalOpen: thread.terminalOpen,
191	211	terminalHeight: thread.terminalHeight,
192		− messages: thread.messages,
	212	+ messages: thread.messages.map((message) => ({
	213	+ id: message.id,
	214	+ role: message.role,
	215	+ text: message.text,
	216	+ ...(message.attachments && message.attachments.length > 0
	217	+ ? {
	218	+ attachments: message.attachments.map((attachment) => ({
	219	+ type: attachment.type,
	220	+ id: attachment.id,
	221	+ name: attachment.name,
	222	+ mimeType: attachment.mimeType,
	223	+ sizeBytes: attachment.sizeBytes,
	224	+ })),
	225	+ }
	226	+ : {}),
	227	+ createdAt: message.createdAt,
	228	+ streaming: message.streaming,
	229	+ })),
193	230	createdAt: thread.createdAt,
194	231	branch: thread.branch,
195	232	worktreePath: thread.worktreePath,

apps/web/src/store.ts+11−1

@@ -13,6 +13,7 @@ import { resolveModelSlug } from "./model-logic"; · 12 unmodified lines
13	13	import { hydratePersistedState, toPersistedState } from "./persistenceSchema";
14	14	import { applyEventToMessages, asObject, asString, evolveSession } from "./session-logic";
15	15	import {
	16	+ type ChatAttachment,
16	17	DEFAULT_RUNTIME_MODE,
17	18	type Project,
18	19	type RuntimeMode,
@@ -37,7 +38,13 @@ type Action = · 18 unmodified lines
37	38	activeAssistantItemRef: { current: string \| null };
38	39	}
39	40	\| { type: "UPDATE_SESSION"; threadId: string; session: ProviderSession }
40		− \| { type: "PUSH_USER_MESSAGE"; threadId: string; id: string; text: string }
	41	+ \| {
	42	+ type: "PUSH_USER_MESSAGE";
	43	+ threadId: string;
	44	+ id: string;
	45	+ text: string;
	46	+ attachments?: ChatAttachment[];
	47	+ }
41	48	\| { type: "SET_ERROR"; threadId: string; error: string \| null }
42	49	\| { type: "SET_THREAD_TITLE"; threadId: string; title: string }
43	50	\| { type: "SET_THREAD_MODEL"; threadId: string; model: string }
@@ -380,6 +387,9 @@ export function reducer(state: AppState, action: Action): AppState { · 336 unmodified lines
380	387	id: action.id,
381	388	role: "user" as const,
382	389	text: action.text,
	390	+ ...(action.attachments && action.attachments.length > 0
	391	+ ? { attachments: action.attachments }
	392	+ : {}),
383	393	createdAt: new Date().toISOString(),
384	394	streaming: false,
385	395	},

apps/web/src/types.ts+12−0

@@ -5,10 +5,22 @@ export type RuntimeMode = "approval-required" \| "full-access"; · 4 unmodified lines
5	5	export const DEFAULT_RUNTIME_MODE: RuntimeMode = "full-access";
6	6	export const DEFAULT_THREAD_TERMINAL_HEIGHT = 280;
7	7
	8	+export interface ChatImageAttachment {
	9	+ type: "image";
	10	+ id: string;
	11	+ name: string;
	12	+ mimeType: string;
	13	+ sizeBytes: number;
	14	+ previewUrl?: string;
	15	+}
	16	+
	17	+export type ChatAttachment = ChatImageAttachment;
	18	+
8	19	export interface ChatMessage {
9	20	id: string;
10	21	role: "user" \| "assistant";
11	22	text: string;
	23	+ attachments?: ChatAttachment[];
12	24	createdAt: string;
13	25	streaming: boolean;
14	26	}

