adk-js
adk-js copied to clipboard
Published
3 weeks ago
•
google
google
Unable to submit request because it must include at least one parts field
Describe the bug A root agent is unable to transfer to subagents.
To Reproduce Steps to reproduce the behavior:
- Install
@google/adkv0.2.0 - Add:
import { LlmAgent } from '@google/adk';
export const jokeAgent = new LlmAgent({
name: 'tell_a_joke_agent',
model: 'gemini-2.0-flash',
description: 'Tell a joke',
instruction: `You are a helpful assistant that tells a joke.`,
});
export const philosophicalAgent = new LlmAgent({
name: 'philosophical_agent',
model: 'gemini-2.0-flash',
description: 'Say a philosophical statement',
instruction: `You are a helpful assistant that says a philosophical statement.`,
});
export const rootAgent = new LlmAgent({
name: 'root_agent',
model: 'gemini-2.0-flash',
description: 'Root agent',
instruction: `You are a helpful assistant that is the root agent.`,
subAgents: [jokeAgent, philosophicalAgent],
});
- Start the local web UI and trigger one of the subagents, like
tell me a joke - See the error appear:
Unable to submit request because it must include at least one parts field, which describes the prompt input. Learn more: https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/gemini
Expected behavior It should be able to transfer to a subagent
Screenshots
Desktop (please complete the following information):
- OS: macOS (Darwin 25.1.0, arm64)
- TS version/environment: 5.9.3
- ADK version(see maven dependency): 0.2.0
Hello, thanks for submitting the issue.
I tried to replicate that and it's not reproduced on my side:
Can you please double-check that or provide some additional context, thanks!
I have the same issue. The first prompt always seems to work and then it somehow breaks and I get the same response all the time.
"@google/adk": "^0.2.1", "@google/adk-devtools": "^0.2.1"
google_adk % npx @google/adk-devtools run gitlab_workflow/index.ts
[[email protected]] injecting env (3) from .env -- tip: ⚙️ load multiple .env files with { path: ['.env.local', '.env'] }
Running agent coordinator_agent, type exit to exit.
[user]: what can you do?
[ADK INFO]: event: {"invocationId":"e-4eb3b16f-1b8b-4b27-b55f-92940f9dab69","author":"user","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"content":{"role":"user","parts":[{"text":"what can you do?"}]},"id":"li9wSLls","longRunningToolIds":[],"timestamp":1766052954063}
[ADK INFO]: Sending out request, model: gemini-3-flash-preview, backend: VERTEX_AI, stream: false
[coordinator_agent]: I am the **Project Setup Coordinator**, and I can automate the end-to-end provisioning of new cloud projects on **Google Cloud Platform (GCP)** or **SAP Business Technology Platform (BTP)**.
Here is what I can do for you:
1. **GitLab Project Creation**: I automatically create a new GitLab repository for your project using approved corporate templates.
2. **Infrastructure as Code (IaC)**: I coordinate the update of the central Landing Zone repositories. This includes creating a feature branch, adding your project definition to the Terraform configuration, and opening a Merge Request.
3. **Lifecycle Management**: I monitor the Merge Request for feedback and ensure the infrastructure setup proceeds correctly.
4. **Environment Configuration**: Once the project is ready, I inject the required CI/CD variables (like Project IDs, Regions, or Credentials) into your new GitLab project so you can start developing immediately.
**To get started, please provide me with:**
* **Project Name**
* **Target Cloud** (Google or BTP)
* **Jira Ticket ID**
[user]: Please create a new project called "Agent Test" on gcp. Jira Ticket Number is SOR-23455.
