2. アシスタントを設定してカスタマイズする [GA]

API を有効にする - これは Bumble の管理者が行う操作です。参加者は Google Cloud コンソールにアクセスできない可能性があるため、これは参加者向けではないことを明記してください。

クラウド管理者が完了するワークショップの前提条件として追加する必要があります

パーソナライズを構成する

[Consumer Software & Mobile App Development] がプルダウンにないため、[Custom] を選択して追加する必要があります。

ポップアップを閉じる必要があります。

3. データコネクタを構成する [一般提供]

これは参加者ではなく、Gemini Enterprise 管理者が行います。ワークショップの前提条件として、この点を明記してください。

4. Jira コネクタの基本的な操作 [一般提供]

まず、各参加者が Jira と GitHub のコネクタを使用できるように、コネクタを承認する必要があります。以下はそのスクリーンショットです。

image-1-jira-authozie.png

image-2-jira-authoziation-page.png

image-3-after-jira-authoziation.png

github についても同様に処理します。

Jira コネクタまたは GitHub コネクタを使用している場合は、Google 検索を無効にするよう依頼する

image-4-disable-google-search.png

1. Jira プロジェクト キーを確認する - Gemini に質問するチャットを開く方法を説明します。以下は参考用のスクリーンショットです。

Image-5-jira-list-projects.png

タスクの作成、コメントの追加、ステータスの更新など、他のテストについても同様に処理します。また、それに応じて関連するスクリーンショットも更新しました

image-6-jira-create-projects-1.png

image-7-jira-create-projects-2.png

image-8-jira-create-projects-3.png

image-9-jira-comment-1.png

image-10-jira-comment-2.png

image-11-jira-comment-3.png

image-12-jira-status-1.png

image-13-jira-status-2.png

image-14-jira-status-3.png

5. GitHub Connector の基本的な操作 [一般提供]

To do リストをコミットするには、リポジトリ名を知る必要があります。リポジトリ名を取得するには、次のようなプロンプトを使用します。

このコマンドを実行する前に、参加者に他のコネクタを無効にして GitHub のみを有効にするよう指示する必要があります。

すべてのテストについて、関連する画像を取得して追加します。以下は、スクリーンショットのファイル名です。

image-15-github-enable-1.png

image-16-github-list-repo-2.png

image-17-github-list-commit-3.png

image-18-github-list-pull-request-4.png

image-19-github-list-pull-request-summary-5.png

6. 実行とドキュメントの相互参照 [一般提供]

[Upload Requirements Document] のステップ 2「Upload a file named requirements.txt with the following content」を最初に行う必要があります。これにより、参加者はノートパソコンでファイルを作成し、それをアップロードできるようになります。Mac で .txt ファイルを作成するには、内蔵のテキストエディット アプリを使用します。テキストエディットを開き、上部のメニューで [フォーマット] > [標準テキストにする] をクリックします。テキストを入力し、[ファイル] > [保存] をクリックして、ファイルに名前を付け(末尾が .txt になっていることを確認)、保存場所を選択してクリックします。他のノートパソコンについても同様の手順を説明します

また、「Gemini Enterprise のチャット インターフェースで、+(プラス)アイコンをクリックするか、ファイルをドラッグ&ドロップしてアップロードします。」という記述は、この段階では関連性がありません。

要件テキストを作成したら、新しいチャット ウィンドウを開き、他のコネクタを無効にして Jira のみを有効にします。次に、Gemini を使用してサンプルデータを作成し、分析を実行します。この時点で、txt ファイルを Gemini アプリにアップロードする必要があります。このテストのスクリーンショットについては、以下のスクリーンショットのファイル名を参照してください。

image-20-jira-req-txt-1.png

Gemini を使用してサンプルデータを作成するの画像

image-21-jira-create-ticket-2.png

image-22-jira-create-ticket-confirmation-3.png

image-23-jira-create-ticket-done-status-confirmation-4.png

image-24-jira-create-ticket-confirmation-5.png

image-25-jira-create-ticket-inprogress-status-confirmation-6.png

image-26-jira-create-ticket-created-6.png

image-27-jira-create-tinjira-portal-7.png

分析を実行するの画像

image-28-jira-run-analysis-1.png

image-29-jira-run-analysis-output-2.png

「回答に記載されている引用をクリックすると、Gemini が参照しているファイルまたは Jira チケットのどの部分かを確認できます。」とありますが、引用が表示されません。

7. NotebookLM でスプリント バックログを分析する [一般提供]

同様に、まず txt ファイルを作成し、その手順を説明してから、[新しいノートブックを作成] をクリックします。次に、次のモックメモを含む retro.txt という名前のファイルをアップロードします。

