Enhance A2A protocol support and fix various bugs by Dallas98 · Pull Request #2822 · ModelEngine-Group/nexent

Dallas98 · 2026-04-18T03:00:29Z

No description provided.

…2776) * 🐛 Bugfix: skill deletion failed occasionally * ♻️ Simplify how run_skill_script() receives additional parameters * 🧪 Add test files * 🧪 Fix test files * ♻️ Update system prompt to better support models with lower parameters * ♻️ Update system prompt to better support models with lower parameters * ♻️ Remove unnecessary rely on re to reduce the risk of DoS * ♻️ Remove unnecessary rely on re to reduce the risk of DoS

…ent discovery (#2773) * Feat: Add A2A protocol support for agent publishing and external agent discovery - A2A Server: Enable publishing platform agents as A2A agents with configurable endpoints, including database schema for agent registration and management - A2A Client: Support discovering and invoking external A2A agents through agent discovery modal, HTTP client, and agent adapter integration - Database: Add `a2a_agent` and `a2a_agent_endpoint` tables with foreign key relationships to existing agent tables - SDK: Implement `A2AAgentProxy` for SDK-level A2A agent invocation support - Frontend: Add A2A Server Settings Panel, Agent Discovery Modal, and corresponding hooks (`useA2AServerAgents`, `useExternalAgents`) - Backend API: Add `a2a_client_app.py` and `a2a_server_app.py` endpoints with service layer implementations - i18n: Update English and Chinese localization files with A2A-related translations - System Prompts: Update manager system prompt templates to support A2A protocol context * Potential fix for pull request finding 'CodeQL / Information exposure through an exception' Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com> * Potential fix for pull request finding 'CodeQL / Information exposure through an exception' Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com> * Bugfix: temporarily not support grpc for external 2a2 agent * Bugfix: 解决在 AgentConfig 类中（第36-52行）引用了 ExternalA2AAgentConfig，但后者在第100行才定义。这构成了循环引用/前向引用 * delete debug logger * delete unsafe log * Bugfix: 使用 A2A 1.0 封套格式检测流式终止状态 * Bugfix: 删除对a2a 0.3版本对kind字段的兼容 * Bugfix: 修复了只读模式下关联代理信息完全隐藏的问题 * Bugfix: 当 supported_interfaces 同时包含两种协议接口且 jsonrpcInterface 有自定义 url 时，之前会错误地使用 REST 的 url，现在会正确使用 JSON-RPC 的 url * Bugfix: 当触发 aiohttp.ClientError 或通用 Exception 时，代码会抛出 NameError: name 'agent_id' is not defined，根本走不到 return 语句 * Bugfix: 确保了 HTTP 连接的正确释放，避免在高负载场景下出现连接泄漏问题 * Bugfix: 修复数据库允许 task_id 为 NULL，但 ORM 层强制要求非空的问题 * update sql, make sure sql is consistent with orm modal * delete unused code * 把详情日志降级为 debug 级别，只在 INFO 保留计数 * Refactor this function to reduce its Cognitive Complexity from 30 to the 15 allowed. * Use "Annotated" type hints for FastAPI dependency injection * fix code analysis * Replace the unused local variable "user_id" with "_" * Bugfix: 修复代码圈复杂度&定义常量避免重复 * 修复圈复杂度 * 修复sonarlcoud问题 * 修复报错 * 修复循环依赖 * 修复单元测试问题 * 修复单元测试 * 新增northbound单元测试 * 修复代码报错 * 优化前端展示数据样式 * 修改报错 * 新增单元测试 * 新增测试用例 * 新增单元测试 * 新增单元测试 * 新增单元测试 * 新增测试用例 * 删除无法识别的UT * 新增单元测试 * 现在 MinioClient 在导入时不会立即连接 MinIO，只在实际调用方法时才会初始化连接。 * 修改 MinioClient，在懒加载时同时设置 storage_config 属性 --------- Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>

…n the resource management page. #2777 (#2782) [Specification Detail] 1. When adding a model, the connectivity check should call the /temporary_healthcheck interface. 2. The `/manage/create` interface should pass the `model_factory` parameter.

Copilot

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

This PR expands A2A protocol support end-to-end (backend A2A client/server, SDK external agent wrappers, and frontend management UI), while also addressing several bugs and prompt-format updates.

