replicating-vibe-search

name	replicating-vibe-search
description	Orchestrates the replication, environment provisioning, container builds, and deployment of the Shutter Vibe Engine on Google Cloud. Use when the user requests deployment, replication, or troubleshooting of the multimodal vibe search system.

Replicating the Shutter Vibe Engine

When to use this skill

• Fresh Bootstrapping: Setting up the Shutter Vibe Engine in a brand-new, empty workspace.
• GCP Resource Provisioning: Creating GCS buckets, binding IAM roles, or enabling APIs.
• Continuous Deployment: Building and deploying the Frontend Web App or Ingestion Worker containers.
• Pipeline Testing & Logs: Running end-to-end diagnostics and tailing container logs.

---

🛠️ The Orchestrator's Pre-Flight Checklist

Before executing any // turbo blocks from the workflow, verify the following prerequisites:

• Python Package Manager: Ensure uv is installed and active.
• Active gcloud CLI: Check that the developer is logged in:

  gcloud auth list
  

• Workspace Parameter Check: Confirm .env exists in the workspace root and contains valid project parameters (and is strictly omitted from Git via .gitignore).

---

🤖 Step-by-Step Deployment Runbook

1. Project Initialization & Sourcing

Make sure the target paths match original directories. The workflow // turbo block in Phase 1 executes the copy. Verify that the files exist in:

• multimodal-search/app/main.py
• multimodal-search/app/templates/index_v2.html
• multimodal-search/app/static/

2. Idempotent Environment Provisioning

When configuring APIs and Service Accounts, always ensure actions are idempotent (check existence first):

• API Verification:

  gcloud services list --enabled --filter="name:(run.googleapis.com OR eventarc.googleapis.com)"
  

• Service Account Check:

  gcloud iam service-accounts describe ${ENVATO_SA_NAME}@${GOOGLE_CLOUD_PROJECT}.iam.gserviceaccount.com
  

3. Parallelized Cloud Container Builds

We compile images using Google Cloud Build with static YAML parameters to prevent local docker engine dependencies:

• Web App Builder: Runs antigravity/deploy/cloudbuild.app.yaml.
• Worker Builder: Runs antigravity/deploy/cloudbuild.ingest.yaml.

Always swap the appropriate .gcloudignore files before compiling to prevent copying heavy .venv folders or temporary local test artifacts!

---

🔍 Validation & Live Diagnostics

Local Verification

Run the verification harness using uv run to assert that all template paths are complete and variable substitutions are complete:

uv run python antigravity/verify_replication.py

Eventarc Pipeline Trigger Testing

After successful container deployments, instruct the developer to perform a live ingestion test:

1. Ingestion Command:

   gcloud storage cp some-sample.mp4 gs://${ENVATO_GCS_BUCKET}/ingest/
   

2. Monitor Processing Output:

   gcloud run services logs read envato-vibe-ingest --region=${GOOGLE_CLOUD_LOCATION} --limit=50
   

3. Inspect Results: Check Firestore for the generated embeddings index entry.

---

🛑 Common Failure Troubleshooting

• API Enablement Latency: Occasionally, newly enabled APIs can take up to 2 minutes to propagate. If a deployment fails immediately after enabling an API, wait 60 seconds and re-run.
• Eventarc Service Agent Permissions: If files are finalized in GCS but the ingest worker isn't triggered, check that the GCS service account has roles/pubsub.publisher assigned.
• Memory Allocation Crash: FFmpeg ingestion can be memory-intensive. Ensure the ingest worker is deployed with --memory=2Gi or higher.