feat: add explicit gateway task case hints for openclaw-gateway-e2e-regression
This commit is contained in:
Haitao Pan
parent
96fbea6d63
commit
ab0ecdd005
@ -16,11 +16,11 @@ legacy `sessionKey` compatibility field.
|
||||
|
||||
## 覆盖目标
|
||||
|
||||
- 连续出图：7 张连续风格 PNG。
|
||||
- 模板出图：参考附件模板生成 7 张连续 PNG。
|
||||
- PDF：拆章节、逐章生成图、汇总排版并输出 PDF。
|
||||
- 视频：围绕同一安全演进主线制作测试视频。
|
||||
- 视频流水线：拆章节、逐章调用 Codex/GPT Images、汇总排版并制作视频。
|
||||
- case1：采集最新 AI 资讯，输出 Markdown 文件。
|
||||
- case2：带图片附件制作视频，输出 MP4，并保留素材 manifest。
|
||||
- case3：围绕安全演进主线制作 7 张连续风格 PNG。
|
||||
- case4：围绕安全演进主线输出 5 份平台 Markdown 文案。
|
||||
- case5：安全演进拆 7 章，每章调用 Codex/GPT Images 生成图，汇总排版输出 PDF。
|
||||
|
||||
## 自动化落点
|
||||
|
||||
@ -35,6 +35,18 @@ legacy `sessionKey` compatibility field.
|
||||
|
||||
以下提示词按原始 E2E 输入记录，作为长期回归 case 的 canonical prompt。
|
||||
|
||||
### `OPENCLAW-E2E-000` AI 资讯 Markdown
|
||||
|
||||
```text
|
||||
采集最新AI资讯，保存在md文件
|
||||
```
|
||||
|
||||
期望结果：
|
||||
|
||||
- 输出 `reports/ai-news-digest.md` 和 `reports/sources.md`。
|
||||
- 资讯包含日期、来源链接和摘要。
|
||||
- 不把浏览器缓存、临时截图或 scratch JSON 当最终 artifact。
|
||||
|
||||
### `OPENCLAW-E2E-001` 连续出图
|
||||
|
||||
```text
|
||||
@ -65,15 +77,15 @@ legacy `sessionKey` compatibility field.
|
||||
### `OPENCLAW-E2E-003` PDF
|
||||
|
||||
```text
|
||||
围绕
|
||||
从单机权限 → 网络边界 → Web安全 → 云身份 → Zero Trust → AI Agent 身份 → AI模型与知识保护 演进
|
||||
拆章节 -> 每章调用 Codex -> 每章 GPT images2 生成图 -> 汇总排版 -> 输出 PDF
|
||||
|
||||
右侧 artifact栏 显示的陈旧文件
|
||||
```
|
||||
|
||||
期望结果：
|
||||
|
||||
- 每章图片素材和最终 PDF 归属当前 task scope。
|
||||
- PDF 或相关素材出现在当前任务 artifact 区。
|
||||
- 输出 `exports/final.pdf`，并且 `assets/images/` 下有 7 张真实 PNG。
|
||||
- 回归缺陷点：右侧 artifact 栏不能显示其他 run 或历史 workspace 的陈旧文件。
|
||||
- 如果 OpenClaw 没有实际导出文件，App 显示 no exported artifacts，而不是旧文件。
|
||||
|
||||
@ -91,19 +103,19 @@ legacy `sessionKey` compatibility field.
