How this aligns with Google’s behavior (key points)

Push delivery is wrapped JSON by default: endpoint receives POST with { message: { data, attributes, ... } }, and data is base64. Can also opt into payload unwrapping if you prefer raw body; this template assumes the default wrapped format.

GCS→Pub/Sub notifications put bucket/object info in message.attributes (e.g., eventType, bucketId, objectId, objectGeneration) and can also include a JSON representation of the object in message.data (when payloadFormat=JSON_API_V1).

We read the exact object generation for idempotency and correctness; generation changes whenever content changes (metageneration changes on metadata updates).

The Storage client can pin reads to a generation (and even enforce if_generation_match); that’s what this template uses.

For auth, a push subscription can attach an OIDC ID token (Bearer) and Cloud Run can validate it automatically when you restrict the service and grant run.invoker to the push identity; manual verification is optional.

Handler should return 2xx (commonly 204) on success to ack the message; non-2xx triggers Pub/Sub retries with backoff.

Local Testing

Env Setup

pip install -r requirements.txt
uvicorn app.main:app --reload --port 8080

Send a sample Pub/Sub push payload (wrapped):

curl -s -X POST http://localhost:8080/ -H "Content-Type: application/json" -d  @data.json

Env Variables

COMPONENT_NAME=ai-scorer

EXPECTED_EVENT_TYPE=OBJECT_FINALIZE

OBJECT_PREFIX=reports/ (if you want to scope)

OUTPUT_PREFIX=outputs/ai-scorer/

REQUIRE_JWT=true and PUBSUB_ALLOWED_AUDIENCE=<your push endpoint URL> if you want in-app token verification (optional; Cloud Run can enforce it without this).

Steps for creating an AI Agent

This set up assumes

• We are creating an event based trigger for the AI agent
• Event based trigger uses PUB SUB as the techincal implementation
• The code is python based

This set up creates

• All necessary infrastructure required for the agent on GCP
• Uses Pub sub
• Assumes code is written in Python

Run the following bat file

create-agent.bat

Steps for removing the AI Agent created above

Run the following bat file

teardown.bat

Misc files

smoke.json - Empty json file for testing the functionality