deer-flow/backend/tests/test_view_image_middleware.py
DanielWalnut cd5bedaa74
feat: MiniMax provider for image/video/podcast skills + new music-generation skill (#3437)
* docs(spec): MiniMax integration for generation skills + new music skill

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(plan): MiniMax generation providers implementation plan

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* test(skills): add importlib loader + FakeResp for skill tests

* test(skills): register loaded module in sys.modules; raise requests.HTTPError in FakeResp

* feat(image-generation): add MiniMax provider with env auto-detect

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(image-generation): guard unknown provider, derive ref MIME, strengthen tests

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(video-generation): add MiniMax provider with async poll/download

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(video-generation): surface base_resp errors while polling; add timeout test

* feat(podcast-generation): add MiniMax t2a_v2 provider with env auto-detect

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor(podcast-generation): restore TTS credential guard; add volcengine + voice tests

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(music-generation): new MiniMax music skill via skill-creator

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* refactor(music-generation): treat empty lyrics as absent; test no-audio-data path

* refactor(skills): add request timeouts to MiniMax network calls

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Potential fix for pull request finding 'Explicit returns mixed with implicit (fall through) returns'

Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>

* fix(models): strip inconsistent user-message names for MiniMax chat

DeerFlow middlewares tag user messages with provenance names (user-input, summary, loop_warning); langchain serializes them into the OpenAI-compatible payload and MiniMax rejects mismatched user-message names with "user name must be consistent (2013)". PatchedChatMiniMax now drops the per-message name from user-role messages. Point the config.example MiniMax models at PatchedChatMiniMax so they also get reasoning_content mapping.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(image-generation): MiniMax sends JSON prompt field, guard 1500-char limit

MiniMax image-01 takes one text string capped at 1500 chars, but the skill was sending the whole structured JSON. The MiniMax provider now extracts the JSON `prompt` field (relying on prompt_optimizer to expand it) and fails fast with a clear error before calling the API when that field exceeds 1500 chars. Authoring stays provider-agnostic; Gemini still receives the full JSON.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(podcast-generation): per-provider TTS concurrency and retry/backoff

Each TTS provider owns its concurrency internally — MiniMax runs single-threaded to reduce rate-limit failures, Volcengine keeps 4 workers — with automatic retry and backoff on transient HTTP and base_resp errors. No caller-facing concurrency knob.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(skills): address Copilot review comments on generation skills

- video: add raise_for_status + timeout to the Gemini download/POST/poll calls so non-2xx responses surface as clear HTTP errors instead of JSON/KeyError or hangs
- video: check the task Fail status before the generic base_resp check so the failure keeps its task_id context
- video/image: create the output file parent directory before writing (matching music-generation) so nested output paths do not raise FileNotFoundError
- music: require a non-empty prompt and fail fast with ValueError instead of sending an empty prompt to the API

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(scripts): reclaim dev ports across worktrees in make stop/dev

All deer-flow worktrees (main checkout + linked worktrees) hardcode the same dev ports (8001/3000/2026), so a service started from any worktree must be reclaimable from another. stop_all now resolves the set of worktree roots (DEERFLOW_ROOTS) and treats a process as deer-flow-owned when its open files live under any of them. It also force-kills survivors on 2026 alongside 8001/3000, fixing `make dev` aborting on the nginx port preflight when a prior nginx lingered on 2026.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(view-image): hide the injected image-context message from the UI

ViewImageMiddleware injects a HumanMessage (text + base64 images) so the vision model can see viewed images, but it was the only internal injector that set neither hide_from_ui nor a hidden name, so it leaked into the chat UI (and IM channels) as a user bubble reading "Here are the images you've viewed:". Mark it with additional_kwargs={"hide_from_ui": True}, matching todo/dynamic_context injections, which the frontend isHiddenFromUIMessage and the channel sender already honor. The model still receives the full content.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(minimax): mark M2.7 models as text-only (no vision)

MiniMax M2.7 / M2.7-highspeed do not support vision; only M3 does. The
provider config asserted vision support for M2.7 in four places.