packages/contracts/src/model.ts+8−1

@@ -1,10 +1,17 @@
1		−export const MODEL_OPTIONS = ["gpt-5.3-codex", "gpt-5.2-codex", "gpt-5.2"] as const;
	1	+export const MODEL_OPTIONS = [
	2	+ "gpt-5.3-codex",
	3	+ "gpt-5.3-codex-spark",
	4	+ "gpt-5.2-codex",
	5	+ "gpt-5.2",
	6	+] as const;
2	7
3	8	export const DEFAULT_MODEL = "gpt-5.3-codex";
4	9
5	10	export const MODEL_SLUG_ALIASES: Record<string, string> = {
6	11	"5.3": "gpt-5.3-codex",
7	12	"gpt-5.3": "gpt-5.3-codex",
	13	+ "5.3-spark": "gpt-5.3-codex-spark",
	14	+ "gpt-5.3-spark": "gpt-5.3-codex-spark",
8	15	};
9	16
10	17	export function normalizeModelSlug(model: string \| null \| undefined): string \| null {

packages/contracts/src/provider.test.ts+59−0

@@ -1,6 +1,7 @@
1	1	import { describe, expect, it } from "vitest";
2	2
3	3	import {
	4	+ PROVIDER_SEND_TURN_MAX_ATTACHMENTS,
4	5	providerEventSchema,
5	6	providerRespondToRequestInputSchema,
6	7	providerSendTurnInputSchema,
@@ -40,9 +41,67 @@ describe("providerSendTurnInputSchema", () => { · 33 unmodified lines
40	41	effort: " high ",
41	42	});
42	43	expect(parsed.input).toBe("summarize this repo");
	44	+ expect(parsed.attachments).toEqual([]);
43	45	expect(parsed.model).toBe("gpt-5.2-codex");
44	46	expect(parsed.effort).toBe("high");
45	47	});
	48	+
	49	+ it("accepts image-only turns", () => {
	50	+ const parsed = providerSendTurnInputSchema.parse({
	51	+ sessionId: "sess_1",
	52	+ attachments: [
	53	+ {
	54	+ type: "image",
	55	+ name: "diagram.png",
	56	+ mimeType: "image/png",
	57	+ sizeBytes: 1_024,
	58	+ dataUrl: "data:image/png;base64,AAAA",
	59	+ },
	60	+ ],
	61	+ });
	62	+ expect(parsed.input).toBeUndefined();
	63	+ expect(parsed.attachments).toHaveLength(1);
	64	+ });
	65	+
	66	+ it("rejects turns with neither text nor attachments", () => {
	67	+ expect(() =>
	68	+ providerSendTurnInputSchema.parse({
	69	+ sessionId: "sess_1",
	70	+ }),
	71	+ ).toThrow();
	72	+ });
	73	+
	74	+ it("rejects non-image data urls", () => {
	75	+ expect(() =>
	76	+ providerSendTurnInputSchema.parse({
	77	+ sessionId: "sess_1",
	78	+ attachments: [
	79	+ {
	80	+ type: "image",
	81	+ name: "not-image.txt",
	82	+ mimeType: "text/plain",
	83	+ sizeBytes: 25,
	84	+ dataUrl: "data:text/plain;base64,QQ==",
	85	+ },
	86	+ ],
	87	+ }),
	88	+ ).toThrow();
	89	+ });
	90	+
	91	+ it("rejects more than the max attachment count", () => {
	92	+ expect(() =>
	93	+ providerSendTurnInputSchema.parse({
	94	+ sessionId: "sess_1",
	95	+ attachments: Array.from({ length: PROVIDER_SEND_TURN_MAX_ATTACHMENTS + 1 }, (_, index) => ({
	96	+ type: "image" as const,
	97	+ name: `image-${index}.png`,
	98	+ mimeType: "image/png",
	99	+ sizeBytes: 1_024,
	100	+ dataUrl: "data:image/png;base64,AAAA",
	101	+ })),
	102	+ }),
	103	+ ).toThrow();
	104	+ });
46	105	});
47	106
48	107	describe("providerEventSchema", () => {