|
||||
- 输出视频帧、配置或 MP4 时，artifact 只属于当前任务。
|
||||
- 失败时释放 active slot 并继续 drain 后续任务。
|
||||
|
||||
### `OPENCLAW-E2E-005` 视频流水线
|
||||
### `OPENCLAW-E2E-005` PDF 流水线
|
||||
|
||||
```text
|
||||
围绕
|
||||
|
||||
从单机权限 → 网络边界 → Web安全 → 云身份 → Zero Trust → AI Agent 身份 → AI模型与知识保护 演进
|
||||
|
||||
拆章节 -> 每章调用 Codex -> 每章 GPT images2 生成图 -> 汇总排版 -> 制作视频
|
||||
拆章节 -> 每章调用 Codex -> 每章 GPT images2 生成图 -> 汇总排版 -> 输出 PDF
|
||||
```
|
||||
|
||||
期望结果：
|
||||
|
||||
- 图片、manifest、视频配置、MP4/ffprobe 等产物按当前 run 隔离。
|
||||
- 图片、manifest、`article.md`、`workflow.plan.md`、`exports/final.pdf` 等产物按当前 run 隔离。
|
||||
- Bridge 和 OpenClaw Gateway 只建立稳定连接，不重复并发握手。
|
||||
- 不出现 `invalid handshake: first request must be connect`、`SOCKET_CLOSED`、`ACP_HTTP_CONNECTION_CLOSED`。
|
||||
|
||||
|
||||
@ -482,6 +482,7 @@ extension AppControllerDesktopThreadActions on AppController {
|
||||
gatewayTaskMetadataWithArtifactContractInternal(
|
||||
baseMetadata: dispatch.metadata,
|
||||
sessionKey: normalizedSessionKey,
|
||||
userPrompt: message,
|
||||
localWorkingDirectory: workingDirectory,
|
||||
executionWorkingDirectory: executionWorkingDirectory,
|
||||
remoteWorkingDirectoryHint: remoteWorkingDirectoryHint,
|
||||
@ -1015,6 +1016,7 @@ extension AppControllerDesktopThreadActions on AppController {
|
||||
Map<String, dynamic> gatewayTaskMetadataWithArtifactContractInternal({
|
||||
required Map<String, dynamic> baseMetadata,
|
||||
required String sessionKey,
|
||||
required String userPrompt,
|
||||
required String localWorkingDirectory,
|
||||
required String executionWorkingDirectory,
|
||||
required String remoteWorkingDirectoryHint,
|
||||
@ -1022,8 +1024,25 @@ extension AppControllerDesktopThreadActions on AppController {
|
||||
final localWorkspace = localWorkingDirectory.trim();
|
||||
final executionWorkspace = executionWorkingDirectory.trim();
|
||||
final remoteHint = remoteWorkingDirectoryHint.trim();
|
||||
return <String, dynamic>{
|
||||
final caseHint = gatewayCaseHintForPromptInternal(userPrompt);
|
||||
final metadata = <String, dynamic>{
|
||||
...baseMetadata,
|
||||
if (caseHint.caseId.isNotEmpty) 'xworkmateCaseId': caseHint.caseId,
|
||||
if (caseHint.taskLoadClass.isNotEmpty)
|
||||
'taskLoadClass': caseHint.taskLoadClass,
|
||||
if (caseHint.requiredArtifactExtensions.isNotEmpty)
|
||||
'requiredArtifactExtensions': caseHint.requiredArtifactExtensions,
|
||||
if (caseHint.expectedArtifactExtensions.isNotEmpty)
|
||||
'expectedArtifactExtensions': caseHint.expectedArtifactExtensions,
|
||||
};
|
||||
if (caseHint.expectedFileCountByExtension.isNotEmpty) {
|
||||
metadata['xworkmateArtifactConstraints'] = <String, dynamic>{
|
||||
'schemaVersion': 1,
|
||||
'expectedFileCountByExtension': caseHint.expectedFileCountByExtension,
|
||||
};
|
||||
}
|
||||
return <String, dynamic>{
|
||||
...metadata,
|
||||
'xworkmateTaskArtifactContract': <String, dynamic>{
|
||||
'schemaVersion': 1,
|
||||
'appThreadKey': sessionKey,
|
||||
@ -1031,14 +1050,20 @@ extension AppControllerDesktopThreadActions on AppController {
|
||||
'finalDeliverableDetection': 'remote-runtime',
|
||||
'requiresExportBeforeFinalResponse': true,
|
||||
'rejectTextOnlyFileClaims': true,
|
||||
'expectedArtifactDirs': const <String>[