Changes:

Add A2A server endpoints (northbound), A2A client management APIs, and DB models for external agents/relations/tasks.
Extend SDK agent configuration to support external A2A sub-agents and update code-tag parsing/rendering.
Update frontend to manage A2A discovery, publishing-as-A2A, and display A2A server settings; plus minor bug fixes/tests.

Reviewed changes

Copilot reviewed 72 out of 85 changed files in this pull request and generated 15 comments.

Show a summary per file

File	Description
test/backend/database/test_skill_db.py	Adds coverage for deleting already-deleted skills.
test/backend/database/test_client.py	Updates MinioClient tests for lazy init; resets singleton flags.
test/backend/consts/test_a2a_models.py	Adds unit tests for new A2A Pydantic models.
test/backend/app/test_agent_app.py	Updates publish-version test to include `publish_as_a2a`.
test/backend/agents/test_create_agent_info.py	Adds tests for external A2A agent config extraction/building.
sdk/nexent/skills/skill_manager.py	Changes script params to raw CLI string; uses `shlex.split`.
sdk/nexent/memory/init.py	Exposes memory service APIs via package init.
sdk/nexent/core/tools/run_skill_script_tool.py	Aligns tool signature with raw CLI params string.
sdk/nexent/core/agents/nexent_agent.py	Wraps external A2A agents as managed agents via proxy wrapper.
sdk/nexent/core/agents/core_agent.py	Switches execution/display code parsing to `<code>` / `<DISPLAY:...>`.
sdk/nexent/core/agents/agent_model.py	Adds external agent config model + protocol constants; extends AgentConfig.
frontend/types/agentConfig.ts	Adds `is_a2a_server` flag to agent type.
frontend/stores/agentConfigStore.ts	Stores external sub-agent IDs and exposes updater.
frontend/services/modelService.ts	Adds `model_factory` support to model create/update payloads.
frontend/services/api.ts	Adds API endpoint definitions for A2A client/server management.
frontend/services/agentVersionService.ts	Extends publish response to include A2A agent/card info.
frontend/services/agentConfigService.ts	Maps backend `is_a2a_server` to frontend model.
frontend/public/locales/zh/common.json	Adds zh translations for A2A discovery/settings/publish and related errors.
frontend/public/locales/en/common.json	Adds en translations for A2A discovery/settings/publish and related errors.
frontend/hooks/agent/useExternalAgents.ts	Adds hook to query available external A2A agents.
frontend/hooks/agent/useA2AServerAgents.ts	Adds hooks/util to query A2A server agents & check agent status.
frontend/components/ui/markdownRenderer.tsx	Converts custom code tags to markdown fences before rendering.
frontend/app/[locale]/tenant-resources/components/resources/ModelList.tsx	Switches connectivity check to `verifyCustomModel`.
frontend/app/[locale]/models/components/model/ModelEditDialog.tsx	Removes tenantId-specific connectivity branch; uses unified verification.
frontend/app/[locale]/models/components/model/ModelAddDialog.tsx	Removes tenantId-specific connectivity branch; uses unified verification.
frontend/app/[locale]/chat/streaming/taskWindow.tsx	Adds streaming-safe conversion/extraction for new custom code tags.
frontend/app/[locale]/chat/streaming/chatStreamFinalMessage.tsx	Converts custom code tags to markdown fences for final message.
frontend/app/[locale]/agents/versions/AgentVersionPubulishModal.tsx	Adds “Publish as A2A Agent” switch + post-publish A2A preview modal.
frontend/app/[locale]/agents/components/agentManage/AgentList.tsx	Adds A2A settings modal entry point for A2A server agents.
frontend/app/[locale]/agents/components/agentConfig/CollaborativeAgent.tsx	Adds selection/removal of external A2A sub-agents alongside internal ones.
frontend/app/[locale]/agents/components/a2a/A2AServerSettingsPanel.tsx	New UI panel showing A2A endpoint/card URLs & usage hints.
frontend/app/[locale]/agents/components/AgentConfigComp.tsx	Adds entry point to A2A discovery modal and external agents invalidation.
docker/docker-compose.prod.yml	Exposes northbound port 5013 for external A2A access.
docker/.env.example	Documents `NORTHBOUND_EXTERNAL_URL` for public agent-card URLs.
backend/utils/a2a_http_client.py	New aiohttp-based A2A HTTP client with retries + SSE streaming.
backend/services/vectordatabase_service.py	Uses timezone-aware UTC timestamps (timezone.utc).
backend/services/agent_version_service.py	Adds `publish_as_a2a` option and returns A2A card preview data.
backend/services/agent_service.py	Adds `is_a2a_server` enrichment in agent list.
backend/services/a2a_agent_adapter.py	New adapter mapping internal agent IO to A2A protocol structures/events.
backend/prompts/utils/prompt_generate_zh.yaml	Updates execution/display code format to `<code>` / `<DISPLAY>`.
backend/prompts/utils/prompt_generate_en.yaml	Updates execution/display code format to `<code>` / `<DISPLAY>`.
backend/prompts/skill_creation_simple_en.yaml	Adds a new prompt template for simple skill creation.
backend/prompts/manager_system_prompt_template_zh.yaml	Updates tool/script calling instructions + adds external agent guidance.
backend/prompts/manager_system_prompt_template_en.yaml	Updates code formatting + adds external agent guidance.
backend/prompts/managed_system_prompt_template_zh.yaml	Updates code formatting + external agent guidance; trims prompt content.
backend/prompts/managed_system_prompt_template_en.yaml	Updates code formatting + external agent guidance.
backend/database/skill_db.py	Prevents deleting already-deleted skills by filtering `delete_flag != 'Y'`.
backend/database/db_models.py	Adds DB models for A2A configs, external agents, relations, server agents, tasks/messages/artifacts.
backend/database/client.py	Makes MinioClient lazily initialize storage client on first use.
backend/consts/model.py	Adds `model_factory` to model requests; adds `publish_as_a2a` to publish request.
backend/consts/exceptions.py	Adds A2A-specific exceptions (TaskNotFound/UnsupportedOperation).
backend/consts/const.py	Adds `NORTHBOUND_EXTERNAL_URL` and bumps app version.
backend/consts/a2a_models.py	Introduces Pydantic models for A2A protocol & management APIs.
backend/apps/remote_mcp_app.py	Removes unused `user_id` binding in remote MCP tool list endpoint.
backend/apps/northbound_base_app.py	New northbound base app wiring A2A discovery + JSON-RPC/REST endpoints.
backend/apps/northbound_app.py	Exports `_get_northbound_context` for reuse by A2A routes.
backend/apps/config_app.py	Registers A2A client/server routers into the main FastAPI app.
backend/apps/agent_app.py	Passes `publish_as_a2a` into publish service.
backend/apps/a2a_server_app.py	Adds internal management endpoints for A2A server registrations/settings.
backend/apps/a2a_client_app.py	Adds endpoints for external agent discovery, management, relations, and Nacos configs.
backend/agents/create_agent_info.py	Adds external A2A sub-agent loading into agent config and templates.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-18T03:03:52Z

