Files
copilotkit__copilotkit/showcase/integrations/ms-agent-python/tests/e2e/multimodal.spec.ts
Alem Tuzlak 5cd77233b7 feat(showcase/ms-agent-python): LGP parity sweep — 33/37 cells green
Brings ms-agent-python to one-to-one parity with langgraph-python (the D5
north star). Playwright e2e suite goes from 49/108 (~26%) → 164/178 (~92%),
33 of 37 cells fully green.

Manifest parity:
- Drop 4 MAF-only cells with no LGP analog: agentic-chat-reasoning,
  hitl-in-chat-booking, shared-state-write, reasoning-default-render.
  Reasoning is handled by reasoning-default + reasoning-custom (LGP);
  booking pill folds into hitl-in-chat; shared-state-write was a TODO stub.
- Rename byoc-hashbrown → declarative-hashbrown and byoc-json-render →
  declarative-json-render. Demo dir, API route dir, and frontend agent id
  follow LGP's naming. Python module files retain the legacy `byoc_*`
  prefix and FastAPI paths stay `/byoc-hashbrown` / `/byoc-json-render`
  (matches LGP's "module name retains legacy graph id" convention).
- Port LGP `_shared/`, `_shared/interrupt-fallback-slots.ts`, and
  `demos/layout.tsx` for one-to-one parity.

Cells ported verbatim from LGP (page + spec):
- agentic-chat, auth, beautiful-chat, chat-customization-css, chat-slots,
  declarative-gen-ui, declarative-hashbrown, declarative-json-render,
  frontend-tools, frontend-tools-async, gen-ui-agent, gen-ui-interrupt,
  gen-ui-tool-based, headless-complete, headless-simple, hitl-in-app,
  hitl-in-chat, shared-state-read, shared-state-read-write,
  shared-state-streaming, subagents, tool-rendering, plus all four
  tool-rendering* variants, a2ui-fixed-schema, agent-config, mcp-apps,
  multimodal, open-gen-ui, open-gen-ui-advanced, prebuilt-popup,
  prebuilt-sidebar, readonly-state-agent-context, reasoning-default,
  reasoning-custom, voice.

Backend infrastructure:
- Swap shared `OpenAIChatClient` (Responses API) → `OpenAIChatCompletionClient`
  (ChatCompletions). Root cause of the cross-cell post-tool ChatClientException
  family: Responses API is stateful and only sends NEW items per leg,
  relying on `previous_response_id` for history. aimock has no view of
  that server-side state, so second-leg requests arrived without the
  user message — fixture matchers keyed on `userMessage` couldn't fire
  and the run fell through to real OpenAI. ChatCompletions sends full
  history every leg, matching the LGP wire shape.
- Bump @ag-ui/client ^0.0.43 → ^0.0.53 (matches google-adk/LGP). Fixes
  the REASONING_* Zod discriminator trap on the catch-all agent.
- Regenerate package-lock.json in isolation outside the pnpm monorepo so
  npm-arborist doesn't resolve transitives against pnpm's hoisted
  symlinks (avoid 40+ `../../../node_modules/.pnpm/...` paths in the
  lockfile that break `npm ci` inside Docker).
- Add `yaml` (^2.8.4) for the new `src/app/demos/layout.tsx` that reads
  manifest.yaml for per-cell page titles (LGP parity).

New / re-added MAF agent backends with LGP-equivalent behavior:
- reasoning_agent.py (uses Responses API explicitly — the only chat
  client that emits AG-UI REASONING_MESSAGE_* events; rest of the
  integration stays on ChatCompletions).
- tool_rendering_agent.py (non-reasoning sibling of the existing
  reasoning_chain variant; shares tool surface via direct imports so
  they can never drift apart; routes the three catchall cells to a
  non-reasoning backend so the default renderer spec stops failing on
  leaked reasoning blocks).
- gen_ui_agent.py — `set_steps` tool + `steps` state schema +
  `predict_state_config` mirrors LGP's StateStreamingMiddleware shape.
- shared_state_streaming.py — `write_document` tool with
  `predict_state_config` that streams the `document` arg into
  `state.document` per-token.
- readonly_state_agent_context.py — minimal agent that consumes
  frontend-provided `useAgentContext` entries; no tools.
- headless_complete_agent.py — three deterministic tools (`get_weather`,
  `get_stock_price`, `get_revenue_chart`) mounted at /headless-complete
  on the mcp-apps runtime (was routing to catch-all sales agent, which
  returned seeded-random weather instead of the deterministic 68°F the
  test asserts on).