分析情報を生成する際にクリックする内容について指示します。

これらのスクリーンショットの画像ファイルは次のとおりです。

image-30-notebook-lm-output.png

image-31-notebook-lm-mindmap.png

実際のテスト実行のスクリーンショットを確認し、不足しているものがあれば手順を更新します。スクリーンショット画像のファイル名。また、完了するまでに時間がかかるため、準備が整うまでトリガーして次のラボに進むことも伝えます。また、レポートで確認する内容、

deep-research-1.png

deep-research-2.png

deep-research-3.png

9. アイデア生成によるブレインストーミング [パブリック プレビュー]

実際のテスト実行のスクリーンショットを確認し、不足しているものがあれば手順を更新します。スクリーンショット画像のファイル名。また、完了するまでに時間がかかるため、準備が整うまでトリガーして次のラボに進むことも伝えます。また、レポートで確認する内容、

idea-agent-1

idea-agent-2

idea-agent-3

idea-agent-4

idea-agent-5

10. マルチステップのスクラム マスター アシスタントを構築する [パブリック プレビュー]

最初の手順は、[新しいエージェント] を選択したときに [チャット エージェント] をクリックすることです。これは、[新しいエージェント] に [ワークフロー エージェント] のプレビュー オプションがあるためです。このセクションでは、[ワークフロー エージェント] については説明しません。

画像 scrum-master-1.png を参照してください。

エージェントの名前を変更するには、中央のボックスをクリックしてサイドパネルを開く必要があります。画像 scrum-master-3.png と scrum-master-4.png を参照してください。

プロンプトを入力する次のステップについては、画像(scrum-master-5.png)を参照してください。

ステップ 3 の「Give it these instructions」では、実際には Google 検索が有効になっているエージェントが作成され、エージェントの名前が「Lead Scrum Master」に変更されます。画像 scrum-master-6.png を参照してください。プロンプトでエージェントの名前が変更されるため、エージェントの名前を「Bumble-Lead-Scrum-Master」に変更するステップ 2 は削除しますが、必ず「リード スクラム マスター」という名前に変更されるようにします。

また、プロンプトで、Google 検索を使用せずに Jira コネクタと GitHub コネクタのみを使用するように指示します。

また、残りの手順の「Bumble-Lead-Scrum-Master」を現在のエージェント名に更新します。

専門サブエージェントを追加する

ステップ 1

手順で、プロジェクト キーが以前に 4 で検索されたことを言及します。Jira Connector の基本的な操作 [一般提供] ラボステップ 1

Google 検索のチェックを外し、Jira のみチェックを入れます。
以下の画像を参照し、手順を適宜更新してください。

scrum-master-7.png

Scrum-master-8.png

ステップ 2

手順で、リポジトリ名が以前に 5 で検索されたことを明記します。GitHub Connector の基本的な操作 [一般提供] ステップ 1

google 検索のチェックを外し、github のみをオンにします。
以下の画像を参照し、手順を適宜更新してください。

scrum-master-9.png

scrum-master-10.png

scrum-master-11.png

このプロンプトは実際には公開の GitHub を検索し、「過去 24 時間の Bumble プロジェクトの優先度の高い更新を生成する」という結果を取得しています。これは完全に間違っています。エージェントは、私が指定した Jira プロジェクトと GitHub リポジトリのデータのみを検索する必要があります。理由はわかりません。

出力結果については、こちらのスクリーンショット(scrum-master-12.png)を参照してください。実際の出力は次のとおりです。

エージェントで構成された Jira プロジェクトと GitHub レポートのみを検索するように、手順を正しく更新してください。

Generate the high-priority update for the Bumble project for the past 24 hours.
Response complete
Gemini replied
Here is the high-priority daily update for the Bumble project ecosystem over the past 24 hours. [107, 109, 928, 934]
-
🚀 Overall Progress Summary
The past 24 hours saw significant progress in integrating Google’s Bumble Bluetooth stack across downstream tools, notably introducing native Windows support for NXBT 

 and adding optional BLE pairing transports to smpclient [107]. Additionally, core refactoring in Bumble's AVDTP (Audio/Video Distribution) layers has successfully eliminated resource-locking bugs [928]. Meanwhile, the community is actively tackling high-priority Classic Bluetooth pairing failures and Windows boot-up driver issues [59, 934].
-
✨ Key Achievements & Code Progress (GitHub)
Repository / Project
Contribution
Details
intercreate/smpclient
BLE Transport with Pairing (PR #107)
Added an optional bumble extra (smpclient[bumble]) providing SMPBumbleTransport [107]. This BLE transport drives external HCI USB controllers, supporting secure scanning, pairing PIN flows, and stored LTK encryption [107].
google/bumble
AVDTP In-Use Management (PR #928)
Merged! Avoids explicit stream in_use flag management by delegating to a dynamic property checking if stream state is IDLE [928]. This resolves previous edge-case resource leaks [923].
typenoob/nxbt
Native Windows Support (Issue #4)
Announced a major integration with Bumble enabling native Windows support for NXBT controllers via hci-socket and usb (WinUSB) transports 

