feat: add configurable screenshot resolution limit

.env.example

@@ -19,6 +19,8 @@ ARTEMIS_VIDEO_ENABLED=false
 ARTEMIS_RECORDING_ENABLED=false
 ARTEMIS_OUTPUTTER_ENABLED=true
 ARTEMIS_STEP_SUMMARIZER_ENABLED=true
+# 默认将任务截图长边压至 2048；false 保留设备原始截图。长边值可在 config/artemis.jsonc 的 agent.screenshot 中覆盖。
+ARTEMIS_SCREENSHOT_LIMIT_ENABLED=true
 # Pro 连续无工具调用或参数校验失败的最大次数；耗尽后明确失败，避免空转。
 ARTEMIS_OPERATOR_MAX_NO_PROGRESS_ATTEMPTS=3
 # Flash 连续无工具调用的最大次数；无效回复不会重复送回模型。

README.md

@@ -52,6 +52,8 @@ cd Artemis-Eapil
 
 也可在项目 `.env` 中设置 `OPENAI_API_KEY`、`OPENAI_BASE_URL` 和 `OPENAI_MODEL`，其他参数见 [.env.example](.env.example)。新安装默认接口地址为 `https://api-n-cd.playones.com/v1`；已保存的地址不变。直接编辑环境文件后需重启相应服务。不要提交自己的密钥。
 
+“功能开关”默认将任务截图长边限制为 2048 像素；关闭后保留设备原始分辨率。可在 [config/artemis.jsonc](config/artemis.jsonc) 的 `agent.screenshot` 中覆盖 `enabled` 和 `max_long_edge`，重启后对新任务生效。
+
 报告会区分“执行完成”和“验证通过”；没有验证证据的结果标为“未验证”。任务步骤与截图可查看，设备动作回放不提供。
 
 ## 让 AI 使用

apps/admin_console/services/model_settings_service.py

@@ -9,6 +9,7 @@ from urllib.parse import urlsplit
 
 from pydantic import BaseModel, Field, SecretStr, field_validator
 
+from artemis.config.agent import load_agent_config
 from artemis.config.paths import get_env_file
 from artemis.config.settings import settings
 
@@ -137,14 +138,19 @@ class CapabilitySettings(BaseModel):
     recording_enabled: bool = False
     outputter_enabled: bool = True
     step_summarizer_enabled: bool = True
+    screenshot_limit_enabled: bool = True
+    screenshot_max_long_edge: int = 2048
 
 
 def get_capability_settings() -> CapabilitySettings:
+    screenshot = load_agent_config().screenshot
     return CapabilitySettings(
         video_enabled=settings.ARTEMIS_VIDEO_ENABLED,
         recording_enabled=settings.ARTEMIS_RECORDING_ENABLED,
         outputter_enabled=settings.ARTEMIS_OUTPUTTER_ENABLED,
         step_summarizer_enabled=settings.ARTEMIS_STEP_SUMMARIZER_ENABLED,
+        screenshot_limit_enabled=screenshot.enabled,
+        screenshot_max_long_edge=screenshot.max_long_edge,
     )
 
 
@@ -154,6 +160,7 @@ def save_capability_settings(update: CapabilitySettings) -> CapabilitySettings:
         "ARTEMIS_RECORDING_ENABLED": update.recording_enabled,
         "ARTEMIS_OUTPUTTER_ENABLED": update.outputter_enabled,
         "ARTEMIS_STEP_SUMMARIZER_ENABLED": update.step_summarizer_enabled,
+        "ARTEMIS_SCREENSHOT_LIMIT_ENABLED": update.screenshot_limit_enabled,
     }
     persist_environment({key: str(value).lower() for key, value in values.items()})
     for key, value in values.items():

apps/showcase_ui/src/app/components/model-settings/capability-settings.component.ts

@@ -8,6 +8,8 @@ interface Capabilities {
   recording_enabled: boolean;
   outputter_enabled: boolean;
   step_summarizer_enabled: boolean;
+  screenshot_limit_enabled: boolean;
+  screenshot_max_long_edge: number;
 }
 