|
||||
'artifacts/',
|
||||
'reports/',
|
||||
'exports/',
|
||||
'assets/',
|
||||
'assets/images/',
|
||||
'dist/',
|
||||
],
|
||||
'expectedArtifactDirs': caseHint.expectedArtifactDirs.isNotEmpty
|
||||
? caseHint.expectedArtifactDirs
|
||||
: const <String>[
|
||||
'artifacts/',
|
||||
'reports/',
|
||||
'exports/',
|
||||
'assets/',
|
||||
'assets/images/',
|
||||
'dist/',
|
||||
],
|
||||
if (caseHint.requiredArtifactExtensions.isNotEmpty)
|
||||
'requiredArtifactExtensions': caseHint.requiredArtifactExtensions,
|
||||
if (caseHint.expectedFileCountByExtension.isNotEmpty)
|
||||
'expectedFileCountByExtension': caseHint.expectedFileCountByExtension,
|
||||
'currentTaskWorkspace': executionWorkspace.isNotEmpty
|
||||
? executionWorkspace
|
||||
: (remoteHint.isNotEmpty ? remoteHint : localWorkspace),
|
||||
@ -1048,6 +1073,115 @@ extension AppControllerDesktopThreadActions on AppController {
|
||||
};
|
||||
}
|
||||
|
||||
GatewayTaskCaseHintInternal gatewayCaseHintForPromptInternal(String prompt) {
|
||||
final text = prompt.trim();
|
||||
final lower = text.toLowerCase();
|
||||
final hasSecurityEvolution =
|
||||
text.contains('从单机权限') &&
|
||||
text.contains('网络边界') &&
|
||||
text.contains('Web安全') &&
|
||||
text.contains('云身份') &&
|
||||
text.contains('Zero Trust') &&
|
||||
text.contains('AI Agent') &&
|
||||
text.contains('AI模型');
|
||||
final wantsAiNewsMd =
|
||||
(text.contains('AI资讯') ||
|
||||
text.contains('AI 资讯') ||
|
||||
text.contains('AI新闻') ||
|
||||
lower.contains('ai news')) &&
|
||||
(text.contains('md文件') ||
|
||||
text.contains('Markdown') ||
|
||||
lower.contains('.md') ||
|
||||
lower.contains('markdown'));
|
||||
if (wantsAiNewsMd) {
|
||||
return const GatewayTaskCaseHintInternal(
|
||||
caseId: 'case1-ai-news-md',
|
||||
taskLoadClass: 'long_task',
|
||||
requiredArtifactExtensions: <String>['md'],
|
||||
expectedArtifactExtensions: <String>['md'],
|
||||
expectedArtifactDirs: <String>['reports/', 'artifacts/'],
|
||||
);
|
||||
}
|
||||
final wantsVideo =
|
||||
text.contains('制作视频') ||
|
||||
text.contains('测试制作视频') ||
|
||||
lower.contains('make video') ||
|
||||
lower.contains('mp4');
|
||||
final wantsImages =
|
||||
text.contains('7张') ||
|
||||
text.contains('七张') ||
|
||||
text.contains('连续制作') ||
|
||||
text.contains('系列图片') ||
|
||||
text.contains('一些列图片') ||
|
||||
text.contains('一系列图片');
|
||||
final wantsPdf =
|
||||
text.contains('输出 PDF') ||
|
||||
text.contains('输出PDF') ||
|
||||
text.contains('PDF文件') ||
|
||||
lower.contains('final.pdf');
|
||||
final wantsSocialCopy =
|
||||
text.contains('微信公众号短图文') ||
|
||||
text.contains('小红书风格') ||
|
||||
text.contains('X文案串') ||
|
||||
text.contains('头条号长文');
|
||||
final wantsChapteredImages =
|
||||
text.contains('拆章节') ||
|
||||
text.contains('每章') ||
|
||||
lower.contains('gpt images') ||
|
||||
lower.contains('images2');
|
||||
if (hasSecurityEvolution && wantsPdf && wantsChapteredImages) {
|
||||
return const GatewayTaskCaseHintInternal(
|
||||
caseId: 'case5-security-evolution-pdf',
|
||||
taskLoadClass: 'complex_chain_task',
|
||||
requiredArtifactExtensions: <String>['pdf', 'png', 'md'],
|
||||
expectedArtifactExtensions: <String>['pdf', 'png', 'md'],
|
||||
