What this skill does
Generates a 1-page ADR (Architecture Decision Record): context, options considered, trade-offs, the call, risks, rollback plan. So that in 6 months, when someone asks, you still know why you chose X.
ADR structure
- Status: proposed / accepted / superseded
- Context: the problem, in 3-5 lines
- Options considered: 2-4 options with pros and cons
- Decision: what you picked, in 1 sentence
- Reasoning: why this one and not the others
- Accepted risks: what will be bad, and why that is OK
- Rollback plan: how to get out if it goes wrong
When to use
- Before picking a database / framework / important SaaS
- Before changing architecture (monolith → micro / SQL → NoSQL)
- When you are about to make a choice that will hurt to reverse
How it works
- You describe the problem in 1-3 sentences
- The skill asks 4 key questions: deadline, cost, lock-in, does the team know it?
- Suggests 3 options to evaluate (based on the problem you described)
- For each one: pros, cons, estimated cost, lock-in level
- Delivers the ADR already filled in with your choice + reasoning
- Includes an explicit rollback plan
Output example
ADR-007: Database for purchase storage
Status: proposed
Date: 2026-05-06
Owner: Ivan
Context
Today purchases are logged with console.log (Vercel logs only).
Expected volume: 50-200 purchases/month. Needs: query by
email, audit trail (refunds), 5-year retention (LGPD).
Options considered
1. Supabase Postgres (free tier 500MB)
+ Familiar, RLS out of the box, versioned migrations
- +1 SaaS to maintain, US-East region latency
2. Vercel KV (Redis)
+ Same provider, low latency, simple
- Not relational, queries by email cost a scan
3. JSON file on Vercel Blob
+ Zero infra
- Does not scale, race conditions, no queries
Decision
Supabase Postgres free tier.
Reasoning
Query volume (by email, by date range) needs SQL. The free tier
covers 5+ years of the expected volume. Migrating to paid
(R$ 25/month) is trivial.
Accepted risks
- ~200ms latency for BR users (fine for audit, not for UI)
- Moderate lock-in (Supabase-specific RLS)
Rollback plan
If Supabase fails / gets expensive: migration script to
self-hosted Postgres on Hetzner (R$ 30/month). The
/scripts/migrate-pg/ folder already has a template.
Estimated: 4h of work.Pre-reqs
None.
