feat(aws-ec2-instance-type): add AWS EC2 instance type parameter module

[phorcys420]

AWS EC2 Instance Type

A coder_parameter helper that presents AWS EC2 instance types with human-readable labels (vCPU, RAM, architecture, and GPU count), filtered by instance family. Follows the same pattern as coder/aws-region.

Specs come straight from the AWS EC2 API (aws ec2 describe-instance-types); .scripts/update.sh regenerates instance-types.json, so vCPU/RAM/GPU/arch are never hand-maintained.

Usage

module "aws_ec2_instance_type" {
  count   = data.coder_workspace.me.start_count
  source  = "registry.coder.com/coder/aws-ec2-instance-type/coder"
  version = "1.0.0"
  include = ["t3", "m5"]
  default = "t3.medium"
}

resource "aws_instance" "dev" {
  instance_type = module.aws_ec2_instance_type[0].value
}

Behavior

• Family filter — include (default ["t3"], validated non-empty) selects which families appear in the picker. Bundled catalog: t3, t4g, m5, m7g, c5, r5, i3, g4dn.
• Readable options — each option is named "<vCPU> vCPU, <RAM> GiB RAM[, <n> GPU] (<arch>)" with the instance type as its tooltip. Labels shared across families are disambiguated with the instance type.
• Architecture aware — the instances output exposes arch (x86_64/arm64) and coder_arch (amd64/arm64) per type, so templates can pick a matching AMI and set coder_agent.arch.
• Overrides — custom_metadata overrides an option's name/description per instance type.
• Optional picker — create_parameter = false skips the coder_parameter (returns default) while keeping the instances catalog.

Outputs

• value — the selected instance type (or default when the picker is disabled).
• instances — the full catalog keyed by instance type (vcpus, memory_mib, gpus, arch, coder_arch).

Every type used by the AWS templates (t3.micro–t3.2xlarge, used by aws-linux/aws-windows) is in the default t3 family; a test pins that coverage.

Checks

terraform fmt, terraform validate, terraform test (13/13), prettier --check, and shellcheck on .scripts/update.sh. The dynamic-parameters render was captured from a live Coder instance (see preview).

Design decisions

• Family-based include over categories — earlier drafts used purpose categories (general/compute/…) with an exclude list. Replaced with a flat family include (default t3): simpler to reason about and maps directly to how AWS names instances.
• Specs sourced from AWS — .scripts/update.sh enumerates sizes per family via describe-instance-types with an inline --query that emits the final JSON shape (no jq), so new sizes are picked up automatically. Curated family list rather than the entire EC2 catalog.
• JSON shape — each entry is { type, vcpus, memory_mib, gpus, arch }; coder_arch is derived in Terraform (AWS uses x86_64, Coder uses amd64).
• custom_metadata — single map(object({ name = optional(string), description = optional(string) })) replacing separate custom_names/custom_descriptions. Requires Terraform >= 1.3 for optional().
• file() fallback — jsondecode(try(file("${path.module}/instance-types.json"), file("instance-types.json"))), because Coder's dynamic-parameters preview resolves file() from the module dir differently than real Terraform (matches merged aws-region).
• Tests — native .tftest.hcl only (13 runs). No main.test.ts: this is a parameter-only module with no runtime script to exercise end to end.
• Metadata — verified: false, initial version 1.0.0.

Preview

---

🤖 Opened by Coder Agents on behalf of @phorcys420.

[github-actions]

Module Scorecard Check

coder/aws-ec2-instance-type: first scorecard, 100 / 100

No existing scorecard discussion found for AWS EC2 Instance Type; this is the initial score. A dedicated discussion is created after merge.

Full scorecard for this PR

Presentation & Onboarding	Credential Hygiene	Restricted-Environment Readiness	Engineering Quality	Overall
17 / 17	20 / 20	N/A	10 / 10	100 / 100

Drilldown

Presentation & Onboarding — 17 / 17

Criterion	Max	Score	Notes
Configuration-mode examples	12	12	README documents five distinct modes: default t3-only picker, multi-family include, custom_metadata label overrides, architecture-aware provisioning via the instances output, and create_parameter = false for catalog-only use. Each example shows sensible defaults and is self-contained.
Visual preview	5	5	 is embedded in the README; file verified to exist (80.6 KB).

Credential Hygiene — 20 / 20

Criterion	Max	Score	Notes
Secrets marked sensitive	16	16	No sensitive inputs exist in this module (all variables are display names, instance-type strings, booleans, or metadata maps). README examples contain no inline secrets or placeholder keys.
Non-hardcoded auth path	4	4	The module performs no authentication; it creates a Coder parameter and reads a local JSON catalog. No keys, tokens, or credentials are involved, so the criterion is vacuously satisfied.

Restricted-Environment Readiness — N/A

The module downloads nothing, installs nothing, and contacts no external endpoints at plan or apply time. It reads a local instance-types.json file and creates a coder_parameter. The .scripts/update.sh is a developer maintenance tool (run manually to regenerate the catalog), not part of the module's runtime. Per the rubric, "a module that downloads nothing" is the canonical N/A example for this theme. All four sub-criteria (Mirrorable artifact source, Bring-your-own binary, Egress transparency, Runs without sudo) are excluded.

Engineering Quality — 10 / 10

Criterion	Max	Score	Notes
Input quality	6	6	All eight variables carry clear description strings and sensible defaults. include has a validation block rejecting empty lists. custom_metadata uses a typed map(object({...})) with optional fields.
Test coverage	4	4	main.tftest.hcl contains 12 run blocks covering parameter naming, order propagation, default/fallback values, create_parameter = false path, family filtering (single and multi), label disambiguation, custom_metadata overrides, instances output shape (vcpus, memory_mib, gpus, coder_arch derivation for x86_64→amd64 and arm64), and value-key consistency. No TypeScript layer exists, so no e2e TS tests are expected.

Overall — 100 / 100

Raw 47 / 47 → round(47 / 47 × 100) = 100

Track: Utility (parameter/helper module; no agent, IDE, or application component)

Tip

You can run this locally by telling your agent: "review this module against .github/scorecard/SCORECARD.md".

---

Scored against SCORECARD.md with solstice-1. Language-model scores are advisory.

[DevelopmentCats]

LGTM I will load up into my deployment as well to test it out but I cant imagine it wouldn't work 😸

[phorcys420]

ive been testing it like this if that helps

module "aws-ec2-instance-type" {
  source = "git::https://github.com/coder/registry.git//registry/coder/modules/aws-ec2-instance-type?ref=phorcys/aws-ec2-instance-type&depth=1"

  default = "t3.medium"
  description = trimspace(<<-EOT
    t3.medium is the smallest that works: the NixOS AMI configures no swap and
    the Nix store shares the root volume, so a rebuild that has to compile
    anything will exhaust a 1-2 GiB instance.
  EOT
  )
}

[DevelopmentCats]

Works fine for me I will let you decide when to merge and release since I see you are still committing 😄

[phorcys420]

i'm trying to get that scorecard up hahaha