 @Component({
@@ -24,6 +26,8 @@ interface Capabilities {
         <span>最终报告生成<small>根据执行记录生成任务结果报告。</small></span></label>
       <label><input type="checkbox" role="switch" [(ngModel)]="config.step_summarizer_enabled" [disabled]="busy() || loading() || !loaded()" />
         <span>步骤摘要压缩<small>Flash 模式下在后台生成摘要，减少长任务的上下文占用。</small></span></label>
+      <label><input type="checkbox" role="switch" [(ngModel)]="config.screenshot_limit_enabled" [disabled]="busy() || loading() || !loaded()" />
+        <span>限制截图分辨率<small>开启时长边最多 {{ config.screenshot_max_long_edge }} 像素；关闭后使用设备原始分辨率。JSON 配置可覆盖开关和上限。</small></span></label>
       <button type="button" (click)="save()" [disabled]="busy() || loading() || !loaded()">{{ busy() ? '正在保存…' : '保存功能设置' }}</button>
       @if (message()) { <p role="status">{{ message() }}</p> }
       @if (error()) { <p role="alert" class="error">{{ error() }}</p> }
@@ -39,7 +43,7 @@ interface Capabilities {
 export class CapabilitySettingsComponent implements OnInit {
   changed = output<Capabilities>();
   private http = inject(HttpClient);
-  config: Capabilities = { video_enabled: false, recording_enabled: false, outputter_enabled: true, step_summarizer_enabled: true };
+  config: Capabilities = { video_enabled: false, recording_enabled: false, outputter_enabled: true, step_summarizer_enabled: true, screenshot_limit_enabled: true, screenshot_max_long_edge: 2048 };
   busy = signal(false); loading = signal(true); loaded = signal(false); message = signal(''); error = signal('');
   ngOnInit(): void {
     this.http.get<Capabilities>('/api/system/capability-settings').pipe(finalize(() => this.loading.set(false))).subscribe({

apps/showcase_ui/src/app/components/model-settings/model-settings.component.spec.ts

@@ -110,17 +110,19 @@ describe('持久化模型与功能设置', () => {
   });
 
   it('录屏与视频分析独立保存，重新打开后保留各自状态', () => {
-    const capabilities = { video_enabled: true, recording_enabled: false, outputter_enabled: true, step_summarizer_enabled: false };
+    const capabilities = { video_enabled: true, recording_enabled: false, outputter_enabled: true, step_summarizer_enabled: false, screenshot_limit_enabled: true, screenshot_max_long_edge: 2048 };
     const fixture = TestBed.createComponent(CapabilitySettingsComponent);
     fixture.detectChanges();
     http.expectOne('/api/system/capability-settings').flush(capabilities);
     fixture.componentInstance.config.video_enabled = false;
     fixture.componentInstance.config.recording_enabled = true;
+    fixture.componentInstance.config.screenshot_limit_enabled = false;
     fixture.componentInstance.save();
     const request = http.expectOne('/api/system/capability-settings');
     expect(request.request.body.video_enabled).toBeFalse();
     expect(request.request.body.recording_enabled).toBeTrue();
-    const saved = { ...capabilities, video_enabled: false, recording_enabled: true };
+    expect(request.request.body.screenshot_limit_enabled).toBeFalse();
+    const saved = { ...capabilities, video_enabled: false, recording_enabled: true, screenshot_limit_enabled: false };
     request.flush(saved);
     fixture.destroy();
     const reopened = TestBed.createComponent(CapabilitySettingsComponent);
@@ -128,5 +130,6 @@ describe('持久化模型与功能设置', () => {
     http.expectOne('/api/system/capability-settings').flush(saved);
     expect(reopened.componentInstance.config.video_enabled).toBeFalse();
     expect(reopened.componentInstance.config.recording_enabled).toBeTrue();
+    expect(reopened.componentInstance.config.screenshot_limit_enabled).toBeFalse();
   });
 });

artemis/config/agent.py

@@ -835,6 +835,13 @@ class ProProfileConfig(BaseModel):
     model_config = {"extra": "allow"}
 
 
+class ScreenshotConfig(BaseModel):
+    """Limit screenshots sent to agents while retaining physical device coordinates."""
+
+    enabled: bool = Field(default_factory=lambda: settings.ARTEMIS_SCREENSHOT_LIMIT_ENABLED)
+    max_long_edge: int = Field(default=2048, ge=256, le=8192)
+
+
 class AgentGlobalConfig(BaseModel):
     """Global configuration parsed from artemis.jsonc / agent_config.json."""
 
@@ -889,6 +896,10 @@ class AgentGlobalConfig(BaseModel):
         default_factory=MemoryConfig,
         description="Unified step-memory runtime and transcript scrub options.",
     )
+    screenshot: ScreenshotConfig = Field(
+        default_factory=ScreenshotConfig,
+        description="Agent screenshot size limit; device/UI coordinates remain native.",
+    )
 
     model_config = {"extra": "allow"}
 

artemis/config/settings.py

@@ -112,6 +112,7 @@ class Settings(BaseSettings):
     ARTEMIS_RECORDING_ENABLED: bool = False
     ARTEMIS_OUTPUTTER_ENABLED: bool = True
     ARTEMIS_STEP_SUMMARIZER_ENABLED: bool = True
+    ARTEMIS_SCREENSHOT_LIMIT_ENABLED: bool = True
     # Consecutive Pro responses without a valid tool call before failing the task.
     ARTEMIS_OPERATOR_MAX_NO_PROGRESS_ATTEMPTS: int = Field(default=3, ge=1, le=20)
     # Consecutive Flash responses without any tool call before failing the task.

