Imported from tomevault-io/tomes (ryanthedev--code-foundations/skills/cc-routine-and-class-design/SKILL.md). Install upstream with npx skills add tomevault-io/tomes --skill cc-routine-and-class-design. Copyright stays with the author.

Skill: cc-routine-and-class-design

STOP - Crisis Invariants (NEVER SKIP)

Prerequisites

This skill assumes:

• Object-oriented programming with class-based inheritance
• Mutable state (objects that change over time)
• Synchronous or simple async patterns

For functional programming, prototype-based inheritance, or heavily concurrent code, adapt principles rather than applying literally. See "Pattern-Specific Guidance" section.

When NOT to Use

• Scripting/automation code - One-off scripts don't benefit from class design rigor
• Prototyping phase - When exploring ideas before committing to design (time-box to max 1 week)
• Simple data transfer objects - Pure DTOs without behavior are exempt from ADT requirements (but DTOs with validation, toString, equals are NOT exempt)
• Framework-mandated patterns - When framework REQUIRES inheritance to function (e.g., Android Activity won't work without extending Activity). "Framework supports inheritance" or "framework examples use inheritance" is NOT mandated
• Performance-critical inner loops - Where accessor overhead matters (measure first - must show profiling data)
• Test doubles (mocks, stubs, fakes) - Test code intentionally violates design rules; empty overrides and minimal abstractions are appropriate

Crisis Invariants - NEVER SKIP

These checks are NON-NEGOTIABLE regardless of deadline pressure:

Why these three? Violations create problems that CANNOT be easily fixed post-crisis. They require architectural changes, not patches.

Design fixes deferred are rarely completed. Technical debt from wrong inheritance is architectural, not patchable.

Modes

CHECKER

Purpose: Execute design checklists against routines and classes Triggers:

• "review my class design"
• "check routine quality"
• "audit class interfaces"
• "evaluate cohesion" Non-Triggers:
• "how should I design this class" -> APPLIER
• "should I use inheritance here" -> APPLIER
• "refactor this routine" -> cc-refactoring-guidance Checklist: See checklists.md Output Format: | Item | Status | Evidence | Location | |------|--------|----------|----------|

Severity Classification Rubric:

Note: Test suite status is IRRELEVANT to checker results. A class can pass all tests and still have VIOLATION-level design issues that predict future maintenance problems.

APPLIER

Purpose: Guide class interface design, inheritance decisions, and routine creation Triggers:

• "how should I design this class"
• "should I use inheritance or containment"
• "what cohesion type is this routine"
• "how many parameters is too many"
• "should I extract this to a routine" Non-Triggers:
• "review my class" -> CHECKER
• "check this routine" -> CHECKER
• "optimize this code" -> performance-optimization Produces: Class interface designs, inheritance/containment decisions, routine signatures, cohesion classifications Constraints:
• [p.133] Default to containment; only inherit if "is-a" is literally true (LSP)
• [p.143] Inheritance depth: target <3 levels, WARNING at 3, VIOLATION at 4+, SEVERE at 6+
• [p.178] Parameter limit: 7 maximum (8+ is VIOLATION requiring redesign)
• [p.168] Target functional cohesion for routines
• [p.139] Asking "What should this class hide?" drives good design

Parameter Guidelines

Threshold Table:

"About 7" means: 7 is the limit. 8 is over. Count ALL parameters including optional ones with defaults. Variadic args (*args/...) count as 1.

Ordering convention (implies data flow sequence - ORDER MATTERS):

1. Input-only parameters first
2. Input-and-output (modify) parameters second
3. Output-only parameters third

Key Definitions

LSP (Liskov Substitution Principle)

If A inherits from B, then everywhere code uses B, you can substitute A without breaking anything.

Test: Does calling every method on the base class make sense for the derived class?

