🐛 Bugfix: Auto-clean account when exists in Supabase but not in postgresql

[xuyaqist]

1. 清除仅存在supabse但是不存在我们数据库的用户
2. 删除A2AAgent的type字段（在1.0协议版本中移除该字段）
3. 延长session refresh的逻辑：如果用户在活跃状态时，如果session<30分钟过期，那么就会refresh token
4. 当后端返回401，前端直接弹出登录页面

[Copilot]

Pull request overview

This PR attempts to address inconsistent user state between Supabase and the local PostgreSQL DB by deleting “orphaned” Supabase accounts when the corresponding local tenant relationship is missing. It also changes the A2A response payload shape for message/task parts and adjusts image sizing in the agent-development documentation.

Changes:

• Add Supabase admin-client deletion logic in get_user_info when user_tenant is missing.
• Modify A2A adapter/server responses to omit type: "text" in generated parts objects (and in one place omit mediaType too).
• Increase documentation screenshot image widths from 50% to 80% (EN/ZH).

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
doc/docs/zh/user-guide/agent-development.md	Increases embedded image width for A2A discovery screenshots.
doc/docs/en/user-guide/agent-development.md	Same image width adjustment for English docs.
backend/services/user_management_service.py	Deletes Supabase users when local tenant relationship is missing during user info lookup.
backend/services/a2a_server_service.py	Changes emitted A2A parts objects (drops type, and in one path drops mediaType).
backend/services/a2a_agent_adapter.py	Changes default/converted A2A parts shape (drops type, sometimes mediaType), impacting existing contract/tests.

Comments suppressed due to low confidence (4)

backend/services/a2a_agent_adapter.py:272

• build_a2a_message_response now creates default text parts without a type field. Repository tests and existing payloads consistently include type: "text" for parts; dropping it will cause incompatibilities unless every consumer is updated. Consider restoring type (and keeping mediaType) for generated text parts.

        if parts:
            message_parts = parts
        elif text:
            message_parts = [{"text": text, "mediaType": _MEDIA_TYPE_TEXT}]
        else:
            message_parts = [{"text": "", "mediaType": _MEDIA_TYPE_TEXT}]

backend/services/a2a_agent_adapter.py:298

• _content_to_artifact_parts returns parts without type, while other parts in the codebase/tests expect type: "text" (and often mediaType). To keep A2A artifact parts consistent and avoid downstream parsing issues, include type for generated text parts (and preserve mediaType).

        """Convert content/parts into artifact parts format."""
        if parts:
            return parts
        if isinstance(content, dict):
            if content.get("type") == "text":
                return [{"text": content.get("text", ""), "mediaType": _MEDIA_TYPE_TEXT}]
        return [{"text": str(content), "mediaType": _MEDIA_TYPE_TEXT}]

backend/services/a2a_agent_adapter.py:347

• _message_to_parts_format now emits parts: [{"text": ...}] without type (and without mediaType). This will fail existing unit tests (they assert parts[0]["type"] == "text") and makes status messages inconsistent with other A2A responses. Please include type: "text" (and ideally mediaType: "text/plain") when synthesizing parts here.

            text = str(message)
        return {
            "role": role,
            "parts": [{"text": text}]
        }

backend/services/a2a_server_service.py:884

• get_task now returns artifact parts as [{"text": ...}] (no type and no mediaType), while other responses in this module/adapter typically include at least mediaType: "text/plain" (and historically type). This inconsistency can break clients that rely on uniform Part objects. Consider including mediaType (and type if required) for artifact parts.

            if message:
                task_obj["artifacts"] = [{
                    "parts": [{"text": str(message)}],
                    "lastChunk": True
                }]

---

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