- config.example.yaml: 4 M2.7 entries -> supports_vision: false
- backend/docs/CONFIGURATION.md: M2.7 + highspeed -> supports_vision: false
- wizard: add LLMProvider.model_vision_overrides + extra_config_for() so
  selecting an M2.7 model writes supports_vision: false while M3 (default)
  keeps vision; wire it through setup_wizard.py
- tests: M2.7-highspeed fixture -> supports_vision=False; add
  test_minimax_vision_is_per_model

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>
2026-06-08 22:04:38 +08:00

402 lines
17 KiB
Python

"""Unit tests for ViewImageMiddleware.
Tests cover the middleware's ability to inject image details (including base64
payloads) as a HumanMessage before the next LLM call, triggered only when the
previous assistant turn contained `view_image` tool calls that have all been
completed with corresponding ToolMessages.
Covered behavior:
- `_get_last_assistant_message` returns the most recent AIMessage (or None).
- `_has_view_image_tool` only matches assistant messages with `view_image` tool calls.
- `_all_tools_completed` verifies every tool call id has a matching ToolMessage.
- `_create_image_details_message` produces correctly structured content blocks.
- `_should_inject_image_message` gates injection on all preconditions, including
deduplication when an image-details message was already added.
- `_inject_image_message` returns a state update with a HumanMessage, or None
when injection is not warranted.
- `before_model` and `abefore_model` expose the same behavior sync/async.
"""
from types import SimpleNamespace
from unittest.mock import MagicMock
import pytest
from langchain_core.messages import AIMessage, HumanMessage, SystemMessage, ToolMessage
from deerflow.agents.middlewares.view_image_middleware import ViewImageMiddleware
def _view_image_call(call_id: str = "call_1", path: str = "/mnt/user-data/uploads/img.png") -> dict:
return {"name": "view_image", "id": call_id, "args": {"image_path": path}}
def _other_tool_call(call_id: str = "call_other", name: str = "bash") -> dict:
return {"name": name, "id": call_id, "args": {"command": "ls"}}
def _runtime() -> MagicMock:
"""Minimal Runtime stub. The middleware doesn't use it today, but the
interface requires it."""
return MagicMock()
class TestGetLastAssistantMessage:
def test_returns_none_on_empty_list(self):
mw = ViewImageMiddleware()
assert mw._get_last_assistant_message([]) is None
def test_returns_none_when_no_ai_message(self):
mw = ViewImageMiddleware()
messages = [
SystemMessage(content="sys"),
HumanMessage(content="hi"),
]
assert mw._get_last_assistant_message(messages) is None
def test_returns_most_recent_ai_message(self):
mw = ViewImageMiddleware()
older = AIMessage(content="older")
newer = AIMessage(content="newer")
messages = [HumanMessage(content="q"), older, HumanMessage(content="q2"), newer]
assert mw._get_last_assistant_message(messages) is newer
class TestHasViewImageTool:
def test_returns_false_when_tool_calls_attr_missing(self):
"""Exercise the `not hasattr(message, "tool_calls")` guard.
AIMessage always has a `tool_calls` attribute, so we use a plain
object that truly lacks the attribute to cover this branch.
"""
mw = ViewImageMiddleware()
msg = SimpleNamespace(content="just text") # no tool_calls attribute
assert not hasattr(msg, "tool_calls") # precondition
assert mw._has_view_image_tool(msg) is False
def test_returns_false_when_ai_message_has_no_tool_calls(self):
"""AIMessage without tool_calls kwarg defaults to an empty list."""
mw = ViewImageMiddleware()
msg = AIMessage(content="just text")
assert mw._has_view_image_tool(msg) is False
def test_returns_false_when_tool_calls_empty(self):
mw = ViewImageMiddleware()
msg = AIMessage(content="", tool_calls=[])
assert mw._has_view_image_tool(msg) is False
def test_returns_true_when_view_image_present(self):
mw = ViewImageMiddleware()
msg = AIMessage(content="", tool_calls=[_view_image_call()])
assert mw._has_view_image_tool(msg) is True
def test_returns_true_when_view_image_mixed_with_others(self):
mw = ViewImageMiddleware()
msg = AIMessage(
content="",
tool_calls=[_other_tool_call(), _view_image_call(call_id="call_vi")],
)
assert mw._has_view_image_tool(msg) is True
def test_returns_false_when_only_other_tools(self):
mw = ViewImageMiddleware()
msg = AIMessage(content="", tool_calls=[_other_tool_call()])
assert mw._has_view_image_tool(msg) is False
class TestAllToolsCompleted:
def test_returns_false_when_no_tool_calls(self):
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[])
assert mw._all_tools_completed([assistant], assistant) is False
def test_returns_true_when_all_completed(self):
mw = ViewImageMiddleware()
assistant = AIMessage(
content="",
tool_calls=[_view_image_call("c1"), _view_image_call("c2", "/p2.png")],
)
messages = [
assistant,
ToolMessage(content="ok", tool_call_id="c1"),
ToolMessage(content="ok", tool_call_id="c2"),