[ADK INFO]: event: {"invocationId":"e-b1fca1d7-06f6-4059-8682-e06ba090d943","author":"user","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"content":{"role":"user","parts":[{"text":"Please create a new project called \"Agent Test\" on gcp. Jira Ticket Number is XXX"}]},"id":"kX6Ig3xM","longRunningToolIds":[],"timestamp":1766052987543}
[ADK INFO]: event: {"invocationId":"e-4eb3b16f-1b8b-4b27-b55f-92940f9dab69","author":"coordinator_agent","id":"U99OjX9p","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"longRunningToolIds":[],"timestamp":1766052954068,"content":{"role":"model","parts":[{"text":"I am the **Project Setup Coordinator**, and I can automate the end-to-end provisioning of new cloud projects on **Google Cloud Platform (GCP)** or **SAP Business Technology Platform (BTP)**.\n\nHere is what I can do for you:\n\n1. **GitLab Project Creation**: I automatically create a new GitLab repository for your project using approved corporate templates.\n2. **Infrastructure as Code (IaC)**: I coordinate the update of the central Landing Zone repositories. This includes creating a feature branch, adding your project definition to the Terraform configuration, and opening a Merge Request.\n3. **Lifecycle Management**: I monitor the Merge Request for feedback and ensure the infrastructure setup proceeds correctly.\n4. **Environment Configuration**: Once the project is ready, I inject the required CI/CD variables (like Project IDs, Regions, or Credentials) into your new GitLab project so you can start developing immediately.\n\n**To get started, please provide me with:**\n* **Project Name**\n* **Target Cloud** (Google or BTP)\n* **Jira Ticket ID**","thoughtSignature":"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"}]},"usageMetadata":{"promptTokenCount":733,"candidatesTokenCount":233,"totalTokenCount":1153,"trafficType":"ON_DEMAND","promptTokensDetails":[{"modality":"TEXT","tokenCount":733}],"candidatesTokensDetails":[{"modality":"TEXT","tokenCount":233}],"thoughtsTokenCount":187},"finishReason":"STOP"}
[ADK INFO]: Sending out request, model: gemini-3-flash-preview, backend: VERTEX_AI, stream: false
[ADK INFO]: event: {"invocationId":"e-00408b19-6e74-48fe-ab47-6eb49e9f447e","author":"user","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"content":{"role":"user","parts":[{"text":"Create a new project named 'Agent Test' in the namespace 'nla/blub' using the template 'template'. Please return the project ID and Web URL."}]},"id":"XC63xyXn","longRunningToolIds":[],"timestamp":1766052990638}
[ADK INFO]: Sending out request, model: gemini-3-flash-preview, backend: VERTEX_AI, stream: false
[ADK INFO]: event: {"invocationId":"e-ce438170-8973-46b5-9c95-9e48e14c7c85","author":"user","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"content":{"role":"user","parts":[{"text":"Please create a new GitLab project. \nProject Name: Agent Test\nNamespace: nla/blub\nTemplate: template\nReturn the project ID and the Web URL."}]},"id":"HU0C4Edt","longRunningToolIds":[],"timestamp":1766052993377}
[ADK INFO]: Sending out request, model: gemini-3-flash-preview, backend: VERTEX_AI, stream: false
[user]: what is the status?
[ADK INFO]: event: {"invocationId":"e-c7973a12-e436-47ed-806e-e791238e3dac","author":"user","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"content":{"role":"user","parts":[{"text":"what is the status?"}]},"id":"GrMpWB3K","longRunningToolIds":[],"timestamp":1766053709090}
[ADK INFO]: event: {"invocationId":"e-b1fca1d7-06f6-4059-8682-e06ba090d943","author":"coordinator_agent","id":"ohpc9Dq6","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"longRunningToolIds":[],"timestamp":1766052993379,"errorCode":"400","errorMessage":"Unable to submit request because it must include at least one parts field, which describes the prompt input. Learn more: https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/gemini"}
[ADK INFO]: Sending out request, model: gemini-3-flash-preview, backend: VERTEX_AI, stream: false
[user]:
@kalenkevich thanks for the answer! That's strange though. Here is my complete setup:
google-adk-demo/
├── package.json
├── tsconfig.json
└── src/
└── agent.ts
package.json
{
"name": "google-adk-demo",
"version": "1.0.0",
"description": "",
"type": "module",
"main": "index.js",
"scripts": {
"start": "npx ts-node --esm src/index.ts",
"adk": "npx adk web"
},
"keywords": [],
"author": "",
"license": "ISC",
"dependencies": {
"@google/adk": "^0.2.1"
},
"devDependencies": {
"@google/adk-devtools": "^0.2.1",
"@types/node": "^25.0.3",
"ts-node": "^10.9.2",
"typescript": "^5.9.3"
}
}
tsconfig.json
{
"compilerOptions": {
"target": "ES2022",
"module": "NodeNext",
"moduleResolution": "NodeNext",
"esModuleInterop": true,
"strict": true,
"skipLibCheck": true,
"outDir": "./dist"
},
"include": ["src/**/*"]
}
src/agent.ts
import { LlmAgent } from '@google/adk';
export const jokeAgent = new LlmAgent({
name: 'tell_a_joke_agent',
model: 'gemini-2.0-flash',
description: 'Tell a joke',
instruction: `You are a helpful assistant that tells a joke.`,
});
export const philosophicalAgent = new LlmAgent({
name: 'philosophical_agent',
model: 'gemini-2.0-flash',
description: 'Say a philosophical statement',
instruction: `You are a helpful assistant that says a philosophical statement.`,
});
export const rootAgent = new LlmAgent({
name: 'root_agent',
model: 'gemini-2.0-flash',
description: 'Root agent',
instruction: `You are a helpful assistant that is the root agent.`,
subAgents: [jokeAgent, philosophicalAgent],
});
Then I run it with pnpm exec adk web src. Is there any other context information that would be helpful to share?