Inheritance requires BOTH:

1. Semantic test: "A is a B" makes English sense to domain experts
   • YES: "Dog is an Animal", "Manager is an Employee"
   • NO: "EmployeeCensus is a ListContainer", "UserSession is a Logger"
2. LSP test: Every method of B works correctly when A is substituted
   • NO empty overrides (override to do nothing)
   • NO exceptions that base class doesn't throw
   • NO precondition strengthening

If EITHER fails: use containment instead.

Functional Cohesion

Routine performs one and only one operation.

Name Test: If you need "and" or "then" in the name to describe what the routine does, it has multiple operations.

• PASS: ValidateUserInput(), CalculateTotalPrice(), SendWelcomeEmail()
• FAIL: ValidateAndSaveUser(), ReadFileThenParseJSON(), InitializeAndConnect()

Scope: "One operation" means one action at the routine's declared abstraction level.

• CreateUser() is ONE operation (user creation) even though it involves validation, hashing, insertion
• Those sub-steps are at a LOWER abstraction level

Inheritance vs Containment Decision

Question order rationale: Data-only sharing → containment (simplest). Behavior-only sharing → interface/protocol. Both data and behavior → requires full LSP verification before inheriting.

digraph inheritance_decision {
    rankdir=TB;

    START [label="How should A relate to B?" shape=doublecircle];

    q1 [label="Shares ONLY data?" shape=diamond];
    q2 [label="Shares ONLY behavior\n(method signatures)?" shape=diamond];
    q3 [label="Shares BOTH data\nand behavior?" shape=diamond];
    q4 [label="Is 'A is a B' literally true?\n(LSP: A substitutes for B\neverywhere)" shape=diamond];

    contain1 [label="CONTAIN\ncommon object" shape=box];
    interface1 [label="USE INTERFACE/PROTOCOL\n(not inheritance)" shape=box];
    inherit2 [label="INHERIT" shape=box];
    contain2 [label="CONTAIN" shape=box];
    none [label="NO RELATION\n(reconsider if relationship needed)" shape=box];

    START -> q1;
    q1 -> contain1 [label="yes"];
    q1 -> q2 [label="no"];
    q2 -> interface1 [label="yes"];
    q2 -> q3 [label="no"];
    q3 -> q4 [label="yes"];
    q3 -> none [label="no"];
    q4 -> inherit2 [label="yes"];
    q4 -> contain2 [label="no"];
}

FINAL CHECK (apply AFTER flowchart determines INHERIT, BEFORE committing to inheritance):

• Depth < 3 levels (definitely < 6)?
• No empty overrides needed?
• All base data private (not protected)?

If ANY answer is NO → use CONTAIN instead.

Cohesion Classification

Classification algorithm: Questions test for highest-quality cohesion first. Stop at first YES answer - that's your classification. This prevents under-classifying (e.g., calling functional cohesion "sequential").

Target: Functional cohesion - routine performs one and only one operation.

digraph cohesion_classification {
    rankdir=TB;

    START [label="Classify routine cohesion" shape=doublecircle];

    q1 [label="Single operation?\n(Name Test passes)" shape=diamond];
    q2 [label="Ordered operations\nsharing data step-to-step?" shape=diamond];
    q3 [label="Operations share\ndata only?" shape=diamond];
    q4 [label="Same-time operations\n(startup/shutdown)?" shape=diamond];
    q5 [label="Orchestrates calls\n(not direct work)?" shape=diamond];
    q6 [label="Order from external\nrequirement (UI)?" shape=diamond];
    q7 [label="Control flag selects\noperation?" shape=diamond];

    functional [label="FUNCTIONAL\n[ACCEPT]" shape=box style=filled fillcolor=lightgreen];
    sequential [label="SEQUENTIAL\n[ACCEPT w/caution]" shape=box style=filled fillcolor=lightyellow];
    communicational [label="COMMUNICATIONAL\n[ACCEPT w/caution]" shape=box style=filled fillcolor=lightyellow];
    temporal_ok [label="TEMPORAL\n[ACCEPT]" shape=box style=filled fillcolor=lightgreen];
    temporal_fix [label="TEMPORAL\n[FIX: orchestrate]" shape=box style=filled fillcolor=orange];
    procedural [label="PROCEDURAL\n[REJECT]" shape=box style=filled fillcolor=lightcoral];
    logical [label="LOGICAL\n[REJECT]" shape=box style=filled fillcolor=lightcoral];
    coincidental [label="COINCIDENTAL\n[REDESIGN]" shape=box style=filled fillcolor=red fontcolor=white];