expectedArtifactDirs: <String>[
|
||||
'exports/',
|
||||
'assets/',
|
||||
'assets/images/',
|
||||
'prompts/',
|
||||
'reports/',
|
||||
],
|
||||
expectedFileCountByExtension: <String, int>{'pdf': 1, 'png': 7},
|
||||
);
|
||||
}
|
||||
if (hasSecurityEvolution && wantsVideo) {
|
||||
return const GatewayTaskCaseHintInternal(
|
||||
caseId: 'case2-security-evolution-video',
|
||||
taskLoadClass: 'complex_chain_task',
|
||||
requiredArtifactExtensions: <String>['mp4'],
|
||||
expectedArtifactExtensions: <String>['mp4', 'png', 'md'],
|
||||
expectedArtifactDirs: <String>[
|
||||
'renders/',
|
||||
'assets/',
|
||||
'assets/images/',
|
||||
'exports/',
|
||||
],
|
||||
);
|
||||
}
|
||||
if (hasSecurityEvolution && wantsImages) {
|
||||
return const GatewayTaskCaseHintInternal(
|
||||
caseId: 'case3-security-evolution-seven-images',
|
||||
taskLoadClass: 'complex_chain_task',
|
||||
requiredArtifactExtensions: <String>['png'],
|
||||
expectedArtifactExtensions: <String>['png', 'md'],
|
||||
expectedArtifactDirs: <String>['assets/', 'assets/images/', 'prompts/'],
|
||||
expectedFileCountByExtension: <String, int>{'png': 7},
|
||||
);
|
||||
}
|
||||
if (hasSecurityEvolution && wantsSocialCopy) {
|
||||
return const GatewayTaskCaseHintInternal(
|
||||
caseId: 'case4-security-evolution-social-copy',
|
||||
taskLoadClass: 'long_task',
|
||||
requiredArtifactExtensions: <String>['md'],
|
||||
expectedArtifactExtensions: <String>['md'],
|
||||
expectedArtifactDirs: <String>['reports/', 'artifacts/'],
|
||||
expectedFileCountByExtension: <String, int>{'md': 5},
|
||||
);
|
||||
}
|
||||
return const GatewayTaskCaseHintInternal();
|
||||
}
|
||||
|
||||
bool usesOpenClawGatewayQueueInternal(
|
||||
AssistantExecutionTarget target,
|
||||
SingleAgentProvider provider,
|
||||
@ -1725,3 +1859,21 @@ extension AppControllerDesktopThreadActions on AppController {
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
class GatewayTaskCaseHintInternal {
|
||||
const GatewayTaskCaseHintInternal({
|
||||
this.caseId = '',
|
||||
this.taskLoadClass = '',
|
||||
this.requiredArtifactExtensions = const <String>[],
|
||||
this.expectedArtifactExtensions = const <String>[],
|
||||
this.expectedArtifactDirs = const <String>[],
|
||||
this.expectedFileCountByExtension = const <String, int>{},
|
||||
});
|
||||
|
||||
final String caseId;
|
||||
final String taskLoadClass;
|
||||
final List<String> requiredArtifactExtensions;
|
||||
final List<String> expectedArtifactExtensions;
|
||||
final List<String> expectedArtifactDirs;
|
||||
final Map<String, int> expectedFileCountByExtension;
|
||||
}
|
||||
|
||||
@ -1358,7 +1358,7 @@ void main() {
|
||||
});
|
||||
|
||||
test(
|
||||
'sendChatMessage leaves Gateway task classification to the remote runtime',
|
||||
'sendChatMessage adds explicit case metadata for security video pipelines',
|
||||
() async {
|
||||
final fakeGoTaskService = _RecordingGoTaskServiceClient();
|
||||
final controller = _connectedGatewayController(fakeGoTaskService);
|
||||
@ -1376,8 +1376,17 @@ void main() {
|
||||
|
||||
expect(fakeGoTaskService.requests, hasLength(1));
|
||||
final request = fakeGoTaskService.requests.single;
|
||||
expect(request.metadata, isNot(contains('taskLoadClass')));