If it's of any help, here are some versions I'm on:
| Component | Version |
|----------------------|------------------------------|
| OS | macOS (Darwin 25.1.0, arm64) |
| Node.js | v22.12.0 |
| pnpm | 10.2.0 |
| @google/adk | 0.2.1 |
| @google/adk-devtools | 0.2.1 |
| TypeScript | 5.9.3 |
| Google Cloud SDK | 539.0.0 |
and using Vertex AI via Application Default Credentials (which works since I can get the root agent to respond).
I copied your agent definitions into an app and am able to reproduce locally. It does happen after the first message. We will take a look and see why the subagent transfer is breaking. Thank you for reporting and we'll let you know.
So far I've discovered that using gemini-3-flash-preview works fine and so does gemini-2.5-flash.
I'm not sure yet what the root cause is for 2.0 but if you can use 2.5 or 3 for now you will be able to continue developing. Newer models are significantly better, so I would encourage you to use the latest one if possible. I will keep this open for now but since the mitigation is easy it will be lower priority.
Thanks @ScottMansfield! I tried 2.5 (I don't have access to 3 preview yet as per company policy), but still getting the same error unfortunately.
[ADK INFO]: event: {"invocationId":"e-cc7efc0b-fb92-4bf7-95f3-b25629d6b57c","author":"user","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"content":{"role":"user","parts":[{"text":"what can you do"}]},"id":"h7tZFiK7","longRunningToolIds":[],"timestamp":1766128312240}
[ADK INFO]: Sending out request, model: gemini-2.5-flash, backend: VERTEX_AI, stream: false
[ADK INFO]: event: {"invocationId":"e-4b0ff6c4-22bf-42ac-b723-b134f0918fce","author":"user","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"content":{"role":"user","parts":[{"text":"tell me a joke"}]},"id":"Oiljvuri","longRunningToolIds":[],"timestamp":1766128318115}
[ADK INFO]: event: {"invocationId":"e-cc7efc0b-fb92-4bf7-95f3-b25629d6b57c","author":"root_agent","id":"XLHYled8","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"longRunningToolIds":[],"timestamp":1766128312243,"content":{"role":"model","parts":[{"text":"I am the root agent. I can answer general questions and help you navigate to other specialized agents. I can transfer you to an agent that can tell jokes or one that can make philosophical statements. How can I help you today?"}]},"usageMetadata":{"promptTokenCount":205,"candidatesTokenCount":46,"totalTokenCount":419,"trafficType":"ON_DEMAND","promptTokensDetails":[{"modality":"TEXT","tokenCount":205}],"candidatesTokensDetails":[{"modality":"TEXT","tokenCount":46}],"thoughtsTokenCount":168},"finishReason":"STOP"}
[ADK INFO]: Sending out request, model: gemini-2.5-flash, backend: VERTEX_AI, stream: false
[ADK INFO]: Sending out request, model: gemini-2.5-flash, backend: VERTEX_AI, stream: false
Do you have any other ideas what the problem might be? I wonder what else there could be on my end that causes this issue.
I don't want to just say "works on my machine" so I will give you what I'm working from so you might find something I can't yet see. I tried the web just now with 2.5 flash and it worked fine.
[ADK INFO]: event: {"invocationId":"e-7d308d96-c81c-462d-bb5b-f10694907da4","author":"user","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"content":{"role":"user","parts":[{"text":"what can you do?"}]},"id":"rnLG6NLr","longRunningToolIds":[],"timestamp":1766128735383}
[ADK INFO]: Sending out request, model: gemini-2.5-flash, backend: GEMINI_API, stream: false
[ADK INFO]: event: {"invocationId":"e-083ee11c-a834-4ad0-9b71-e31702bbccea","author":"user","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"content":{"role":"user","parts":[{"text":"tell me a joke"}]},"id":"wKcevqDM","longRunningToolIds":[],"timestamp":1766128740015}
[ADK INFO]: event: {"invocationId":"e-7d308d96-c81c-462d-bb5b-f10694907da4","author":"root_agent","id":"16yERar8","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"longRunningToolIds":[],"timestamp":1766128735386,"content":{"parts":[{"text":"I am a helpful assistant. I can also tell jokes or make philosophical statements if you'd like.\n"}],"role":"model"},"usageMetadata":{"promptTokenCount":232,"candidatesTokenCount":21,"totalTokenCount":253,"promptTokensDetails":[{"modality":"TEXT","tokenCount":232}]},"finishReason":"STOP"}
[ADK INFO]: Sending out request, model: gemini-2.5-flash, backend: GEMINI_API, stream: false
[ADK INFO]: Sending out request, model: gemini-2.5-flash, backend: GEMINI_API, stream: false
One difference I noticed is that I am using GEMINI_API as the backend and you are using VERTEX_AI. I don't think this should cause a problem but it's a difference.