Wiring:
- copilotkit/route.ts: register the new agents, drop the stale
  shared-state-write entry, route all three tool-rendering variants to
  the non-reasoning backend (the reasoning-chain cell keeps its own
  dedicated path), register reasoning-default + reasoning-custom on
  /reasoning, register gen-ui-agent on /gen-ui-agent,
  shared-state-streaming on /shared-state-streaming,
  readonly-state-agent-context on its dedicated path.
- copilotkit-mcp-apps/route.ts: register headless-complete agent (was
  missing — the strict useAgent runtime sync in the newer
  @copilotkit/react-core surfaced the gap).
- copilotkit-declarative-hashbrown/route.ts + copilotkit-declarative-json-render/route.ts:
  new dedicated runtimes; agent IDs and runtime URLs follow LGP.
- copilotkit-declarative-gen-ui/route.ts: drop non-LGP `openGenerativeUI:
  false` for parity.

A2UI tool rename — `render_a2ui` → `_design_a2ui_surface`:
- Ported LGP's `tools/generate_a2ui.py` (LGP renamed the secondary-LLM
  tool to `_design_a2ui_surface` to avoid the A2UI middleware's bypass;
  shared d5-all.json fixtures key the response on this name).
- Renamed every `render_a2ui` occurrence in src/agents/{a2ui_dynamic,
  agent,beautiful_chat}.py and `tools/__init__.py`.
- Updated 4 declarative-gen-ui aimock fixtures to pass `context` arg in
  the first-leg `generate_a2ui` tool call (agent_framework doesn't
  auto-inject AgentSession into our @tool function so `session=None` and
  the secondary-LLM `user_content` was defaulting to a catch-all string
  containing "KPI dashboard" — every pill matched the KPI fixture).

Aimock router patch persisted alongside the integration changes:
hasToolResult matcher restricted to scan only messages after the last
user message (was global). The patch lives in F:/projects/cpk/aimock —
upstream PR pending.

Test infrastructure:
- playwright.config.ts: cap local workers at 4 + retries at 1. CI keeps
  workers=1, retries=2. `agent_framework.Agent` is reused across requests
  and the shared OpenAI HTTP client serialises concurrent SSE streams;
  >4 workers makes 30s timeouts inevitable on a few cells. Confirmed
  with hard data: workers=1 = 164 passed (16.8 min), workers=4+retries=1
  = 164 passed (7.2 min), workers=undefined = 159 passed. Same green
  set, ~2x faster. Long-term upstream fix is per-request Agent
  instantiation in agent_framework_ag_ui.

Remaining 14 failures across 4 cells documented per-cell in the Notion
D5 sweep doc (declarative-gen-ui A2UI surface mounting, multimodal
attachment forwarding, tool-rendering-default-catchall multi-pill chain,
tool-rendering-reasoning-chain multi-leg chains). Each has a specific
next-pass action.
2026-05-19 18:36:01 +02:00