artemis/drivers/android/adb_driver.py

@@ -25,12 +25,14 @@ from artemis.clients.ui_automator_client import (
     UIAutomatorClient,
     _parse_hierarchy_xml_to_elements,
 )
+from artemis.config.agent import load_agent_config
 from artemis.config.paths import get_temp_dir
 from artemis.drivers.base import BaseDeviceDriver, KeyCode, ScreenData, SwipeDirection
 from artemis.toolchain import find_ffmpeg, find_scrcpy
 from artemis.utils.video import build_scrcpy_record_command
 from artemis.utils.ui_filter import filter_ui_hierarchy
 from artemis.utils.logger import get_logger
+from artemis.utils.screenshot_limit import limit_screenshot_long_edge
 
 logger = get_logger(__name__)
 
@@ -88,6 +90,7 @@ class AndroidAdbDriver(BaseDeviceDriver):
         self._device: AdbDevice | None = None
         self._recording_process: asyncio.subprocess.Process | None = None
         self._recording_output_path: Path | None = None
+        self._screenshot_config = load_agent_config().screenshot
 
     @property
     def device_id(self) -> str:
@@ -215,6 +218,13 @@ class AndroidAdbDriver(BaseDeviceDriver):
             if isinstance(screenshot_bytes, bytes) and screenshot_bytes
             else base64.b64decode(clean_b64)
         )
+        if self._screenshot_config.enabled:
+            limited_bytes = limit_screenshot_long_edge(
+                clean_bytes, self._screenshot_config.max_long_edge
+            )
+            if limited_bytes is not clean_bytes:
+                clean_bytes = limited_bytes
+                clean_b64 = base64.b64encode(clean_bytes).decode("utf-8")
 
         return ScreenData(
             screenshot_bytes=clean_bytes,

artemis/resources/config/artemis.jsonc

@@ -27,6 +27,10 @@
     "checker": {}
   },
   "agent": {
+    "screenshot": {
+      "enabled": "${env:ARTEMIS_SCREENSHOT_LIMIT_ENABLED:-true}",
+      "max_long_edge": 2048
+    },
     "flash": {
       "max_turns": 0,
       "explorer_mode": "flash",

artemis/utils/screenshot_limit.py

@@ -0,0 +1,23 @@
+"""Prepare Android screenshots for model input without changing device coordinates."""
+
+from io import BytesIO
+
+from PIL import Image
+
+
+def limit_screenshot_long_edge(image_bytes: bytes, max_long_edge: int) -> bytes:
+    """Downscale oversized screenshots; keep smaller images byte-for-byte unchanged."""
+    try:
+        with Image.open(BytesIO(image_bytes)) as image:
+            long_edge = max(image.size)
+            if long_edge <= max_long_edge:
+                return image_bytes
+            scale = max_long_edge / long_edge
+            size = tuple(max(1, round(side * scale)) for side in image.size)
+            resized = image.convert("RGB").resize(size, Image.Resampling.LANCZOS)
+            output = BytesIO()
+            resized.save(output, format="JPEG", quality=80, optimize=True)
+            return output.getvalue()
+    except OSError:
+        # Preserve the capture path's previous behavior for invalid placeholder images.
+        return image_bytes

config/artemis.jsonc

@@ -27,6 +27,10 @@
     "checker": {}
   },
   "agent": {
+    "screenshot": {
+      "enabled": "${env:ARTEMIS_SCREENSHOT_LIMIT_ENABLED:-true}",
+      "max_long_edge": 2048
+    },
     "flash": {
       "max_turns": 0,
       "explorer_mode": "flash",

