k-aws-cost

Files (1)

• SKILL.md

SKILL.md (first lines)

> [!info] In force on this repo (Team Party map)
> Layout [[adr-25-harness-layout]]: project stack skill (stem names the pinned stack taught).
> Product SSOT: `docs/constitution/PRD.md`. Canonical host: `{{PUBLIC_HOST}}`.
> Citations: [[adr-24-constitution]], [[adr-25-harness-layout]], [[adr-27-guardians-and-delivery]].
> Live AWS names that still contain `astro-drf-aws` are leftover until Gabriel reprovisions — do not invent ARNs.


# k-aws-cost

## Non-negotiable cost opinions (template)

| Decision | Why |
|----------|-----|
| **No NAT Gateway** | Tasks use public IPs — deliberate (`docs/INFRASTRUCTURE.md`) |
| **No Redis / ElastiCache** | `DatabaseCache` only (`docs/CACHE.md`) |
| **Shared ALB** per env | Not one ALB per project |
| **Fargate 256 CPU / 512 MB** baseline | Scale only when measured |
| **desiredCount: 1** | Until load proves otherwise |
| **RDS `db.t4g.micro` single-AZ** | `alvs-*-pg` precedent |
| **dev + prod only** | No staging tier |
| **Log retention short** | 14–30 days default |

Any proposal that adds NAT, Redis, Multi-AZ RDS “because production”, or second ALB must be **rejected** unless the user overrides in-turn with explicit justification.

## When the agent must refuse

- “Move tasks to private subnets + NAT for security”  
- “Add ElastiCache for Django cache”  
- “Multi-AZ RDS for the template default”  
- “Fargate 2 vCPU because why not”  
- “Always-on NAT for ECR pulls” (use public IP + ECR endpoints optional later)

## Allowed cost work

- Cost Explorer by service/tag for account `{{AWS_ACCOUNT_ID}}`  
- Right-size **after** CloudWatch CPU/mem evidence (`k-aws-observability`)  
- Budgets/alerts for unexpected spend  
- ECR lifecycle policies to expire untagged images  
- Spot **not** for this SSR pair (stick to Fargate standard)

## Math rule

Never freehand arithmetic on cost figures — use a short Python script or CLI output.

## Date rule

Before Cost Explorer queries, use the **real current date** (today’s year is not training data).

## Related

`k-aws-containers` · `k-aws-observability` · `docs/INFRASTRUCTURE.md` · `docs/CACHE.md`

From kodexArg/astro-drf-aws, MIT.

More from kodexArg/astro-drf-aws