packages/contracts/src/provider.ts+43−6

@@ -55,14 +55,48 @@ export const providerSessionStartInputSchema = z.object({ · 54 unmodified lines
55	55	});
56	56
57	57	export const PROVIDER_SEND_TURN_MAX_INPUT_CHARS = 120_000;
58		−
59		−export const providerSendTurnInputSchema = z.object({
60		− sessionId: z.string().min(1),
61		− input: z.string().trim().min(1).max(PROVIDER_SEND_TURN_MAX_INPUT_CHARS),
62		− model: z.string().trim().min(1).optional(),
63		− effort: z.string().trim().min(1).optional(),
	58	+export const PROVIDER_SEND_TURN_MAX_ATTACHMENTS = 8;
	59	+export const PROVIDER_SEND_TURN_MAX_IMAGE_BYTES = 10 * 1024 * 1024;
	60	+export const PROVIDER_SEND_TURN_MAX_IMAGE_DATA_URL_CHARS = 14_000_000;
	61	+
	62	+export const providerSendTurnImageAttachmentSchema = z.object({
	63	+ type: z.literal("image"),
	64	+ name: z.string().trim().min(1).max(255),
	65	+ mimeType: z
	66	+ .string()
	67	+ .trim()
	68	+ .min(1)
	69	+ .max(100)
	70	+ .regex(/^image\//i, "mimeType must be an image/* MIME type"),
	71	+ sizeBytes: z.number().int().min(1).max(PROVIDER_SEND_TURN_MAX_IMAGE_BYTES),
	72	+ dataUrl: z.string().trim().min(1).max(PROVIDER_SEND_TURN_MAX_IMAGE_DATA_URL_CHARS),
64	73	});
65	74
	75	+export const providerSendTurnAttachmentInputSchema = z.discriminatedUnion("type", [
	76	+ providerSendTurnImageAttachmentSchema,
	77	+]);
	78	+
	79	+export const providerSendTurnInputSchema = z
	80	+ .object({
	81	+ sessionId: z.string().min(1),
	82	+ input: z.string().trim().min(1).max(PROVIDER_SEND_TURN_MAX_INPUT_CHARS).optional(),
	83	+ attachments: z
	84	+ .array(providerSendTurnAttachmentInputSchema)
	85	+ .max(PROVIDER_SEND_TURN_MAX_ATTACHMENTS)
	86	+ .default([]),
	87	+ model: z.string().trim().min(1).optional(),
	88	+ effort: z.string().trim().min(1).optional(),
	89	+ })
	90	+ .superRefine((value, ctx) => {
	91	+ if (!value.input && value.attachments.length === 0) {
	92	+ ctx.addIssue({
	93	+ code: z.ZodIssueCode.custom,
	94	+ message: "Either input text or at least one attachment is required",
	95	+ path: ["input"],
	96	+ });
	97	+ }
	98	+ });
	99	+
66	100	export const providerTurnStartResultSchema = z.object({
67	101	threadId: z.string().min(1),
68	102	turnId: z.string().min(1),
@@ -110,6 +144,9 @@ export type ProviderApprovalDecision = z.infer<typeof providerApprovalDecisionSc · 41 unmodified lines
110	144	export type ProviderSessionStatus = z.infer<typeof providerSessionStatusSchema>;
111	145	export type ProviderSession = z.infer<typeof providerSessionSchema>;
112	146	export type ProviderSessionStartInput = z.input<typeof providerSessionStartInputSchema>;
	147	+export type ProviderSendTurnImageAttachment = z.infer<typeof providerSendTurnImageAttachmentSchema>;
	148	+export type ProviderSendTurnAttachment = z.infer<typeof providerSendTurnAttachmentInputSchema>;
	149	+export type ProviderSendTurnAttachmentInput = z.input<typeof providerSendTurnAttachmentInputSchema>;
113	150	export type ProviderSendTurnInput = z.input<typeof providerSendTurnInputSchema>;
114	151	export type ProviderTurnStartResult = z.infer<typeof providerTurnStartResultSchema>;
115	152	export type ProviderInterruptTurnInput = z.input<typeof providerInterruptTurnInputSchema>;