195 lines
8.1 KiB
TypeScript

import { test, expect } from "@playwright/test";
import type { Page } from "@playwright/test";
/**
* E2E spec for the Multimodal Attachments demo.
*
* Exercises the sample-file injection path only — the real OS file
* picker isn't automatable through Playwright without a flakey host-
* file dependency. The sample buttons go through the same V2 agent
* surface (`agent.addMessage` + `copilotkit.runAgent`) that the
* paperclip path ultimately feeds, so this suite covers the render +
* round-trip path end-to-end.
*
* Behavior pinned by this suite (each was an actual regression caught
* during the multimodal rewrite — see headers in
* sample-attachment-buttons.tsx and legacy-converter-shim.tsx):
*
* 1. Clicking a sample button auto-sends the canned prompt — no manual
* fill / send required. Earlier the DataTransfer-into-file-input
* approach raced the upload state, the send got rejected, and the
* prompt got eaten.
* 2. The user message keeps showing exactly ONE attachment chip after
* the assistant response arrives. The @ag-ui/langgraph round-trip
* used to duplicate the attachment (modern + re-converted shape) so
* the user saw two thumbnails for one file.
* 3. Sample-PDF specifically renders as a `DocumentAttachment` chip,
* NOT as an `<img>` that errors with "Failed to load image". Earlier
* the round-trip mis-tagged PDFs with `type: "image"` and forced the
* image renderer.
* 4. The PDF flattened text doesn't bleed into the rendered user
* message. Earlier `_PdfFlattenMiddleware` ran in `before_model`
* and persisted the flattened "[Attached document]\n<pdf body>"
* into state, which the chat UI rendered inline.
* 5. Clicking image then PDF in the same session (and vice versa)
* leaves each user message with its own single, correct chip — no
* cross-contamination, no doubling.
*
* Assumes the demo is reachable at BASE_URL/demos/multimodal and
* aimock is configured with the canned prompts:
* - "can you tell me what is in this demo image I just attached"
* - "can you tell me what is in this demo pdf I just attached"
* Both fixtures live in showcase/aimock/feature-parity.json.
*/
const ROUTE = "/demos/multimodal";
const SAMPLE_IMAGE_BTN = '[data-testid="multimodal-sample-image-button"]';
const SAMPLE_PDF_BTN = '[data-testid="multimodal-sample-pdf-button"]';
const CHAT_TEXTAREA = '[data-testid="copilot-chat-textarea"]';
const ADD_MENU_BUTTON = '[data-testid="copilot-add-menu-button"]';
const USER_MESSAGE = '[data-testid="copilot-user-message"]';
const ASSISTANT_MESSAGE = '[data-testid="copilot-assistant-message"]';
// The canned auto-prompts live in sample-attachment-buttons.tsx — kept in
// sync here so the user-message bubble assertion stays accurate.
const IMAGE_PROMPT =
"can you tell me what is in this demo image I just attached";
const PDF_PROMPT = "can you tell me what is in this demo pdf I just attached";
async function openDemo(page: Page) {
await page.goto(ROUTE);
await expect(
page.locator('[data-testid="multimodal-demo-root"]'),
).toBeVisible();
}
/**
* Wait until the sample-injection flow finishes for the nth user message:
* the user bubble shows up, then the matching assistant response. Returns
* both locators so the caller can probe for chips and dedup invariants.
*/
async function waitForRoundTrip(
page: Page,
index = 0,
): Promise<{
userMsg: ReturnType<Page["locator"]>;
asstMsg: ReturnType<Page["locator"]>;
}> {
const userMsg = page.locator(USER_MESSAGE).nth(index);
await expect(userMsg).toBeVisible({ timeout: 60_000 });
const asstMsg = page.locator(ASSISTANT_MESSAGE).nth(index);
await expect(asstMsg).toBeVisible({ timeout: 90_000 });
return { userMsg, asstMsg };
}
test.describe("Multimodal Attachments", () => {
test.beforeEach(async ({ page }) => {
await openDemo(page);
});
test("page loads with sample row, sample buttons, composer, and paperclip", async ({
page,
}) => {
await expect(
page.locator('[data-testid="multimodal-sample-row"]'),
