Crear un agente de IA en 3 clics con Amazon Bedrock AgentCore Harness

Eres un asistente de investigación de lanzamientos de AWS.

Usa el navegador para consultar la documentación oficial de AWS y sus blogs.

En cada respuesta:
1. Indica el estado actual del lanzamiento.
2. Explica qué cambió.
3. Incluye las URLs de las fuentes que usaste.
4. Aclara cuándo la evidencia disponible no respalda una afirmación.

No uses fuentes de terceros.

import os
import time
import uuid
import boto3

REGION = os.getenv("AWS_REGION", "us-west-2")
ROLE_ARN = os.environ["HARNESS_EXECUTION_ROLE_ARN"]
MODEL_ID = os.environ["BEDROCK_MODEL_ID"]

SYSTEM_PROMPT = """
You are an AWS release research assistant.

Use the browser to consult official AWS documentation and AWS blogs.

For every answer:
1. State the current launch status.
2. Explain what changed.
3. Include the source URLs you used.
4. Say when the available evidence does not support a claim.

Do not use third-party sources.
""".strip()

control = boto3.client(
"bedrock-agentcore-control",
region_name=REGION,
)

runtime = boto3.client(
"bedrock-agentcore",
region_name=REGION,
)

created = control.create_harness(
harnessName="aws_release_researcher",
executionRoleArn=ROLE_ARN,
model={
"bedrockModelConfig": {
"modelId": MODEL_ID,
}
},
systemPrompt=[
{
"text": SYSTEM_PROMPT,
}
],
tools=[
{
"type": "agentcore_browser",
"name": "browser",
}
],
maxIterations=12,
timeoutSeconds=300,
)

harness = created["harness"]

while harness["status"] == "CREATING":
time.sleep(3)
harness = control.get_harness(
harnessId=harness["harnessId"],
)["harness"]

if harness["status"] != "READY":
raise RuntimeError(
f"Harness creation failed: {harness.get('failureReason', 'unknown error')}"
)

response = runtime.invoke_harness(
harnessArn=harness["arn"],
runtimeSessionId=str(uuid.uuid4()),
messages=[
{
"role": "user",
"content": [
{
"text": (
"When did Amazon Bedrock AgentCore Harness become "
"generally available, and what was added at GA?"
)
}
],
}
],
)

for event in response["stream"]:
if "contentBlockDelta" in event:
delta = event["contentBlockDelta"].get("delta", {})
if "text" in delta:
print(delta["text"], end="", flush=True)

elif "runtimeClientError" in event:
raise RuntimeError(event["runtimeClientError"]["message"])

AWSTemplateFormatVersion: "2010-09-09"
Description: AWS release research agent built with AgentCore Harness

Parameters:
HarnessExecutionRoleArn:
Type: String

BedrockModelId:
Type: String

Resources:
ReleaseResearcherHarness:
Type: AWS::BedrockAgentCore::Harness
Properties:
HarnessName: aws_release_researcher
ExecutionRoleArn: !Ref HarnessExecutionRoleArn

Model:
BedrockModelConfig:
ModelId: !Ref BedrockModelId
Temperature: 0.2

SystemPrompt:
- Text: |
You are an AWS release research assistant.

Use the browser to consult official AWS documentation
and AWS blogs.

For every answer:
1. State the current launch status.
2. Explain what changed.
3. Include the source URLs you used.
4. Say when the available evidence does not support a claim.

Do not use third-party sources.

Tools:
- Type: agentcore_browser
Name: browser

Memory:
ManagedMemoryConfiguration:
Strategies:
- SEMANTIC
- SUMMARIZATION
EventExpiryDuration: 30

MaxIterations: 12
TimeoutSeconds: 300

Tags:
- Key: application
Value: aws-release-researcher
- Key: managed-by
Value: cloudformation

response = runtime.invoke_harness(
harnessArn=HARNESS_ARN,
runtimeSessionId=str(uuid.uuid4()),
systemPrompt=[
{
"text": (
"You are an AWS release research assistant. "
"Use only official AWS sources. "
"Return at most three paragraphs and include the URLs."
)
}
],
maxIterations=6,
messages=[
{
"role": "user",
"content": [
{
"text": "Explain what AgentCore Harness manages."
}
],
}
],
)

# Inside your AWS::BedrockAgentCore::Harness resource
MaxIterations: 12
MaxTokens: 8000
TimeoutSeconds: 300

Tags:
- Key: application
Value: aws-release-researcher
- Key: environment
Value: production
- Key: owner
Value: platform-team
- Key: cost-center
Value: genai