Architecture Decision Records (ADR)

Purpose
When to Use This Skill
What is an ADR?
Workflow
Common Patterns
Guardrails
Quick Reference

Purpose

Document significant architectural and technical decisions with full context, alternatives considered, trade-offs analyzed, and consequences understood. ADRs create a decision trail that helps teams understand "why" decisions were made, even years later.

When to Use This Skill

Recording architecture decisions (microservices, databases, frameworks)
Documenting infrastructure choices (cloud providers, deployment strategies)
Capturing technology selections (libraries, tools, platforms)
Logging process decisions (branching strategy, deployment process)
Establishing technical standards or conventions
Migrating or sunsetting systems
Making security or compliance choices
Resolving technical debates with documented rationale
Onboarding new team members who need decision history

Trigger phrases: "ADR", "architecture decision", "document this decision", "why did we choose", "decision record", "technical decision log"

What is an ADR?

An Architecture Decision Record is a document capturing a single significant decision. It includes:

Context: What situation necessitates this decision?
Decision: What are we choosing to do?
Alternatives: What other options did we consider?
Consequences: What are the trade-offs and implications?
Status: Proposed, accepted, deprecated, superseded?

Quick Example:

# ADR-042: Use PostgreSQL for Primary Database

**Status:** Accepted
**Date:** 2024-01-15
**Deciders:** Backend team, CTO

## Context
Need to select primary database for new microservices platform.
Requirements: ACID transactions, complex queries, 10k+ QPS at launch.

## Decision
Use PostgreSQL 15+ as primary relational database.

## Alternatives Considered
- MySQL: Weaker JSON support, less robust constraint handling
- MongoDB: No ACID across documents, eventual consistency issues
- CockroachDB: Excellent but adds operational complexity we can't support yet

## Consequences
✓ Strong consistency and data integrity
✓ Excellent JSON support for semi-structured data
✓ Team has deep PostgreSQL experience
✗ Vertical scaling limits (will need read replicas at 50k+ QPS)
✗ More complex to shard than DynamoDB if we need it

Workflow

Copy this checklist and track your progress:

ADR Progress:
- [ ] Step 1: Understand the decision
- [ ] Step 2: Choose ADR template
- [ ] Step 3: Document the decision
- [ ] Step 4: Validate quality
- [ ] Step 5: Deliver and file

Step 1: Understand the decision

Gather decision context: what decision needs to be made, why now, who decides, constraints (budget, timeline, skills, compliance), requirements (functional, non-functional, business), and scope (one service vs organization-wide). This ensures the ADR addresses the right problem.

Step 2: Choose ADR template

For technology selection (frameworks, libraries, databases) → Use resources/template.md. For complex architectural decisions with multiple interdependent choices → Study resources/methodology.md. To see examples → Review resources/examples/ (database-selection.md, microservices-migration.md, api-versioning.md).

Step 3: Document the decision

Create adr-{number}-{short-title}.md with: clear title, metadata (status, date, deciders), context (situation and requirements), decision (specific and actionable), alternatives considered (with pros/cons), consequences (trade-offs, risks, benefits), implementation notes if relevant, and links to related ADRs. See Common Patterns for decision-type specific guidance.

Step 4: Validate quality

Self-check using resources/evaluators/rubric_adr_architecture.json. Verify: context explains WHY, decision is specific and actionable, 2-3+ alternatives documented with trade-offs, consequences include benefits AND drawbacks, technical details accurate, understandable to unfamiliar readers, honest about downsides. Minimum standard: Score ≥ 3.5 (aim for 4.5+ if controversial/high-impact).

Step 5: Deliver and file

Present the completed ADR file, highlight key trade-offs identified, suggest ADR numbering if not provided, recommend review process for high-stakes decisions, and note any follow-up decisions needed. Filing convention: Store ADRs in docs/adr/ or architecture/decisions/ directory with sequential numbering.

Common Patterns

For technology selection:

Focus on technical capabilities vs requirements
Include performance benchmarks if available
Document team expertise level
Consider operational complexity

For architectural changes:

Include migration strategy in consequences
Document backward compatibility impact
Consider team velocity impact during transition
Note monitoring and rollback plans

For standards and conventions:

Include examples of the standard in practice
Document exceptions or escape hatches
Consider enforcement mechanisms
Note educational/onboarding implications

For deprecations:

Set status to "Deprecated" or "Superseded"
Link to superseding ADR
Document sunset timeline
Include migration guide

Guardrails

Do:

Be honest about trade-offs (every choice has downsides)
Write for future readers who lack current context
Include specific technical details (versions, configurations)
Acknowledge uncertainty and risks
Keep ADRs immutable (status changes, but content doesn't)
Write one ADR per decision (focused scope)

Don't:

Make decisions sound better than they are
Omit alternatives that were seriously considered
Use jargon without explanation
Write vague consequences ("might improve performance")
Revisit/edit old ADRs (write new superseding ADR instead)
Combine multiple independent decisions in one ADR

Quick Reference

Standard template: resources/template.md
Complex decisions: resources/methodology.md
Examples: resources/examples/database-selection.md, resources/examples/microservices-migration.md, resources/examples/api-versioning.md
Quality rubric: resources/evaluators/rubric_adr_architecture.json

ADR Naming Convention: adr-{number}-{short-kebab-case-title}.md

Example: adr-042-use-postgresql-for-primary-database.md

adr-architecture