+    ) -> aiohttp.ClientResponse:
+        """Execute HTTP request with automatic retry on transient failures.
+
+        Args:
+            method: HTTP method
+            url: Target URL
+            read_response: If True, return (status, body_text). If False, return response object.
+            **kwargs: Additional arguments for the request
+        """
+        last_exception = None
+
+        for attempt in range(self.max_retries):
+            try:
+                async with await self._session.request(method, url, **kwargs) as response:
+                    if response.status < 500 and not read_response:
+                        return response
+                    body = await response.read()


aiohttp.ClientSession.request(...) should not be awaited before async with (use async with self._session.request(...) as response:). Also, returning response from inside an async with will return a closed response object after context exit, and the function annotation claims aiohttp.ClientResponse but returns tuples in other branches—please refactor to return a consistent type (e.g., always (status, body_bytes)), or avoid async with when you intend to return the live response.

Suggested change

) -> aiohttp.ClientResponse:

"""Execute HTTP request with automatic retry on transient failures.

Args:

method: HTTP method

url: Target URL

read_response: If True, return (status, body_text). If False, return response object.

**kwargs: Additional arguments for the request

"""

last_exception = None

for attempt in range(self.max_retries):

try:

async with await self._session.request(method, url, **kwargs) as response:

if response.status < 500 and not read_response:

return response

body = await response.read()

) -> aiohttp.ClientResponse | tuple[int, bytes]:

"""Execute HTTP request with automatic retry on transient failures.

Args:

method: HTTP method

url: Target URL

read_response: If True, return ``(status, body_bytes)``.

If False, return an open ``aiohttp.ClientResponse`` and the caller

is responsible for consuming and closing it.

