Feature Request: Cloud Pub/Sub Skill

[unrealandychan]

Feature Request: Cloud Pub/Sub Skill

Summary

Cloud Pub/Sub is the backbone of event-driven and asynchronous architectures on Google Cloud. Existing skills cover individual services (BigQuery, Cloud Run, GKE) but do not explain how to connect them via Pub/Sub — leaving agents unable to design reliable messaging pipelines.

The Gap

A typical agent workflow today:

User asks 'How do I trigger a Cloud Run job when a file lands in a bucket?' → Agent may answer with Cloud Functions or direct Eventarc, but misses Pub/Sub as the preferred decoupled alternative for high-throughput or multi-consumer scenarios.

There is no skill covering:

• Topic vs subscription architecture (fan-out, fan-in, dead-letter queues)
• Push vs Pull delivery semantics and retry policies
• Ordering keys, exactly-once delivery, and message deduplication
• BigQuery subscription for direct ingestion without intermediate compute
• Schema validation (Avro / Protobuf) for strongly typed message contracts
• Flow control, batching, and throughput tuning
• Security: Pub/Sub IAM, VPC-SC, CMEK encryption

Proposed Skill

A pubsub-basics skill that agents load when users mention: Pub/Sub, event-driven, message queue, topic, subscription, async pipeline, streaming ingestion, dead letter queue, or BigQuery subscription.

Suggested SKILL.md frontmatter

---
name: pubsub-basics
description: >
  Use when the user asks about event-driven messaging, asynchronous pipelines,
  decoupling services, or streaming data ingestion on Google Cloud. Covers Cloud Pub/Sub
  topics, subscriptions (push/pull/BigQuery), message ordering, exactly-once delivery,
  dead-letter queues, schema validation, IAM, and integration with Cloud Run, GKE,
  Dataflow, and BigQuery. WHEN: create topic, create subscription, event-driven,
  message queue, async job, dead letter, BigQuery subscription, Pub/Sub schema.
compatibility: Requires pubsub.editor IAM role and Cloud Pub/Sub API enabled.
---

Key reference topics

1. Core Concepts
   • Publisher → Topic → Subscription(s) → Subscriber
   • Fan-out (multiple subscriptions per topic) vs fan-in (single subscription per topic)
   • Message retention: default 7 days, configurable up to 31 days
2. Subscription Types
   • Pull: Manual ack, batch processing, GKE / compute workloads
   • Push: HTTP POST to endpoint, Cloud Run / Cloud Functions
   • BigQuery: Direct streaming write without code, schema mapping
   • Cloud Storage: Export messages to GCS (integration with analytics)
3. Reliability Patterns
   • Dead-letter topics (DLQ) and max delivery attempts
   • Retry policies: immediate vs exponential backoff
   • Ordering keys: when to use, throughput limitations
   • Exactly-once delivery: semantics, prerequisites, idempotency
4. Schema & Governance
   • Avro / Protobuf / JSON schema definitions
   • Schema evolution and backward compatibility
   • Constraining publishers to enforce contracts
5. Performance & Cost
   • Flow control (max outstanding messages / bytes)
   • Batching and compression for high throughput
   • Pricing model: ingestion + egress + storage
6. Security
   • IAM roles: publisher, subscriber, viewer
   • VPC Service Controls for private-only access
   • CMEK encryption at rest

Why Now?

• Pub/Sub is implicitly required by at least 4 existing skills (Cloud Run event triggers, BigQuery streaming, GKE async jobs, Dataflow ingestion) but never taught as a first-class skill.
• BigQuery Subscription is a powerful zero-code pattern that is under-documented across the ecosystem.
• The rise of agentic workflows increases demand for reliable async handoff between tasks; Pub/Sub is the canonical GCP answer.

Reference Implementation

Google Cloud official docs:

• https://cloud.google.com/pubsub/docs
• https://cloud.google.com/pubsub/docs/bigquery
• https://cloud.google.com/pubsub/docs/handling-failures

Happy to contribute a SKILL.md draft if this direction is accepted.