]
assert mw._all_tools_completed(messages, assistant) is True
def test_returns_false_when_some_tool_call_unanswered(self):
mw = ViewImageMiddleware()
assistant = AIMessage(
content="",
tool_calls=[_view_image_call("c1"), _view_image_call("c2", "/p2.png")],
)
messages = [assistant, ToolMessage(content="ok", tool_call_id="c1")]
assert mw._all_tools_completed(messages, assistant) is False
def test_returns_false_when_assistant_not_in_messages(self):
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[_view_image_call("c1")])
# assistant is not part of the list, so messages.index() will raise and be caught
messages = [HumanMessage(content="hi")]
assert mw._all_tools_completed(messages, assistant) is False
def test_ignores_tool_messages_before_assistant(self):
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[_view_image_call("c1")])
# A stale ToolMessage with matching id appears BEFORE the assistant turn.
# It should not count — only ToolMessages after the assistant close the call.
messages = [
ToolMessage(content="stale", tool_call_id="c1"),
assistant,
]
assert mw._all_tools_completed(messages, assistant) is False
class TestCreateImageDetailsMessage:
def test_returns_placeholder_when_no_images(self):
mw = ViewImageMiddleware()
state = {"viewed_images": {}}
blocks = mw._create_image_details_message(state)
assert blocks == [{"type": "text", "text": "No images have been viewed."}]
def test_returns_placeholder_when_state_missing_key(self):
mw = ViewImageMiddleware()
blocks = mw._create_image_details_message({})
assert blocks == [{"type": "text", "text": "No images have been viewed."}]
def test_builds_blocks_for_single_image(self):
mw = ViewImageMiddleware()
state = {
"viewed_images": {
"/path/to/cat.png": {"base64": "BASE64DATA", "mime_type": "image/png"},
}
}
blocks = mw._create_image_details_message(state)
# header text + per-image description text + per-image image_url block
assert len(blocks) == 3
assert blocks[0] == {"type": "text", "text": "Here are the images you've viewed:"}
assert blocks[1]["type"] == "text"
assert "/path/to/cat.png" in blocks[1]["text"]
assert "image/png" in blocks[1]["text"]
assert blocks[2] == {
"type": "image_url",
"image_url": {"url": "data:image/png;base64,BASE64DATA"},
}
def test_builds_blocks_for_multiple_images(self):
mw = ViewImageMiddleware()
state = {
"viewed_images": {
"/a.png": {"base64": "AAA", "mime_type": "image/png"},
"/b.jpg": {"base64": "BBB", "mime_type": "image/jpeg"},
}
}
blocks = mw._create_image_details_message(state)
# 1 header + (1 description + 1 image_url) per image = 5 blocks
assert len(blocks) == 5
image_url_blocks = [b for b in blocks if isinstance(b, dict) and b.get("type") == "image_url"]
assert len(image_url_blocks) == 2
urls = {b["image_url"]["url"] for b in image_url_blocks}
assert "data:image/png;base64,AAA" in urls
assert "data:image/jpeg;base64,BBB" in urls
def test_omits_image_url_block_when_base64_missing(self):
mw = ViewImageMiddleware()
state = {
"viewed_images": {
"/broken.png": {"base64": "", "mime_type": "image/png"},
}
}
blocks = mw._create_image_details_message(state)
# header + description only (no image_url since base64 is empty)
assert len(blocks) == 2
assert all(not (isinstance(b, dict) and b.get("type") == "image_url") for b in blocks)
def test_uses_unknown_mime_type_when_missing(self):
mw = ViewImageMiddleware()
state = {
"viewed_images": {
"/mystery.bin": {"base64": "XYZ"}, # no mime_type key
}
}
blocks = mw._create_image_details_message(state)
# The description block should mention unknown
description_blocks = [b for b in blocks if b.get("type") == "text" and "/mystery.bin" in b.get("text", "")]
assert len(description_blocks) == 1
assert "unknown" in description_blocks[0]["text"]
class TestShouldInjectImageMessage:
def test_false_when_no_messages(self):
mw = ViewImageMiddleware()
assert mw._should_inject_image_message({"messages": []}) is False
def test_false_when_messages_key_missing(self):
mw = ViewImageMiddleware()
assert mw._should_inject_image_message({}) is False
def test_false_when_no_assistant_message(self):
mw = ViewImageMiddleware()
state = {"messages": [HumanMessage(content="hello")]}
assert mw._should_inject_image_message(state) is False
def test_false_when_no_view_image_tool_call(self):
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[_other_tool_call()])
state = {
"messages": [assistant, ToolMessage(content="ok", tool_call_id="call_other")],
}
assert mw._should_inject_image_message(state) is False
def test_false_when_tool_not_completed(self):
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[_view_image_call("c1")])
state = {"messages": [assistant]} # no ToolMessage yet
assert mw._should_inject_image_message(state) is False
def test_true_when_all_preconditions_met(self):
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[_view_image_call("c1")])
state = {