tests/unit/admin_console/test_model_settings.py

@@ -27,6 +27,7 @@ def isolated_settings(monkeypatch, tmp_path):
         "ARTEMIS_VIDEO_ENABLED",
         "ARTEMIS_OUTPUTTER_ENABLED",
         "ARTEMIS_STEP_SUMMARIZER_ENABLED",
+        "ARTEMIS_SCREENSHOT_LIMIT_ENABLED",
     ]
     for key in keys:
         monkeypatch.setattr(settings, key, getattr(settings, key))
@@ -93,21 +94,38 @@ def test_rejects_invalid_base_url(url):
 def test_capabilities_persist_and_apply_to_new_tasks(isolated_settings):
     result = service.save_capability_settings(
         service.CapabilitySettings(
-            video_enabled=False, outputter_enabled=False, step_summarizer_enabled=False
+            video_enabled=False,
+            outputter_enabled=False,
+            step_summarizer_enabled=False,
+            screenshot_limit_enabled=False,
         )
     )
     assert result.video_enabled is False
     saved = dotenv_values(isolated_settings)
     assert saved["ARTEMIS_VIDEO_ENABLED"] == "false"
+    assert saved["ARTEMIS_SCREENSHOT_LIMIT_ENABLED"] == "false"
     config = load_agent_config()
     assert config.video_analyzer.enabled is False
     assert config.pro.video_analyzer.enabled is False
     assert config.outputter.enabled is False
     assert config.flash.step_summarizer.enabled is False
+    assert config.screenshot.enabled is False
     service.save_capability_settings(service.CapabilitySettings(video_enabled=True))
     assert load_agent_config().video_analyzer.enabled is True
 
 
+def test_screenshot_json_can_override_environment(monkeypatch, tmp_path):
+    config_path = tmp_path / "artemis.jsonc"
+    config_path.write_text(
+        '{"agent": {"screenshot": {"enabled": true, "max_long_edge": 1024}}}',
+        encoding="utf-8",
+    )
+    monkeypatch.setenv("ARTEMIS_SCREENSHOT_LIMIT_ENABLED", "false")
+    screenshot = load_agent_config(config_path).screenshot
+    assert screenshot.enabled is True
+    assert screenshot.max_long_edge == 1024
+
+
 def test_failed_write_keeps_running_settings(monkeypatch, isolated_settings):
     previous = settings.OPENAI_MODEL
     monkeypatch.setattr(

tests/unit/test_drivers.py

@@ -15,14 +15,18 @@
 """Unit tests for Unified Base Device Driver and implementations."""
 
 import base64
+from io import BytesIO
 from types import SimpleNamespace
 from unittest.mock import AsyncMock, MagicMock
 
 import pytest
+from PIL import Image
 
 from artemis.drivers.base import KeyCode, ScreenData, SwipeDirection
 from artemis.drivers.mock.mock_driver import MockDeviceDriver
 from artemis.drivers.android.adb_driver import AndroidAdbDriver
+from artemis.config.agent import ScreenshotConfig
+from artemis.utils.screenshot_limit import limit_screenshot_long_edge
 
 
 _ONE_PIXEL_PNG = base64.b64encode(
@@ -32,6 +36,13 @@ _ONE_PIXEL_PNG = base64.b64encode(
 ).decode("ascii")
 
 
+def test_screenshot_limit_keeps_small_and_invalid_images_unchanged():
+    small_image = base64.b64decode(_ONE_PIXEL_PNG)
+    assert limit_screenshot_long_edge(small_image, 2048) is small_image
+    invalid_image = b"not an image"
+    assert limit_screenshot_long_edge(invalid_image, 2048) is invalid_image
+
+
 @pytest.mark.asyncio
 async def test_mock_driver_actions():
     """Verify MockDeviceDriver records and performs all core actions."""
@@ -141,6 +152,36 @@ async def test_android_driver_preserves_elements_from_combined_screen_data():
     mock_ui_client.get_hierarchy.assert_not_called()
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("enabled", [True, False])
+async def test_android_screenshot_limit_preserves_native_coordinates(enabled):
+    image = Image.new("RGB", (1200, 2600), "white")
+    buffer = BytesIO()
+    image.save(buffer, format="JPEG")
+    original = buffer.getvalue()
+    elements = [{"text": "目标", "bounds": "[100,200][300,400]"}]
+    ui_client = MagicMock()
+    ui_client.get_screen_data.return_value = SimpleNamespace(
+        base64=base64.b64encode(original).decode("ascii"),
+        hierarchy_xml="<hierarchy />",
+        elements=elements,
+        width=1200,
+        height=2600,
+    )
+    driver = AndroidAdbDriver("emulator-5554", MagicMock(), ui_adb_client=ui_client)
+    driver._screenshot_config = ScreenshotConfig(enabled=enabled, max_long_edge=2048)
+
+    screen = await driver.get_screen_data(skip_settling=True)
+
+    with Image.open(BytesIO(screen.screenshot_bytes)) as result:
+        assert result.size == ((945, 2048) if enabled else (1200, 2600))
+    assert screen.width == 1200 and screen.height == 2600
+    assert screen.ui_elements == elements
+    assert screen.screenshot_base64 == base64.b64encode(screen.screenshot_bytes).decode("ascii")
+    if not enabled:
+        assert screen.screenshot_bytes == original
+
+
 @pytest.mark.asyncio
 async def test_android_driver_parses_xml_hierarchy_fallback():
     """XML-only clients remain supported without discarding their hierarchy."""