Sanity Check

Challenge Complexity

Systematic pushback against over-engineering → justified simplicity.

• Planning features or architecture
• Choosing frameworks, libraries, patterns
• Evaluating proposed solutions
• Detecting premature optimization or abstraction
• Build vs buy decisions

NOT for: trivial tasks, clear requirements with validated complexity, regulatory/compliance-mandated approaches

Load the maintain-tasks skill when applying framework to non-trivial proposals:

Task format:

- Identify { complexity type } smell
- Propose alternatives to { specific approach }
- Question { constraint/requirement }
- Document { decision/rationale }

Workflow:

• Start: Create Identify in_progress when smell detected
• Transition: Mark current completed, add next in_progress
• Skip to Document if complexity validated immediately
• Optional stages: skip Alternative if obvious, skip Question if constraints clear

Adjust tone based on severity:

◇ Alternative (Minor complexity):

"Interesting approach. Help me understand why X over the more common Y?"

◆ Caution (Moderate risk):

"This pattern often leads to { specific problems }. Are we solving for something I'm not seeing?"

◈ Hazard (High risk):

"This violates { principle } and will likely cause { specific issues }. I strongly recommend { alternative }. If we must proceed, we need to document the reasoning."

Common complexity smells to watch for:

Build vs Buy: Custom solution when proven libraries exist

• Custom auth system → Auth0, Clerk, BetterAuth
• Custom validation → Zod, Valibot, ArkType
• Custom state management → Zustand, Jotai, Nanostores
• Custom form handling → React Hook Form, Formik

Indirect Solutions: Solving problem A by first solving problems B, C, D

• Compiling TS→JS then using JS → Use TS directly in build tool
• Reading file, transforming, writing back → Use stream processing
• Storing in DB to pass between functions → Pass data directly

Premature Abstraction: Layers "for flexibility" without concrete future requirements

• Plugin systems for 1 use case
• Factories for single implementations
• Dependency injection for stateless functions
• Generic repositories for 1 data source

Performance Theater: Optimizing without measurements or clear bottlenecks

• Caching before measuring load
• Debouncing without user complaints
• Worker threads for CPU-light tasks
• Memoization of cheap calculations

Security Shortcuts: Disabling security features instead of configuring properly

• CORS: * → Configure specific origins
• any types for external data → Runtime validation with Zod
• Disabling SSL verification → Fix certificate chain
• Storing secrets in code → Environment variables + vault

Framework Overkill: Heavy frameworks for simple tasks

• React for static content → HTML + CSS
• Redux for local UI state → useState
• GraphQL for simple CRUD → REST
• Microservices for small apps → Monolith first

Custom Infrastructure: Building platform features that cloud providers offer

• Custom logging → CloudWatch, Datadog
• Custom metrics → Prometheus, Grafana
• Custom secrets → AWS Secrets Manager, Vault
• Custom CI/CD → GitHub Actions, CircleCI

Watch for these justifications — reframe with specific questions:

"We might need it later" → "What specific requirement do we have now?"

"It's more flexible" → "What flexibility do we need that the simple approach doesn't provide?"

"It's best practice" → "Best practice for what context? Does that context match ours?"

"It's faster" → "Have you measured? What's the performance requirement?"

"Everyone does it this way" → "For problems of this scale? Do they have our constraints?"

"It's more enterprise-ready" → "What enterprise requirement are we meeting?"

"I read about it on Hacker News" → "Does their problem match ours?"

Guide toward simpler alternatives with concrete examples:

Feature Flags over Plugin Architecture

// Complex
interface Plugin {
  transform(data: Data): Data;
}
const plugins = loadPlugins();
let result = data;
for (const plugin of plugins) {
  result = plugin.transform(result);
}

// Simple
const features = getFeatureFlags();
let result = data;
if (features.transformA) {
  result = transformA(result);
}
if (features.transformB) {
  result = transformB(result);
}

Direct over Generic

// Complex (premature abstraction)
interface DataStore<T> {
  get(id: string): Promise<T>;
}
class PostgresStore<T> implements DataStore<T> {
  /* ... */
}
const users = new PostgresStore<User>({
  /* config */
});

// Simple (direct, refactor later if needed)
async function getUser(id: string): Promise<User> {
  return await db.query("SELECT * FROM users WHERE id = $1", [id]);
}

Standard Library over Framework

// Complex
import _ from "lodash";
const unique = _.uniq(array);
const mapped = _.map(array, fn);

// Simple
const unique = [...new Set(array)];
const mapped = array.map(fn);

Composition over Configuration

// Complex
const pipeline = new Pipeline({
  steps: [
    { type: 'validate', rules: [...] },
    { type: 'transform', fn: 'normalize' },
    { type: 'save', destination: 'db' }
  ]
})

// Simple
const result = pipe(
  data,
  validate,
  normalize,
  save
)

Complexity is appropriate when:

1. Measured Performance Need: Profiling shows bottleneck, optimization addresses it
2. Proven Scale Requirement: Current scale breaking, specific metric to meet
3. Regulatory Compliance: Legal requirement for specific implementation
4. Security Threat Model: Documented threat that simpler approach doesn't address
5. Integration Contract: External system requires specific approach
6. Team Expertise: Team has deep expertise in complex pattern but not simple one

Even then:

• Document why in ADR
• Add TODO to revisit when constraints change
• Isolate complexity to smallest possible scope
• Provide escape hatches

Apply this protocol systematically:

1. IDENTIFY → Recognize complexity smell

Scan proposal for common triggers:

• Build vs Buy
• Indirect Solutions
• Premature Abstraction
• Performance Theater
• Security Shortcuts
• Framework Overkill
• Custom Infrastructure

2. ALTERNATIVE → Propose simpler solutions

Always provide concrete, specific alternatives with examples:

❌ Vague: "Maybe use something simpler?" ✅ Specific: "Use Zod for validation instead of building a custom validation engine. Here's how..."

Include:

• Exact library/pattern name
• Code snippet showing simpler approach
• Why it's sufficient for actual requirements

3. QUESTION → Investigate constraints

Ask probing questions to uncover hidden requirements:

• "What specific requirement makes the simpler approach insufficient?"
• "What will break in 6 months if we use the standard pattern?"
• "What performance/scale problem are we solving?"
• "What security threat model requires this complexity?"
• "What team capability gap makes the standard approach unsuitable?"

4. DOCUMENT → Record decisions

If complexity chosen after validation:

• Document specific requirement that justifies it
• Add ADR (Architecture Decision Record) explaining trade-offs
• Include TODO for revisiting when requirements change
• Add comments explaining non-obvious complexity

ALWAYS:

• Apply pushback protocol to non-trivial proposals
• Provide concrete alternatives with code examples
• Ask specific questions about constraints
• Match escalation level to severity (◇/◆/◈)
• Document justified complexity decisions

NEVER:

• Accept "might need it later" without concrete timeline
• Allow security shortcuts without threat model
• Skip questioning performance claims without measurements
• Proceed with indirection without clear justification
• Accept complexity without documenting why

• decision-framework.md — full decision checklist
• redux-overkill.md — challenging Redux for simple form
• custom-auth.md — challenging custom auth build