I followed the quickstart to get a clean env and put in your agent definitions in place of the ones in the article.
Here's what I am using:
agent.ts:
import { LlmAgent } from '@google/adk';
export const jokeAgent = new LlmAgent({
name: 'tell_a_joke_agent',
// model: 'gemini-2.0-flash',
model: 'gemini-2.5-flash',
// model: 'gemini-3-flash-preview',
description: 'Tell a joke',
instruction: `You are a helpful assistant that tells a joke.`,
});
export const philosophicalAgent = new LlmAgent({
name: 'philosophical_agent',
// model: 'gemini-2.0-flash',
model: 'gemini-2.5-flash',
// model: 'gemini-3-flash-preview',
description: 'Say a philosophical statement',
instruction: `You are a helpful assistant that says a philosophical statement.`,
});
export const rootAgent = new LlmAgent({
name: 'root_agent',
// model: 'gemini-2.0-flash',
model: 'gemini-2.5-flash',
// model: 'gemini-3-flash-preview',
description: 'Root agent',
instruction: `You are a helpful assistant that is the root agent.`,
subAgents: [jokeAgent, philosophicalAgent],
});
tsconfig.json:
{
// Visit https://aka.ms/tsconfig to read more about this file
"compilerOptions": {
// File Layout
// "rootDir": "./src",
// "outDir": "./dist",
// Environment Settings
// See also https://aka.ms/tsconfig/module
"module": "nodenext",
"target": "esnext",
"types": [],
// For nodejs:
// "lib": ["esnext"],
// "types": ["node"],
// and npm install -D @types/node
// Other Outputs
"sourceMap": true,
"declaration": true,
"declarationMap": true,
// Stricter Typechecking Options
"noUncheckedIndexedAccess": true,
"exactOptionalPropertyTypes": true,
// Style Options
// "noImplicitReturns": true,
// "noImplicitOverride": true,
// "noUnusedLocals": true,
// "noUnusedParameters": true,
// "noFallthroughCasesInSwitch": true,
// "noPropertyAccessFromIndexSignature": true,
// Recommended Options
"strict": true,
"jsx": "react-jsx",
"verbatimModuleSyntax": false,
"isolatedModules": true,
"noUncheckedSideEffectImports": true,
"moduleDetection": "force",
"skipLibCheck": true,
}
}
package.json:
{
"name": "my-agent-22",
"version": "1.0.0",
"description": "",
"main": "agent.ts",
"scripts": {
"test": "echo \"Error: no test specified\" && exit 1"
},
"keywords": [],
"author": "",
"license": "ISC",
"type": "commonjs",
"devDependencies": {
"typescript": "^5.9.3"
},
"dependencies": {
"@google/adk": "^0.2.1",
"@google/adk-devtools": "^0.2.1"
}
}
I have re-created your setup @ScottMansfield and I run into the same issue as @ankargren.
Basic Infos:
node: v22.18.0
typescript: 5.9.3
OS: macOS 26.1 (25B78)
@google/adk: 0.2.1
google/adk-devtools: 0.2.1
I increased the log level in my agent.ts to get more logging.
