Wendy Gcp Deployment by wendylabsinc — AI Skill for Any LLM

Overview

Publisher	wendylabsinc
Repository	`claude-skills`
Skill name	`wendy-gcp-deployment`
Stars	62
Forks	5
Bundled files	5
Links	Skill source Repository

Markdown instructions

A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

Works with any LLM

AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

5 bundled files

Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

Open source

Published by wendylabsinc on GitHub. Read the source before you install it.

Installation

Install the Wendy Gcp Deployment AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

Open the app and go to Plugins → Skills.
Choose "Install from GitHub".
Paste the skill folder URL below and confirm.
https://github.com/wendylabsinc/claude-skills/tree/main/wendy-gcp-deployment
Enable the skill in any chat where you want it available.

Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills

git clone --depth 1 https://github.com/wendylabsinc/claude-skills.git /tmp/claude-skills
mkdir -p .claude/skills
cp -r /tmp/claude-skills/wendy-gcp-deployment .claude/skills/wendy-gcp-deployment

Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Wendy Gcp Deployment in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Wendy Gcp Deployment on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Wendy Gcp Deployment is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

Loaded only when it is needed

The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

Switch models mid-chat

Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Wendy GCP Deployment

How Wendy deploys to Google Cloud. Status: intermediate. These conventions hold until v1 of the stack runs and the compute workload is consolidated (possibly k8s/GitOps later). The methodology (Pulumi, OIDC, PoLP) is expected to survive that transition; specific names may not.

Hard rules (non-negotiable)

Requirements before infrastructure. Never design infra without asking the user about the app's requirements first (see interview below). Never invent project names, regions, or scale assumptions.
Cost-consciousness is a MUST, not a preference. Read references/cost.md before choosing compute or a database.
PoLP via custom IAM roles. Prefer a custom role with the exact permissions needed over any predefined broad role. Needing to alter one DNS record never justifies roles/dns.admin. See references/iam-polp.md.
No service account keys. Ever. CI authenticates via GitHub OIDC / Workload Identity Federation (references/github-oidc.md); workloads use attached service accounts. If you are about to write credentials_json, export a key, or suggest one "as a stopgap" — stop; that path does not exist here.
No GCP credentials client-side. Frontends, device apps, anything running outside Wendy's server perimeter never holds GCP credentials of any form. Devices get mediated access via server-side proxies minting short-lived downscoped credentials (wendy-proxy pattern). Authority: the AAA contract, /home/sem/wendy/aaa-contract-*.md (read the latest version when touching auth between services).
dev/prod split by default. Default branch deploys to dev; semver tags (v*) deploy to prod. Skip the split only with a stated good reason. Today each repo keeps every env in one GCP project, split by resource naming (-dev / -prod suffixes): pki-core wendy-pki-secure, cloud cloud-c7e56 (incl. legacy-prod), wendy-auth wendy-auth (its envs are test / prod). Source: gcp:project (wendy-auth foundation: projectId) in each Pulumi.<stack>.yaml, foundation and platform. Every v* tag is a GitHub Release with generated notes. pki-core, wendy-auth and cloud additionally split each env into a foundation stack (network, DNS zones, custom IAM roles; applied by hand) and a platform stack (applied by CI, whose deploy SA is never widened) — see references/iam-polp.md.
Never clash with pre-existing resources. Check before creating — especially DNS records (references/dns.md) and anything in shared projects.

Requirements interview (always first)

Before proposing any design, ask the user (don't assume):

Traffic shape and volume; public or internal?
State: does it truly need a database? What kind/size?
Latency and availability requirements (is scale-to-zero cold start acceptable?)
Which project does it belong in (see map below)? Dev/prod needed?
Anything that must survive redeploys (volumes, IPs, data)?

Shape the design to the answers. When requirements are modest, the infrastructure must be too.

Project map (current truth — verify with `gcloud projects list` if in doubt)

Project	Purpose
`wendy-customer`	All customer-facing apps (apps not part of the cloud stack). Per-app isolation inside this project matters: own SA, own custom role, own resources per app.
`wendy-auth`	wendy-auth
`wendy-pki-secure`	pki-core engine
`wendy-pki-services`	pki-core frontends
`cloud-c7e56`	SaaS cloud stack; hosts the `.wendy.sh` and `.wendy.dev` DNS zones (except pki-core's delegated `pki.wendy.dev` zone — see `references/dns.md`)

Regions are set per repo in each stack's Pulumi yaml (<project>:region). Read it there; don't assume one default:

pki-core: europe-west4 (every foundation and platform stack, dev and prod).
cloud, wendy-auth: us-central1 for compute (every stack, including foundation).
PSC: the pki-core fabric producer (service attachment) is in europe-west4. The consumer endpoints in cloud and wendy-auth are created in the attachment's region, europe-west4, and use global access so us-central1 Cloud Run can reach them. That is a fact about PSC placement, not the compute region of those repos.

Changing a repo's region needs a real use case — ask, don't switch unilaterally.

Compute selection

Cloud Run first. Scale-to-zero, request-based billing, no OS to maintain. Default for services and jobs.
VMs are leaned against, for one specific reason: once spun up, nobody maintains the OS. A VM is acceptable only if the OS is completely maintenance-free (e.g. Container-Optimized OS with automated updates, MIGs with automated image rotation). If the plan involves a human ever SSHing in to patch it, it's not an option.
Databases: the cheapest thing that meets requirements. Read references/cost.md — it exists because of a real incident (expensive Cloud SQL clusters provisioned for apps that didn't need them).

Network shape (preferred for bigger projects)

Source of truth: Linear doc Wendy Network Architecture (WDY-2840, approved 2026-09-03). Anything with multiple services, VMs, or a database gets this shape.