 .
badoo/codeisok
Security Hardening (PR #109)
Enabled the secure flag for authentication cookies in the Login controller to prevent unencrypted session token transmission [109].
 
-
⚠️ Current Blockers & High-Priority Issues
While no tracking issues were found in Jira for "Bumble", the following high-priority issues are currently impeding progress in the GitHub ecosystem:
Classic Bluetooth Pairing Failures (Issue #59 - kindle-hid-passthrough) [59]
Status: Active / Investigation [59]
Impact: Classic-only Bluetooth HID devices (such as the Logitech K811 keyboard) fail during Simple Secure Pairing (SSP) with an AUTHENTICATION_FAILURE_ERROR (0x05) [59]. Bumble appears to bypass the user pairing delegate entirely [59].
Next Steps: Testing a speculative fix on feature/brcm-bt-support that relaxes pairing config constraints for BR/EDR [59].
Windows Startup Power-On Failures (Issue #934 - google/bumble) [934]
Status: Open [934]
Impact: Bluetooth USB dongles fail to power on during the very first boot on Windows systems [934]. The Intel driver receives a disallowed error (HCI_COMMAND_DISALLOWED_ERROR) during the initial HCI_RESET_COMMAND [934].
-
Would you like me to draft any replies to these active issues or prepare a Confluence summary document of the Windows workarounds?
give me the link for this repo intercreate/smpclient
Response complete
Gemini replied
The link for the intercreate/smpclient repository is:
👉 https://github.com/intercreate/smpclient
intercreate/smpclient repo is not in my repo in mumanoha/gemini-code-review-agent-demo. intercreate/smpclient is a public repo why are you getting me all the public repo
Response complete
Gemini replied
I retrieved results from public repositories because the search query for "Bumble" was executed globally across all of GitHub rather than being restricted to your specific repository (mumanoha/gemini-code-review-agent-demo).
When I initially searched your internal Jira projects and repositories for the project "Bumble", it returned no matches. To ensure I didn't miss any relevant information, my search query broadened its scope to look for any active issues or pull requests containing the keyword "Bumble" globally, which pulled in these public repositories (like Google's bumble library and intercreate/smpclient).
-
🔍 Filtering to Your Specific Repository
If "Bumble" is a branch, a specific folder, or a topic within your repository, I can narrow down the search.
Would you like me to run a targeted analysis specifically on mumanoha/gemini-code-review-agent-demo for any updates or commits made in the past 24 hours instead?

「過去 24 時間のプロジェクトの優先度の高い更新を生成して」というプロンプトを使用すると、次の出力が得られます。

画像 scrum-master-13.png を参照してください

Generate the high-priority update for the my projects for the past 24 hours.
Response complete
Gemini replied
📊 Daily Project Status Report (Past 24 Hours)
-
🚀 Overall Progress Overview
Currently, the projects are focused on resolving critical infrastructure blockers on Google Cloud Platform while maintaining a steady baseline on application services. While there has been no new code merged or committed in the last 24 hours, the active development continues to center around service integration, and the infrastructure pipeline is currently held back by two newly updated high-priority Jira issues 

 .
-
🛠️ Code Progress & Achievements (GitHub)
There are no new commits or merged pull requests to report for the past 24 hours. The primary active work-in-progress remains:
Repository / PR
Status
Description
mumanoha/gemini-code-review-agent-demo PR #1: feat: Add menu-service application
🟡 Open
This pull request introduces the core menu-service application to the repository.
 
-
⚠️ Blockers & High-Priority Issues (Jira)
The following high-priority infrastructure issues were updated in the past 24 hours under the ge-bugs project and require immediate attention to prevent further deployment delays on the GCP Migration Strategy 

 :
Key
Priority
Status
Summary
Impact
GB-5
🔴 High
To Do
IAM permission shadowing in VPC
Conflicting IAM policies between the VPC and GKE cluster are causing intermittent permission denied errors for active services 

 .
GB-4
🔴 High
To Do
GKE node pool auto-scaling timeout
The GKE node pool is failing to auto-scale automatically, leading to pod scheduling failures under load 

 .
 