"messages": [assistant, ToolMessage(content="ok", tool_call_id="c1")],
"viewed_images": {
"/img.png": {"base64": "AAA", "mime_type": "image/png"},
},
}
assert mw._should_inject_image_message(state) is True
def test_false_when_already_injected(self):
"""If a HumanMessage with the recognized header is already present after
the assistant turn, we must not inject a duplicate."""
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[_view_image_call("c1")])
already_injected = HumanMessage(content="Here are the images you've viewed: /img.png")
state = {
"messages": [
assistant,
ToolMessage(content="ok", tool_call_id="c1"),
already_injected,
],
"viewed_images": {
"/img.png": {"base64": "AAA", "mime_type": "image/png"},
},
}
assert mw._should_inject_image_message(state) is False
def test_false_when_already_injected_with_list_content(self):
"""Deduplication must recognize the real injected payload shape.
The middleware's own `_inject_image_message` creates a HumanMessage
whose `.content` is a *list* of dicts (text + image_url blocks), not a
plain string. This test reuses `_create_image_details_message` output
to reproduce the realistic shape and confirms `_should_inject_image_message`
still detects the marker via `str(msg.content)`.
"""
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[_view_image_call("c1")])
viewed_images = {"/img.png": {"base64": "AAA", "mime_type": "image/png"}}
# Build content the same way the middleware would.
real_injected_content = mw._create_image_details_message({"viewed_images": viewed_images})
# Sanity: this is a list of blocks, not a plain string.
assert isinstance(real_injected_content, list)
already_injected = HumanMessage(content=real_injected_content)
state = {
"messages": [
assistant,
ToolMessage(content="ok", tool_call_id="c1"),
already_injected,
],
"viewed_images": viewed_images,
}
assert mw._should_inject_image_message(state) is False
def test_false_when_legacy_details_marker_present(self):
"""The middleware also recognizes the legacy 'Here are the details of the
images you've viewed' marker as an already-injected signal."""
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[_view_image_call("c1")])
legacy = HumanMessage(content="Here are the details of the images you've viewed: ...")
state = {
"messages": [
assistant,
ToolMessage(content="ok", tool_call_id="c1"),
legacy,
],
"viewed_images": {
"/img.png": {"base64": "AAA", "mime_type": "image/png"},
},
}
assert mw._should_inject_image_message(state) is False
class TestInjectImageMessage:
def test_returns_none_when_should_not_inject(self):
mw = ViewImageMiddleware()
state = {"messages": []}
assert mw._inject_image_message(state) is None
def test_returns_state_update_with_human_message(self):
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[_view_image_call("c1")])
state = {
"messages": [assistant, ToolMessage(content="ok", tool_call_id="c1")],
"viewed_images": {
"/img.png": {"base64": "AAA", "mime_type": "image/png"},
},
}
result = mw._inject_image_message(state)
assert isinstance(result, dict)
assert "messages" in result
assert len(result["messages"]) == 1
injected = result["messages"][0]
assert isinstance(injected, HumanMessage)
# Mixed-content payload: list of text + image_url blocks
assert isinstance(injected.content, list)
assert any(isinstance(b, dict) and b.get("type") == "image_url" for b in injected.content)
# Internal injection: must be hidden from the chat UI (and IM channels),
# like the other middleware-injected context messages.
assert injected.additional_kwargs.get("hide_from_ui") is True
class TestBeforeModel:
def test_before_model_returns_none_when_preconditions_not_met(self):
mw = ViewImageMiddleware()
state = {"messages": [HumanMessage(content="hi")]}
assert mw.before_model(state, _runtime()) is None
def test_before_model_returns_injection_when_ready(self):
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[_view_image_call("c1")])
state = {
"messages": [assistant, ToolMessage(content="ok", tool_call_id="c1")],
"viewed_images": {
"/img.png": {"base64": "AAA", "mime_type": "image/png"},
},
}
result = mw.before_model(state, _runtime())
assert result is not None
assert isinstance(result["messages"][0], HumanMessage)
@pytest.mark.anyio
async def test_abefore_model_matches_sync_behavior(self):
mw = ViewImageMiddleware()
assistant = AIMessage(content="", tool_calls=[_view_image_call("c1")])
state = {
"messages": [assistant, ToolMessage(content="ok", tool_call_id="c1")],
"viewed_images": {
"/img.png": {"base64": "AAA", "mime_type": "image/png"},
},
}
result = await mw.abefore_model(state, _runtime())
assert result is not None
assert isinstance(result["messages"][0], HumanMessage)
@pytest.mark.anyio
async def test_abefore_model_returns_none_when_no_injection(self):
mw = ViewImageMiddleware()
state = {"messages": []}
assert await mw.abefore_model(state, _runtime()) is None