).toBeVisible();
await expect(page.locator(SAMPLE_IMAGE_BTN)).toBeEnabled();
await expect(page.locator(SAMPLE_PDF_BTN)).toBeEnabled();
await expect(page.locator(CHAT_TEXTAREA)).toBeVisible();
await expect(page.locator(ADD_MENU_BUTTON)).toBeVisible();
});
test("sample image: auto-sends, user msg shows EXACTLY ONE image, assistant references it", async ({
page,
}) => {
await page.locator(SAMPLE_IMAGE_BTN).click();
const { userMsg, asstMsg } = await waitForRoundTrip(page);
// Auto-prompt landed as the user message body — confirms the
// useAgent / runAgent path fired and the prompt wasn't eaten.
await expect(userMsg).toContainText(IMAGE_PROMPT);
// Exactly ONE rendered image — pin the dedupe + normalize behavior
// so the @ag-ui/langgraph round-trip can't quietly start doubling
// attachments again.
await expect(userMsg.locator("img")).toHaveCount(1);
await expect(userMsg.getByText(/Failed to load image/i)).toHaveCount(0);
await expect(asstMsg).toContainText(/copilotkit|logo|image/i);
});
test("sample PDF: auto-sends, user msg shows EXACTLY ONE document chip (not a broken image)", async ({
page,
}) => {
await page.locator(SAMPLE_PDF_BTN).click();
const { userMsg, asstMsg } = await waitForRoundTrip(page);
await expect(userMsg).toContainText(PDF_PROMPT);
// Pinned regression: PDFs round-tripped through @ag-ui/langgraph
// used to come back as `type: "image"` with `mimeType:
// application/pdf`, forcing the image renderer to fail load and
// show "Failed to load image" twice. The page-level dedupe + type-
// normalize subscriber turns them back into `document` parts so
// the icon-+-filename renderer fires exactly once.
await expect(userMsg.getByText(/Failed to load image/i)).toHaveCount(0);
await expect(userMsg.locator("img")).toHaveCount(0);
// DocumentAttachment surfaces a "PDF" label via getDocumentIcon.
await expect(userMsg.getByText(/^PDF$/)).toBeVisible();
// Pinned regression: the PDF flattened text used to bleed into the
// rendered user message when the Python middleware mutated state
// via `before_model`. Switching to `wrap_model_call` scopes the
// rewrite to the model request only — the bracket-tagged dump must
// never appear in the UI bubble.
await expect(userMsg).not.toContainText("[Attached document]");
await expect(asstMsg).toContainText(/copilotkit/i);
});
test("image then PDF in same session: each message keeps its own chip, no doubling", async ({
page,
}) => {
await page.locator(SAMPLE_IMAGE_BTN).click();
const first = await waitForRoundTrip(page, 0);
await expect(first.userMsg.locator("img")).toHaveCount(1);
await page.locator(SAMPLE_PDF_BTN).click();
const second = await waitForRoundTrip(page, 1);
// First message still shows exactly one image — the second send
// must not re-render or double the prior attachment.
await expect(first.userMsg.locator("img")).toHaveCount(1);
// Second message is a clean PDF chip.
await expect(second.userMsg.locator("img")).toHaveCount(0);
await expect(second.userMsg.getByText(/Failed to load image/i)).toHaveCount(
0,
);
await expect(second.userMsg.getByText(/^PDF$/)).toBeVisible();
});
test("PDF then image in same session: each message keeps its own chip, no doubling", async ({
page,
}) => {
await page.locator(SAMPLE_PDF_BTN).click();
const first = await waitForRoundTrip(page, 0);
await expect(first.userMsg.locator("img")).toHaveCount(0);
await expect(first.userMsg.getByText(/^PDF$/)).toBeVisible();
await page.locator(SAMPLE_IMAGE_BTN).click();
const second = await waitForRoundTrip(page, 1);
// First message still shows the PDF chip — no contamination.
await expect(first.userMsg.getByText(/^PDF$/)).toBeVisible();
await expect(first.userMsg.locator("img")).toHaveCount(0);
// Second message has exactly one image, no broken-image fallback.
await expect(second.userMsg.locator("img")).toHaveCount(1);
await expect(second.userMsg.getByText(/Failed to load image/i)).toHaveCount(
0,
);
});
});