-
Let me know if you would like me to help assign these blockers or draft a follow-up message to the infrastructure team!

また、コネクタをルート エージェントに追加する必要があるかどうかもわかりません。ルート エージェントのコネクタを削除してテストしたところ、以下のような出力が得られました。正しい構成はどれですか?正しい構成はご自身で判断してください。


画像(scrum-master-14.png)を参照してください。

Generate the high-priority update for the my projects for the past 24 hours.
Response complete
Gemini replied
High-Priority Update: Project GB (ge-bugs)
Here is the summary of P0/P1 (Highest/High priority) issues and blocked tickets in Project GB (ge-bugs) along with their current impact:
1. P0/P1 Issues (High Priority)
GB-4: GKE node pool auto-scaling timeout
Priority: High (P1) 

Impact: The GKE node pool is failing to auto-scale dynamically under load. This results directly in pod scheduling failures, threatening application reliability and performance during high traffic 

 .
GB-5: IAM permission shadowing in VPC
Priority: High (P1)
Impact: Conflicting IAM policies between the VPC and the GKE cluster are causing intermittent "permission denied" errors for services, leading to unstable internal communication 

 .
2. Blocked Tickets
GB-7: Fix database latency issues
Priority: Medium
Status: Blocked 

Impact: Progress on resolving database latency is halted. The ticket is currently blocked while waiting for the infrastructure team to increase IOPS 

 .
-
I will now hand control back to the root agent to synthesize this update into your unified daily report. Let's make sure this progress is tracked.
I have checked the open Pull Requests in the mumanoha/gemini-code-review-agent-demo repository:
Open Pull Requests:
PR #1: feat: Add menu-service application
State: Open
Review Status: There is currently one review submitted by gemini-code-assist[bot], which has a status of COMMENTED (not CHANGES_REQUESTED). The reviewer provided general code quality, testing, and dependency cleanup suggestions but did not block the PR with an explicit changes-requested status.
Therefore, there are no Pull Requests currently in the "Changes Requested" state.
I will now transfer control back to the root_agent to compile the complete update.Here is your unified daily report for the past 24 hours, highlighting the high-priority Jira blockers and active code progress on GitHub.
I, Gemini Enterprise, have analyzed your active projects and compiled the most critical updates to keep you and your team aligned.
-
🚨 High-Priority Jira Issues & Blockers
There are currently two high-priority bugs active in Project GB (ge-bugs), along with one blocked issue that requires infrastructure intervention.
Issue
Priority
Status
Description & Impact
GB-4: GKE node pool auto-scaling timeout
High
To Do
The GKE node pool is failing to auto-scale under load, resulting in active pod scheduling failures.
GB-5: IAM permission shadowing in VPC
High
To Do
Conflicting IAM policies between the VPC and GKE cluster are causing intermittent service permission errors.
GB-7: Fix database latency issues
Medium
🛑 Blocked
Progress is currently halted. The team is waiting on the infrastructure team to increase IOPS.
 
-
💻 GitHub Code Progress
Active progress continues on the gemini-code-review-agent-demo repository with one open pull request. There are currently no PRs blocked under a "Changes Requested" status.
PR # & Title
State
Review Status
Key Highlights
#1: feat: Add menu-service application
Open
Commented
Adds the menu-service Quarkus application. gemini-code-assist[bot] provided recommendations to clean up pom.xml, improve JPA mapping, utilize Panache's active record pattern, and expand test coverage.
 
-
Would you like me to help you draft a follow-up message to the infrastructure team to expedite the IOPS increase for GB-7, or would you like to dive deeper into any of these updates?

11. Resource Impact Analyzer エージェントをビルドする [パブリック プレビュー]

最初の手順は、[新しいエージェント] を選択したときに [チャット エージェント] をクリックすることです。これは、[新しいエージェント] に [ワークフロー エージェント] のプレビュー オプションがあるためです。このセクションでは、[ワークフロー エージェント] については説明しません。

画像 scrum-master-1.png を参照してください。

[Create a New agent] > [Proceed to builder] ではなく、チャット自体にプロンプトを入力してみましょう。このスクリーンショット リソース(resource-impact-1.png)を参照してください。

エージェントがプロジェクトを作成したら、Google コネクタのチェックを外して [プレビュー] をクリックし、「チームメンバーの Muthu は 3 日間休みです。どのような影響がありますか?」

画像を参照し、手順を適宜更新してください。
resource-impact-2.png と resource-impact-3.png

テストが完了したら、[作成] をクリックしてエージェントを公開します。

12. 高度: ADK を使用してビルドとデプロイを行う [公開プレビュー]