|
||||
expect(request.metadata, isNot(contains('expectedArtifactExtensions')));
|
||||
expect(
|
||||
request.metadata['xworkmateCaseId'],
|
||||
'case2-security-evolution-video',
|
||||
);
|
||||
expect(request.metadata['taskLoadClass'], 'complex_chain_task');
|
||||
expect(request.metadata['requiredArtifactExtensions'], <String>['mp4']);
|
||||
expect(request.metadata['expectedArtifactExtensions'], <String>[
|
||||
'mp4',
|
||||
'png',
|
||||
'md',
|
||||
]);
|
||||
expect(request.metadata, contains('xworkmateTaskArtifactContract'));
|
||||
final artifactContract =
|
||||
(request.metadata['xworkmateTaskArtifactContract'] as Map)
|
||||
@ -1388,14 +1397,12 @@ void main() {
|
||||
expect(artifactContract['finalDeliverableDetection'], 'remote-runtime');
|
||||
expect(artifactContract['requiresExportBeforeFinalResponse'], isTrue);
|
||||
expect(artifactContract['expectedArtifactDirs'], const <String>[
|
||||
'artifacts/',
|
||||
'reports/',
|
||||
'exports/',
|
||||
'renders/',
|
||||
'assets/',
|
||||
'assets/images/',
|
||||
'dist/',
|
||||
'exports/',
|
||||
]);
|
||||
expect(artifactContract, isNot(contains('expectedArtifactExtensions')));
|
||||
expect(artifactContract['requiredArtifactExtensions'], <String>['mp4']);
|
||||
expect(request.prompt, isNot(contains('Task load classification:')));
|
||||
expect(
|
||||
request.prompt,
|
||||
@ -1413,7 +1420,7 @@ void main() {
|
||||
);
|
||||
|
||||
test(
|
||||
'sendChatMessage leaves artifact expectations to the remote runtime',
|
||||
'sendChatMessage adds explicit case metadata for chaptered PDF pipelines',
|
||||
() async {
|
||||
final fakeGoTaskService = _RecordingGoTaskServiceClient();
|
||||
final controller = _connectedGatewayController(fakeGoTaskService);
|
||||
@ -1432,8 +1439,16 @@ void main() {
|
||||
|
||||
expect(fakeGoTaskService.requests, hasLength(1));
|
||||
final request = fakeGoTaskService.requests.single;
|
||||
expect(request.metadata, isNot(contains('taskLoadClass')));
|
||||
expect(request.metadata, isNot(contains('expectedArtifactExtensions')));
|
||||
expect(
|
||||
request.metadata['xworkmateCaseId'],
|
||||
'case5-security-evolution-pdf',
|
||||
);
|
||||
expect(request.metadata['taskLoadClass'], 'complex_chain_task');
|
||||
expect(request.metadata['requiredArtifactExtensions'], <String>[
|
||||
'pdf',
|
||||
'png',
|
||||
'md',
|
||||
]);
|
||||
expect(request.metadata, contains('xworkmateTaskArtifactContract'));
|
||||
final artifactContract =
|
||||
(request.metadata['xworkmateTaskArtifactContract'] as Map)
|
||||
@ -1444,13 +1459,16 @@ void main() {
|
||||
expect(artifactContract['scopeKind'], 'task');
|
||||
expect(artifactContract['rejectTextOnlyFileClaims'], isTrue);
|
||||
expect(artifactContract['expectedArtifactDirs'], const <String>[
|
||||
'artifacts/',
|
||||
'reports/',
|
||||
'exports/',
|
||||
'assets/',
|
||||
'assets/images/',
|
||||
'dist/',
|
||||
'prompts/',
|
||||
'reports/',
|
||||
]);
|
||||
expect(artifactContract['expectedFileCountByExtension'], <String, int>{
|
||||
'pdf': 1,
|
||||
'png': 7,
|
||||
});
|
||||
expect(
|
||||
artifactContract['currentTaskWorkspace'],
|
||||
request.workingDirectory,
|
||||
|
||||
Loading…
Reference in New Issue
Block a user