Carve-out: a single, fully-native deployment (e.g. Firebase Hosting, a lone Cloud Run service) does not carry the strict VPC requirements below. Dual-stack (rule 6) remains required even there.

Per-project custom-mode VPCs, deliberately separate — authority isolation. Never the auto-mode default VPC; delete it from projects we control.
Cross-project service traffic only via Private Service Connect — one published service, one direction, consumer-allowlisted. Never VPC peering (firewall rules can't reference peer SAs/tags; CIDR coupling).
Databases/caches private-only. Cloud SQL: no public IP, sslMode: ENCRYPTED_ONLY, PSC preferred. Valkey/Memorystore: transit encryption + AUTH.
Egress deny-by-default, with per-service-account egress firewall rules (v4/v6 symmetric). Every service gets an explicit egress class: zero / allowlist / broad-443. Cloud NAT for v4 egress; a service granted v6 egress carries external v6 + deny-all ingress.
No public IPs on VMs. Ingress only through LB forwarding rules; operator access via IAP, never public SSH/RDP.
Dual-stack (IPv6+IPv4) is mandatory for everything we build — LB frontends, MIGs, VMs. Internal v6 is ULA (ipv6AccessType=INTERNAL) by default; AAAA records on every public host. Only exception: a written one, for Google-managed endpoints with no v6 path (Cloud SQL, Memorystore internals).
Server certs: endpoints reachable by default-trust clients (browsers, generic tooling) use Let's Encrypt (lego DNS-01); endpoints called exclusively by wendy software may use pki-core-issued certs. The exact machine-endpoint policy is still being finalized on WDY-2840 — apply the principle, don't hardcode per-host assignments.

Deployment workflow

Requirements interview (above).
Design: compute, data, network shape (above), IAM (custom roles per references/iam-polp.md), DNS (references/dns.md), dev/prod stacks.
Bootstrap privileged resources (deploy SA, custom role, WIF binding) via the human's own gcloud, with their approval per command. If the human lacks permission: produce the exact commands + a use-case writeup to hand to an admin (template in references/iam-polp.md).
Pulumi program — Go preferred, GCS state backend, KMS secrets provider, conventions in references/pulumi.md.
GitHub Actions workflow — OIDC only, per references/github-oidc.md. Also invoke the github-actions skill when writing the workflow.

Rationalizations that don't fly

Excuse	Reality
"roles/dns.admin scoped to the zone is minimal enough"	Cloud DNS supports per-RRset IAM conditions. Zone-wide admin is not the floor; the exact records are.
"A predefined role is simpler than a custom role"	Simpler to type, larger to breach. Custom roles are a one-time cost; see the recipe.
"credentials_json just to get it working"	A leaked key is a standing credential. WIF setup takes minutes; there is no keys-temporarily path.
"It'll need Cloud SQL HA when it grows"	Provision for today's requirements. Growth is a config change, not a sunk cost.
"Skipping dev, it's a small app"	Small apps get semver-tagged prod deploys too. Skipping the split is a decision the user makes, with a reason.
"The zone probably has no record with this name"	`gcloud dns record-sets list` takes two seconds. Check.
"VPC peering is simpler than PSC"	Peered firewalls can't reference the peer's SAs/tags, and CIDRs couple. PSC publishes one service, one direction, allowlisted.
"IPv4-only for now, v6 later"	Dual-stack is mandatory from day one. The only exception is a written one for Google-managed endpoints with no v6 path.

Red flags — stop and reconsider

You wrote roles/*.admin for a deploy SA
You wrote credentials_json or gcloud iam service-accounts keys create
A GCP credential (of any form) ends up in frontend code, a device image, or an app config shipped outside Wendy's servers
You picked a region, tier, or project name the user never confirmed
Your design has no -dev stack and the user never said why
You're about to create a DNS record without listing existing ones first
Your design uses the default VPC, VPC peering, or a public IP on a VM or database
An LB frontend, MIG, or public host is IPv4-only with no written exception

Frequently asked questions

What does the Wendy Gcp Deployment AI skill do?

Expert guidance on deploying Wendy applications and infrastructure to Google Cloud. Use when: (1) deploying anything to GCP, (2) writing or changing Pulumi programs, (3) creating GCP service accounts, IAM roles, or custom roles, (4) touching wendy.sh / wendy.dev DNS records, (5) choosing GCP compute or databases, (6) setting up CI/CD deployment to GCP, (7) provisioning Cloud Run, Cloud SQL, or Artifact Registry.

Why use Wendy Gcp Deployment on TypingMind?

Because you install it once and use it with any model. Wendy Gcp Deployment is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Wendy Gcp Deployment in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/wendylabsinc/claude-skills/tree/main/wendy-gcp-deployment. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Wendy Gcp Deployment?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Wendy Gcp Deployment?

As many as you like. As long as a model supports skills, you can use Wendy Gcp Deployment with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Wendy Gcp Deployment AI skill free?

It is published on GitHub by wendylabsinc. Check the repository for licensing terms. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

Overview

Installation

Install in TypingMind

Install in another agent

Use it in TypingMind

Works with any AI model

Skill instructions

Wendy GCP Deployment

Hard rules (non-negotiable)

Requirements interview (always first)

Project map (current truth — verify with gcloud projects list if in doubt)

Compute selection

Network shape (preferred for bigger projects)

Deployment workflow

Rationalizations that don't fly

Red flags — stop and reconsider

Bundled files

Frequently asked questions

Related AI Skills

Set up your own AI workspace now

Project map (current truth — verify with `gcloud projects list` if in doubt)