これらの手順のスクリーンショットを撮りました。スクリーンショットの画像を確認して、手順が正確であることを確認し、各手順のスクリーンショットを更新して、わかりやすくしてください。

ステップ 1: 環境設定

agent-cli-1.png

ステップ 2: 認証とセットアップ

agent-cli-2.png
agent-cli-3.png - この手順には時間がかかります
agent-cli-4.png -

cli がコマンドの実行権限を求めるので、[1 回のみ許可] を選択するよう手順に記載

オプション B: Gemini CLI の方法(自然言語)

コンソールで実際に発生したことに基づいて手順を正しく更新するには、画像を参照してください。必要に応じて、これらのスクリーンショットと関連する手順で Codelab を更新します

確認する画像 - agent-cli-5.png から agent-cli-9.png

以下は CLI の出力です。

Agent Development Kit(ADK)を使用して、Bumble-Ops-Assistant エージェントを構築しました。

1. エージェントの実装
エージェントは bumble-ops-assistant/app/agent.py にあります。これには、任意の環境に対してハードコードされた「GREEN」ステータスを返す check_build_health ツールが含まれています。

1 def check\_build\_health(environment: str) \-\> str:                                                                                                                          
2     """Checks the build health for a given environment.                                                                                                                   
3                                                                                                                                                                           
4     Args:                                                                                                                                                                 
5         environment: The environment to check (e.g., 'prod', 'staging').                                                                                                  
6                                                                                                                                                                           
7     Returns:                                                                                                                                                              
8         The health status of the build.                                                                                                                                   
9     """                                                                                                                                                                   

10 return "GREEN"
11
12 root_agent = Agent(
13 name="Bumble_Ops_Assistant",
14 # ...
15 instruction="You are Bumble-Ops-Assistant, an operations assistant. 本番環境とステージング環境のビルドの健全性を確認できます。」、
16 tools=[check_build_health],
17 )

2. スモークテスト
2 種類のスモークテストを実施しました。

* 単体テスト: bumble-ops-assistant/tests/unit/test_health_tool.py で単体テストを作成して実行し、ツールのロジックを直接検証しました。
* 結果: 3.03 秒で 2 件合格
* 評価スモークテスト: エージェントのインタラクションをテストするために、bumble-ops-assistant/tests/eval/evalsets/smoke_test.evalset.json にシンプルな評価セットをスキャフォールディングしました。
* 注: aiplatform.googleapis.com API が cloudshell-gca プロジェクトで有効になっていないため、評価実行で PERMISSION_DENIED エラーが発生しました。また、この API を有効にする権限もありません。

agent-cli-10.png - [open editor] をクリックしたときに表示され、参加者にファイルを確認して、エージェント CLI が作成した内容を確認するよう求めます。エージェント CLI が作成した各ファイルの説明が表示されます。詳細については、ドキュメントを参照してください。
Playground テスト - デプロイと公開の前にこれらの手順を提供します。
これらの画像を参照してテストを行い、それに応じて手順を作成します。
画像 - agent-cli-11.png から agent-cli-15.png

エージェント CLI にプレイグラウンドの作成とテストを依頼しようとしましたが、問題が発生しています。
セッションが見つかりません: f681747b-c55d-42d6-9324-77f394d0ecb8。ランナーはアプリ名「bumble_ops_assistant」で構成されていますが、ルート エージェントは「/home/admin_/bumble-ops-assistant/app」から読み込まれました。これは、アプリ名が「app」であることを意味します。ランナーの app_name がそのディレクトリと一致していることを確認するか、ランナーの構築時に app_name を明示的に渡します。この不一致により、ランナーはセッションを見つけることができません。セッションがない場合に自動的に作成するには、ランナーの構築時に auto_create_session=True を設定します。

エージェントに修正を依頼したところ、権限の問題が解決し、太ももが機能し始めました。

13. Gemini Enterprise にデプロイして公開する [公開プレビュー]

自然言語を使用して GE
にデプロイして公開したいので、手動の方法と自然言語の方法の両方のオプションを残してください。

エージェントのデプロイをリクエストしたところ、エージェント エンジンにデプロイされました。手順を理解し、関連する手順を作成するには、画像を参照してください。
画像 - agent-cli-deploy-1.png から agent-cli-deploy-4.png

gemini cli に Google Earth アプリへの公開を依頼しました

- Publish to Gemini Enterprise App bumble-workshop-lab-test-1_1778167389086
以下の画像を参照して、この操作がどのように行われたかを理解し、それに応じて手順を作成してください。

agent-cli-publish-1.png から agent-cli-publish-7.png