**kwargs: Additional arguments for the request

"""

last_exception = None

for attempt in range(self.max_retries):

try:

if not read_response:

response = await self._session.request(method, url, **kwargs)

if response.status < 500:

return response

body = await response.read()

await response.release()

if attempt < self.max_retries - 1:

wait_time = RETRY_BACKOFF_FACTOR * (2 ** attempt)

logger.warning(

f"HTTP {response.status} for {url}, "

f"retrying in {wait_time}s (attempt {attempt + 1}/{self.max_retries})"

)

await asyncio.sleep(wait_time)

continue

return (response.status, body)

async with self._session.request(method, url, **kwargs) as response:

body = await response.read()

Copilot · 2026-04-18T03:03:53Z

+        for attempt in range(self.max_retries):
+            try:
+                async with await self._session.request(method, url, **kwargs) as response:
+                    if response.status < 500 and not read_response:
+                        return response
+                    body = await response.read()
+                    if response.status < 500:
+                        return (response.status, body)


aiohttp.ClientSession.request(...) should not be awaited before async with (use async with self._session.request(...) as response:). Also, returning response from inside an async with will return a closed response object after context exit, and the function annotation claims aiohttp.ClientResponse but returns tuples in other branches—please refactor to return a consistent type (e.g., always (status, body_bytes)), or avoid async with when you intend to return the live response.

Copilot · 2026-04-18T03:03:53Z

+    async def post_stream(
+        self,
+        url: str,
+        payload: Dict[str, Any],
+        headers: Optional[Dict[str, str]] = None
+    ) -> AsyncIterator[Dict[str, Any]]:


post_stream does not use an async context manager for the response (or otherwise ensure response.release()/close), which can leak connections and exhaust the connector under load. Consider async with self._session.post(...) as response: and ensure the response is properly closed on normal completion and exceptions; also consider reusing the retry/timeout strategy used elsewhere for transient disconnects.

Copilot · 2026-04-18T03:03:53Z

+            response = await self._session.post(
+                url,
+                json=payload,
+                headers=headers
+            )
+            response.raise_for_status()
+
+            async for line in response.content:
+                decoded = line.decode('utf-8').strip()
+                if decoded.startswith("data: "):
+                    data_str = decoded[6:].strip()
+                    if data_str:
+                        import json
+                        try:
+                            yield json.loads(data_str)
+                        except json.JSONDecodeError:
+                            logger.warning(f"Failed to parse SSE data: {data_str}")


post_stream does not use an async context manager for the response (or otherwise ensure response.release()/close), which can leak connections and exhaust the connector under load. Consider async with self._session.post(...) as response: and ensure the response is properly closed on normal completion and exceptions; also consider reusing the retry/timeout strategy used elsewhere for transient disconnects.

Suggested change

response = await self._session.post(

url,

json=payload,

headers=headers

)

response.raise_for_status()

async for line in response.content:

decoded = line.decode('utf-8').strip()

if decoded.startswith("data: "):

data_str = decoded[6:].strip()

if data_str:

import json

try:

yield json.loads(data_str)

except json.JSONDecodeError:

logger.warning(f"Failed to parse SSE data: {data_str}")

async with self._session.post(

url,

json=payload,

headers=headers

) as response:

response.raise_for_status()

async for line in response.content:

decoded = line.decode('utf-8').strip()

if decoded.startswith("data: "):

data_str = decoded[6:].strip()

if data_str:

import json

try:

yield json.loads(data_str)

except json.JSONDecodeError:

logger.warning(f"Failed to parse SSE data: {data_str}")

Copilot · 2026-04-18T03:03:53Z

+        if headers:
+            request_headers.update(headers)
+
+        logger.info(f"A2A POST request: url={url}, payload={payload}")


Logging full request payloads at INFO can leak sensitive user content and credentials (e.g., API keys in headers/metadata). Consider logging only high-level fields (method/url/request-id) or redacting sensitive keys, and downgrade verbose payload logging to DEBUG with explicit redaction.

Copilot · 2026-04-18T03:03:55Z

+These endpoints are for internal use only (user authentication required).
+They are NOT registered to any FastAPI app - only used by internal code.


This module docstring claims the router is not registered to any FastAPI app, but backend/apps/config_app.py includes a2a_server_router. Please update the docstring to reflect actual registration/usage to prevent operational confusion.

Suggested change

These endpoints are for internal use only (user authentication required).

They are NOT registered to any FastAPI app - only used by internal code.

These endpoints are intended for internal, authenticated management use.

This module defines the router used to expose these endpoints via FastAPI app registration.