import { LlmAgent, setLogLevel, LogLevel } from "@google/adk";
setLogLevel(LogLevel.DEBUG);
export const jokeAgent = new LlmAgent({
name: "tell_a_joke_agent",
// model: 'gemini-2.0-flash',
model: "gemini-2.5-flash",
// model: 'gemini-3-flash-preview',
description: "Tell a joke",
instruction: `You are a helpful assistant that tells a joke.`,
});
export const philosophicalAgent = new LlmAgent({
name: "philosophical_agent",
// model: 'gemini-2.0-flash',
model: "gemini-2.5-flash",
// model: 'gemini-3-flash-preview',
description: "Say a philosophical statement",
instruction: `You are a helpful assistant that says a philosophical statement.`,
});
export const rootAgent = new LlmAgent({
name: "root_agent",
// model: 'gemini-2.0-flash',
model: "gemini-2.5-flash",
// model: 'gemini-3-flash-preview',
description: "Root agent",
instruction: `You are a helpful assistant that is the root agent.`,
subAgents: [jokeAgent, philosophicalAgent],
});
I also use google cloud and vertex ai .env:
GOOGLE_GENAI_USE_VERTEXAI=TRUE
GOOGLE_CLOUD_PROJECT=my-project
GOOGLE_CLOUD_LOCATION=global
logs:
+-----------------------------------------------------------------------------+
| ADK Web Server started |
| |
| For local testing, access at http://localhost:8000. |
+-----------------------------------------------------------------------------+
[ADK INFO]: event: {"invocationId":"e-3ad8740a-0073-412e-af3e-e3456c4beaeb","author":"user","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"content":{"role":"user","parts":[{"text":"What can you do?"}]},"id":"kkr017JC","longRunningToolIds":[],"timestamp":1766147755261}
[ADK INFO]: Sending out request, model: gemini-2.5-flash, backend: VERTEX_AI, stream: false
[ADK DEBUG]: Skipping output save for agent root_agent: outputKey is not set
[ADK INFO]: event: {"invocationId":"e-81b3170f-79eb-4a85-a651-3655720d9a9e","author":"user","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"content":{"role":"user","parts":[{"text":"tell me a joke"}]},"id":"XhJOqmqU","longRunningToolIds":[],"timestamp":1766147764049}
[ADK INFO]: event: {"invocationId":"e-3ad8740a-0073-412e-af3e-e3456c4beaeb","author":"root_agent","id":"gJK07WSg","actions":{"stateDelta":{},"artifactDelta":{},"requestedAuthConfigs":{},"requestedToolConfirmations":{}},"longRunningToolIds":[],"timestamp":1766147755265,"content":{"role":"model","parts":[{"text":"I can tell jokes or philosophical statements. What would you like me to do?"}]},"usageMetadata":{"promptTokenCount":206,"candidatesTokenCount":16,"totalTokenCount":222,"trafficType":"ON_DEMAND","promptTokensDetails":[{"modality":"TEXT","tokenCount":206}],"candidatesTokensDetails":[{"modality":"TEXT","tokenCount":16}]},"finishReason":"STOP"}
[ADK INFO]: Sending out request, model: gemini-2.5-flash, backend: VERTEX_AI, stream: false
[ADK DEBUG]: Skipping output save for agent root_agent: outputKey is not set
[ADK DEBUG]: execute_tool transfer_to_agent
[ADK DEBUG]: callToolAsync transfer_to_agent
[ADK DEBUG]: traceToolCall {
tool: 'transfer_to_agent',
args: { agentName: 'tell_a_joke_agent' },
functionResponseEvent: 'gdcv9ID8'
}
[ADK DEBUG]: Skipping output save for agent root_agent: outputKey is not set
[ADK DEBUG]: Skipping output save for agent root_agent: outputKey is not set
[ADK DEBUG]: Skipping output save for agent root_agent: outputKey is not set
[ADK INFO]: Sending out request, model: gemini-2.5-flash, backend: VERTEX_AI, stream: false
[ADK DEBUG]: Skipping output save for agent tell_a_joke_agent: outputKey is not set
[ADK DEBUG]: Skipping output save for agent root_agent: event authored by tell_a_joke_agent
Since I am currently having issues with a larger agent orchestration, I wanted to check whether the problem lies within Gemini itself or the ADK.
I recreated the same example using the Python ADK, and everything works! Since I am not very fluent in Python—and even Gemini-CLI had trouble recreating my TypeScript agent with the Python ADK—I hope someone can identify and fix the issue. 👍
from google.adk.agents.llm_agent import Agent
joke_agent = Agent(
model='gemini-3-flash-preview',
name='joke_agent',
description='A funny assistant that tells jokes.',
instruction='Tell a funny joke in response to user prompts',
)
philosophy_agent = Agent(
model='gemini-3-flash-preview',
name='philosophy_agent',
description='Say a philosophical statement.',
instruction='You are a helpful assistant that says a philosophical statement.',
)
root_agent = Agent(
model='gemini-3-flash-preview',
name='root_agent',
description='Root agent',
instruction='You are a helpful assistant that is the root agent.',
sub_agents=[joke_agent, philosophy_agent],
)