    START -> q1;
    q1 -> functional [label="yes"];
    q1 -> q2 [label="no"];
    q2 -> sequential [label="yes"];
    q2 -> q3 [label="no"];
    q3 -> communicational [label="yes"];
    q3 -> q4 [label="no"];
    q4 -> q5 [label="yes"];
    q4 -> q6 [label="no"];
    q5 -> temporal_ok [label="yes"];
    q5 -> temporal_fix [label="no"];
    q6 -> procedural [label="yes"];
    q6 -> q7 [label="no"];
    q7 -> logical [label="yes"];
    q7 -> coincidental [label="no"];
}

Evidence: 50% of highly cohesive routines fault-free vs 18% low cohesion [Card et al. 1986, N=450 routines]. This is MAINTENANCE data, not shipping data - the 50% vs 18% gap appears during modifications.

Cohesion Types Reference

Types listed from BEST (top) to WORST (bottom) quality - ORDER MATTERS:

"ACCEPT w/caution" means:

1. Document in code comment WHY this cohesion type is acceptable here
2. Review whether functional cohesion is achievable with reasonable effort
3. Add TODO if cohesion should improve in future refactoring

"Caution" is not permission to ignore - it's permission with accountability.

Detecting Orchestration Pattern

If description uses these verbs, likely ORCHESTRATION (temporal cohesion OK):

• "orchestrates", "coordinates", "delegates", "dispatches", "routes"

If description uses these verbs, likely DIRECT WORK (check cohesion type):

• "handles", "processes", "performs", "executes", "calculates"

Orchestration Example (GOOD):

def startup():
    load_config()       # calls another routine
    init_database()     # calls another routine
    start_server()      # calls another routine
    log_startup_complete()

Direct Work Example (FIX by extracting):

def startup():
    config = json.load(open('config.json'))  # direct work - extract
    db = Database(config['db_url'])          # direct work - extract
    db.connect()                             # direct work - extract
    server.start(config['port'])             # acceptable call

Improving Cohesion

Pattern-Specific Guidance

Builder Pattern

Builders exhibit intentional method chaining (fluent interface). Evaluate cohesion at the class level, not individual methods.

• Methods returning this are acceptable
• Class cohesion = "constructs one type of object"
• Skill's routine-level rules don't apply to builder methods

Mixin/Trait Patterns

Mixins add behavior without "is-a" claims. Evaluate differently:

• Does the mixin have a single, focused responsibility?
• Are there fewer than 3 mixins per class?
• Could containment achieve the same goal more explicitly?
• Prefer containment - mixins hide the relationship; containment makes it explicit

Singleton Pattern

Apply extra scrutiny:

• Does global state truly need to be global?
• Is this hiding a dependency that should be explicit?
• Would dependency injection be cleaner?

Event Handlers and Callbacks

Cohesion evaluation differs for event-driven patterns:

• Handler cohesion: Does it handle ONE event type with ONE response?
• Callback cohesion: Does the callback do ONE thing when triggered?
• Async routines: Evaluate the complete operation, not just what runs before await

Minimum Viable Compliance (Time-Constrained)

When full checklist review is impractical, these 7 items are MANDATORY:

Total: ~4 minutes - This is the floor, not the ceiling.

Skipped items require: Technical debt ticket with specific follow-up date.

Evidence Summary

Why These Rules Apply Even After Success:

Critical insight: This is MAINTENANCE data, not shipping data. All code "works" on day 1. The 50% vs 18% gap appears during modifications, extensions, and bug fixes. Code passing tests on first commit does not predict maintenance quality.

Chain

Converted and distributed by TomeVault — claim your Tome and manage your conversions.