Copilot · 2026-04-18T03:03:55Z

+        default=[]
+    )
+    external_a2a_agents: List["ExternalA2AAgentConfig"] = Field(
+        description="External A2A agents called via HTTP requests",
+        default=[]


Using default=[] for list fields is error-prone and can trigger shared-mutable-default issues (and is discouraged by Pydantic conventions). Prefer default_factory=list for both managed_agents and external_a2a_agents.

Suggested change

default=[]

)

external_a2a_agents: List["ExternalA2AAgentConfig"] = Field(

description="External A2A agents called via HTTP requests",

default=[]

default_factory=list

)

external_a2a_agents: List["ExternalA2AAgentConfig"] = Field(

description="External A2A agents called via HTTP requests",

default_factory=list

Copilot · 2026-04-18T03:03:55Z

+        base_url = str(request.base_url).rstrip("/")
+
+        card = a2a_server_service.get_agent_card(
+            endpoint_id=endpoint_id,
+            base_url=base_url,
+            use_northbound=True
+        )


Using request.base_url will generate incorrect public URLs when the service is behind a reverse proxy (host/port mismatch), which this PR explicitly addresses via NORTHBOUND_EXTERNAL_URL. Consider using NORTHBOUND_EXTERNAL_URL (or honoring X-Forwarded-Host / X-Forwarded-Proto via FastAPI proxy headers middleware) when building the Agent Card base URL.

Copilot · 2026-04-18T03:03:56Z

+const convertToMarkdownCodeFences = (content: string): string => {
+  // Step 1: Handle complete <DISPLAY:language>...</DISPLAY> blocks
+  content = content.replace(/<DISPLAY:(\w+)>([\s\S]*?)<\/DISPLAY>/g, (_match, language, code) => {
+    return `\`\`\`${language}\n${code.trim()}\n\`\`\``;
+  });
+
+  // Step 2: Handle complete <code>...</code> blocks
+  content = content.replace(/<code>([\s\S]*?)<\/code>/g, (_match, code) => {
+    return `\`\`\`python\n${code.trim()}\n\`\`\``;
+  });


The custom-tag → markdown-fence conversion is duplicated across taskWindow.tsx, chatStreamFinalMessage.tsx, and markdownRenderer.tsx. Centralize this into a shared utility (e.g., frontend/lib/markdown/convertCustomCodeTags.ts) to prevent drift and inconsistencies in streaming vs. non-streaming rendering.

Copilot · 2026-04-18T03:03:56Z

+class A2AHttpClient:
+    """HTTP client for A2A protocol communication."""


This introduces a non-trivial retrying HTTP/SSE client (timeouts, backoff, streaming parsing), but no unit tests are added alongside it. Please add tests that cover: retry behavior on transient aiohttp errors, handling of 5xx vs <5xx responses, JSON decode errors, and SSE parsing (data: lines, empty lines, malformed JSON).

WMC001 and others added 6 commits April 17, 2026 11:07

Update default CODEOWNERS to include Dallas98

453b6b3

Merge branch 'main' into release/v2.0.2

31f767f

🔧 Bump APP_VERSION to v2.0.2

d8f1b99

Dallas98 requested a review from WMC001 as a code owner April 18, 2026 03:00

Copilot AI review requested due to automatic review settings April 18, 2026 03:00

Copilot AI reviewed Apr 18, 2026

View reviewed changes

Dallas98 merged commit 5c9504c into main Apr 18, 2026
4 of 5 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Enhance A2A protocol support and fix various bugs#2822

Enhance A2A protocol support and fix various bugs#2822
Dallas98 merged 6 commits intomainfrom
release/v2.0.2

Dallas98 commented Apr 18, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 18, 2026

Uh oh!

Copilot AI Apr 18, 2026

Uh oh!

Copilot AI Apr 18, 2026

Uh oh!

Copilot AI Apr 18, 2026

Uh oh!

Copilot AI Apr 18, 2026

Uh oh!

Copilot AI Apr 18, 2026

Uh oh!

Copilot AI Apr 18, 2026

Uh oh!

Copilot AI Apr 18, 2026

Uh oh!

Copilot AI Apr 18, 2026

Uh oh!

Copilot AI Apr 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

		These endpoints are for internal use only (user authentication required).
		They are NOT registered to any FastAPI app - only used by internal code.

		class A2AHttpClient:
		"""HTTP client for A2A protocol communication."""

Conversation

Dallas98 commented Apr 18, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 18, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 18, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 18, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 18, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 18, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 18, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 18, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 18, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 18, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 18, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants