AgentSkillsCN

adr-writer

创建架构决策记录，记录关键技术决策，包括背景、考虑的备选方案、权衡、后果和决策负责人。适用于“架构决策”、“技术选择”、“设计决策”或“ADR”。

SKILL.md
--- frontmatter
name: adr-writer
description: Creates Architecture Decision Records documenting key technical decisions with context, alternatives considered, tradeoffs, consequences, and decision owners. Use when documenting "architecture decisions", "technical choices", "design decisions", or "ADRs".

ADR Writer

Document architecture decisions with clear context, alternatives, and consequences.

ADR Template

markdown
# ADR-001: [Title of Decision]

**Status:** Proposed | Accepted | Deprecated | Superseded by ADR-XXX
**Date:** 2024-01-15
**Deciders:** Alice (Tech Lead), Bob (Principal Engineer)
**Owner:** Alice

## Context

What is the issue we're facing? What factors are driving this decision?

We need to choose a database for our new analytics service. Current
requirements:

- 10M+ events per day
- Complex aggregation queries
- Real-time dashboards
- Budget: $5k/month
- Team familiar with SQL

## Decision

We will use PostgreSQL with TimescaleDB extension.

## Alternatives Considered

### Option 1: PostgreSQL + TimescaleDB (CHOSEN)

**Pros:**

- Team already knows PostgreSQL
- Time-series optimization for analytics
- Reliable and proven
- Good query performance
- Reasonable cost (~$3k/month)

**Cons:**

- Requires manual scaling effort
- Not purpose-built for analytics

### Option 2: ClickHouse

**Pros:**

- Excellent query performance for analytics
- Built for analytics workloads
- Column-oriented storage

**Cons:**

- Team unfamiliar with ClickHouse
- Steeper learning curve
- Different query syntax

### Option 3: BigQuery

**Pros:**

- Fully managed
- Excellent for analytics
- Scales automatically

**Cons:**

- Higher cost (~$8k/month for our volume)
- Vendor lock-in to GCP
- Less control over optimization

## Tradeoffs

**What we're optimizing for:**

- Team velocity (familiar tech)
- Cost efficiency
- Good enough performance

**What we're sacrificing:**

- Peak analytical performance (vs ClickHouse)
- Fully managed experience (vs BigQuery)

## Consequences

### Positive

- Development can start immediately (no learning curve)
- Lower operational costs
- Can reuse existing PostgreSQL expertise
- Easy integration with current stack

### Negative

- Will need to manually optimize queries
- May need to revisit if we scale 10x
- Requires more operational work than BigQuery

### Risks

- Performance may degrade at 100M+ events/day
- **Mitigation:** Monitor query performance, plan migration at 50M events/day

## Implementation Notes

- Use TimescaleDB hypertables for event storage
- Implement continuous aggregates for common queries
- Set up read replicas for dashboard queries
- Document scaling runbook at 50M events/day

## Follow-up Actions

- [ ] Provision PostgreSQL + TimescaleDB cluster (Alice, by 2024-01-20)
- [ ] Create migration script from MySQL (Bob, by 2024-01-22)
- [ ] Set up monitoring dashboards (Alice, by 2024-01-25)
- [ ] Document scaling thresholds (Alice, by 2024-01-30)

## References

- [TimescaleDB Benchmarks](https://example.com)
- [Cost Analysis Spreadsheet](https://docs.google.com/...)
- [Team Survey Results](https://example.com)

## Revision History

- 2024-01-15: Initial draft (Alice)
- 2024-01-16: Added cost analysis (Bob)
- 2024-01-17: Accepted by architecture review board

ADR Numbering

code
ADR-001: Initial System Architecture
ADR-002: Database Selection for Analytics
ADR-003: Authentication Strategy
...

Status Workflow

code
Proposed → Accepted → Implemented
    ↓
Rejected

Accepted → Deprecated → Superseded by ADR-XXX

Common Decision Types

Technology Selection:

  • •Database choice
  • •Framework selection
  • •Cloud provider
  • •Programming language

Architecture Patterns:

  • •Microservices vs Monolith
  • •Event-driven vs Request-response
  • •Sync vs Async communication

Infrastructure:

  • •Deployment strategy
  • •Monitoring approach
  • •Scaling strategy

Security:

  • •Authentication method
  • •Data encryption approach
  • •Access control model

Quick Start Guide

bash
# 1. Create new ADR
cp templates/adr-template.md docs/adr/ADR-042-title.md

# 2. Fill in sections
# - Context: Why are we making this decision?
# - Decision: What did we decide?
# - Alternatives: What else did we consider?
# - Consequences: What are the impacts?

# 3. Review with team
# - Share in architecture channel
# - Get feedback from stakeholders
# - Iterate on alternatives

# 4. Update status to "Accepted"
# 5. Link from main architecture docs

Best Practices

  1. •Write ADRs for significant decisions: Not every choice needs an ADR
  2. •Document alternatives: Show you considered options
  3. •Be honest about tradeoffs: Every decision has downsides
  4. •Keep it concise: 1-2 pages maximum
  5. •Update status: Mark deprecated/superseded ADRs
  6. •Link related ADRs: Create decision trails
  7. •Include follow-ups: Action items with owners

Anti-Patterns

❌ Too detailed: 10-page ADRs nobody reads ❌ No alternatives: Looks like decision was predetermined ❌ Missing consequences: Ignoring downsides ❌ No owner: Nobody accountable ❌ Outdated status: Old ADRs marked "Proposed"

Review Checklist

  • • Clear problem statement in Context
  • • Decision is stated explicitly
  • • 2+ alternatives considered
  • • Tradeoffs honestly assessed
  • • Consequences (positive and negative) documented
  • • Risks identified with mitigations
  • • Decision owner assigned
  • • Follow-up actions with deadlines
  • • Status is current

Output Checklist

  • • ADR document created
  • • Context explains the problem
  • • Decision clearly stated
  • • 2-3 alternatives documented
  • • Tradeoffs section filled
  • • Consequences listed (positive & negative)
  • • Risks identified with mitigations
  • • Decision date and owners
  • • Follow-up actions assigned
  • • Status is set