Code Quality

Code Quality Skill

Write code that is easy to review, understand, and change. Optimize for the next developer's mental load before optimizing for machine performance, because readable code is cheaper to change, debug, and trust — and most performance wins come from algorithmic choices and profiling, not micro-optimizations.

The three quality axes this skill targets, in priority order:

1. Readability — a reader understands the code top-to-bottom on first pass.
2. Maintainability — the next variant of a concept is a one-file, one-edit

change; existing utilities are reused rather than re-invented; one concept has one canonical home.

1. Pragmatic performance — algorithmic wins by default, micro-optimizations

only when a profiler points at them.

This skill applies in four modes:

1. Plan mode — invoked with Skill("code-quality", "plan"), before any code is written. Reads a plan.md (or proposed approach) and the existing codebase, and verifies the plan's structure follows the existing patterns and avoids predictable design-time risks (premature parallel maps, missing branded primitives, untyped error paths, parameter creep, neighbour mismatch). Returns findings only — no code edits. Used by autonomous-workflow Phase 1. See [rules/plan-mode.md](./rules/plan-mode.md).
2. Authoring mode — when writing new code (e.g., GREEN phase of TDD, new features). Apply principles inline so the first version already meets the bar. Default mode when no argument is passed.
3. Review mode — invoked with Skill("code-quality", "code") or Skill("code-quality", "review"), after code is written. When refactoring, reviewing PRs, or being asked to "clean this up". Diagnose against the rules and propose targeted changes. Returns findings only — no code edits.
4. Simplify mode — invoked with Skill("code-quality", "simplify"), after a feature is implemented and tests are green. Runs the review pass, then applies mechanical refactors (Class M recipes from [rules/refactor-recipes.md](./rules/refactor-recipes.md#recipe-class--mechanical-vs-judgment)) one at a time behind Skill("confidence", "code") ≥ 90 % and a scoped fast-check, reverting on failure. Judgment-class recipes (Class J) stay as proposals. See [rules/simplify-mode.md](./rules/simplify-mode.md). Variants: simplify aggressive also auto-applies Medium-impact mechanical recipes; simplify dry-run reports without writing; simplify scopes to a path instead of the diff.

Detect the mode from the $ARGUMENTS first token: plan → plan mode; code or review → review mode; simplify → simplify mode; anything else (including no argument) → authoring mode. The legacy convention "user said review or audit" still routes to review mode for ad-hoc invocations.

---

The Core Bet

Cognitive complexity — how hard code is to understand — is the single best proxy for long-term maintainability. SonarSource's research showed that cyclomatic complexity (path counting) misses what actually hurts humans: nesting, broken linear flow, and decisions that compound. So the rules below target cognitive load, not theoretical complexity scores. Full citations (SonarSource, Clean Code, Knuth, Alexis King, Gary Bernhardt) and the research grounding for each rule live in [references/citations.md](./references/citations.md).

When in doubt, the heuristic is: can a reader understand this function top-to-bottom on one pass without backtracking? If yes, ship it. If they have to scroll, jump, or hold a stack of conditions in their head, fix it.

The maintainability counterpart: if I add the next obvious variant of this concept, how many files do I have to edit, and will the type system catch me if I miss one? One file with full type coverage is excellent; four hand-synchronised maps in four files is shotgun surgery — fix the structure before adding the variant. See rules/maintainability.md.

---

Quick Reference (load these patterns into working memory)

| Smell | Refactor To | Why | |---|---|---| | Nested if (3+ levels) | Guard clauses + early return | Each indent adds mental load; flat is easier | | Long function (50+ lines, multiple responsibilities) | Extract by intent, name by what not how | One function = one reason to change | | Cryptic name (d, tmp, data) | Domain noun (priceDifference, pendingOrders) | Names ARE documentation | | Boolean parameter | Two named functions or an enum | send(true, false, true) is unreadable | | Comment explaining WHAT | Rename or extract function | Comments rot; names get refactored with code | | Comment explaining WHY (non-obvious constraint) | Keep it | The one comment that earns its place | | Defensive checks for impossible states | Delete | Trust your callers; validate at boundaries | | Flag/option cluster (4+ params) | Object parameter or builder | Working memory holds ~4 chunks | | else after return/throw | Drop the else | Linear flow beats branching | | Magic number/string | Named constant | Future-you will not remember what 7 meant | | Parallel maps over the same union (LABELS, COLORS, ICONS keyed by Status) | One Record | Adding a variant becomes one edit, not N | | Reimplementing a helper that already exists | Search first; use the existing one | Two implementations drift; bugs get fixed in one copy only | | if/else if chain dispatching on a value | Lookup table or single source-of-truth record | Data structure beats control flow; easy to extend | | Same constant (MAX_RETRIES, status strings) duplicated across files | Hoist to one shared module | One concept, one home | | Adding a new variant requires editing 4+ files | Consolidate before adding the variant | Shotgun surgery compounds with every variant | | Separate type User = {...} and userSchema = z.object({...}) for the same shape | One schema; type User = z.infer | Two declarations drift; one cannot | | Re-validating an already-parsed value deep in the stack | Parse once at the boundary; trust the type internally | Validation is a boundary concern, not a per-call concern | | Splitting every nested object into its own sub-schema "for cleanliness" | Keep flat unless the sub-shape is reused or has its own boundary | Premature decomposition; over-engineering | | Function mixes orchestration sentences with low-level mechanics | Extract by abstraction level (R16) | Readers stop at the level they care about | | Runtime check enforcing "these two fields can't both be set" | Discriminated union (R15) | Compiler enforces the invariant; runtime check disappears | | Raw string for Email, UserId, OrderId | Brand the type via the schema (R11) | Mixing IDs becomes a compile error | | any or unjustified cast silencing the type checker | Schema parse, narrowed unknown, or // because: comment (R17) | Unjustified any is the shape most type bugs take | | Function calls Date.now() / Math.random() directly | Inject the clock / RNG (R9) | Pure functions are testable; impure functions are flaky | | Function throws for "not found" or returns sentinels (-1, "") | Total-ise: return null or Result (R10) | Total functions are testable exhaustively | | Every error throws the same Error with a parsed message | Discriminated AppError union (R12) | Structured fields beat string parsing | | Importing a module triggers a side effect | Factory: createX(...) (R20) | Tests stop being accidentally integration tests | | Operation that may be retried is not safe to invoke twice | Idempotency key, upsert, or right HTTP method (R18) | Retries corrupt state otherwise | | Money stored as number | Integer minor units or decimal library (R19) | Floats cannot represent decimal currency exactly | | Floats compared with === | Compare with epsilon, or use integer ticks | 0.1 + 0.2 !== 0.3 | | await in a for loop where Promise.all was meant | Choose serial or parallel consciously | Accidental serialisation is a perf bug | | Every open without a paired close | try/finally or using | Resource leaks compound silently | | New file does not match neighbouring files' patterns | Read 2–3 neighbours and mimic them | Outlier code forces context-switching for every reader | | Refactor PR mixed with feature PR | Split into two PRs | Mixed PRs get rubber-stamped or rejected on the wrong grounds | | Helpers at the top of the file, public function 200 lines down | Public surface first; helpers below in call order | Files read top to bottom | | Reaching a.b.c.d.method() to act on a | Tell, don't ask: put the operation on a | Callers shouldn't walk private structure |

Stack-flagship patterns

A few of the highest-leverage stack-specific rules. Each rule file under rules/stacks/ carries its own Common Mistakes section — load the file when working in that stack.

| Smell | Refactor To | Where | |---|---|---| | React component with 6+ props (especially booleans) controlling visual variations | Compound component with deep dot-notation (Combobox.Content.List.Item) + namespaced control hook (Combobox.useCombobox); shared state via Context | stacks/react/components.md | | useEffect(() => { fetch(...).then(setData) }, [id]) for server data | useQuery({ queryKey, queryFn }) (TanStack Query / SWR) | stacks/react/data-fetching.md | | Mutation waits for the server before updating the UI | Optimistic update in onMutate, rollback in onError, invalidate in onSettled | stacks/react/data-fetching.md | | Form saves on every keystroke | Debounce ~1000 ms + max-wait ~5 s flush + on-blur + on-visibility-hidden; never block input | stacks/react/autosave.md | | Two tabs autosave the same record and one silently overwrites the other | ETag + If-Match; on 412 surface "edited elsewhere — Reload / Keep mine" | stacks/react/autosave.md | | Mutations lost on reload, ad-hoc retries scattered across components | Durable mutation queue in IndexedDB + idempotency key per entry; multi-tab single-writer via navigator.locks | stacks/react/offline-sync.md | | Route handler trusts req.json() body without validation | safeParse with a shared Zod schema (FE + BE import the same module); one error envelope mapped from a discriminated AppError union | stacks/nextjs/endpoints.md |

---

Procedure

The skill operates in four modes — plan (validate a plan), authoring (writing new code), review (refactor / audit, findings only), and simplify (review then auto-apply mechanical recipes). Detect the mode from $ARGUMENTS per the table above; default is authoring.

Load [rules/procedure.md](./rules/procedure.md) for authoring (13-step checklist) and review (8-step pass). Load [rules/simplify-mode.md](./rules/simplify-mode.md) for the simplify procedure (partition → confidence-gate → apply one recipe at a time → scoped fast-check → revert on failure → whole-suite verify at end).

---

Rule Files

Load only what's relevant to the code in front of you. Reading every rule file every time is wasted context. The skill is language-agnostic at its core — the table below splits into framework-neutral rules and stack-specific extensions. Load a stack file only when the code is in that stack.

Language-agnostic rules

| When the code involves... | Load | |---|---| | Conditionals, branching, nesting | rules/control-flow.md | | Long or multi-purpose functions | rules/functions.md | | Variable, function, class names | rules/naming.md | | Comments, docstrings, doc | rules/comments.md | | Performance concerns or hot paths | rules/performance.md | | Error handling, validation, defensive code, schema-first validation, Zod / Pydantic, inferring types from schemas | rules/error-handling.md | | Reviewing existing code | rules/review-checklist.md | | Cognitive complexity scoring | rules/cognitive-complexity.md | | Union types with associated data, parallel maps, duplicated constants, "where should this live", reuse decisions | rules/maintainability.md | | Abstraction levels, type-driven design, illegal states unrepresentable, branded primitives, generics, any/cast discipline | rules/abstraction.md | | Module boundaries, public surface, dependency direction, functional core / imperative shell, DTO ↔ domain ↔ persistence, immutability defaults, side-effecting imports | rules/architecture.md | | Function signatures, parameter ordering, total functions, modeling absence, designing the error type system, tell-don't-ask, file reading order | rules/api-design.md | | Idempotency, money / decimals / floats, dates and time, identifiers, encoding, determinism, assertions, async / concurrency, resource management | rules/correctness.md | | Hard-to-test code, dependency injection of clock / RNG / IDs, when to invoke the tdd skill, UI components locatable by role / label without data-testid | rules/testability.md | | PR scope, neighbour-pattern symmetry, migration & evolution, working with legacy code, diff hygiene | rules/collaboration.md | | Naming a refactor in review output (R1 Consolidate Parallel Maps, R6 Replace Type Declaration with Inferred Type, etc.) | rules/refactor-recipes.md | | Auto-applying mechanical refactors at end of feature (review-then-apply, confidence-gated, revert-on-failure) | rules/simplify-mode.md |

Stack-specific rules

See rules/stacks/README.md for the index and the convention for adding a new stack.

| Stack | When the code involves... | Load | |---|---|---| | React / Next.js | Any UI file (*.tsx, *.jsx, React Native screens) — semantic HTML, ARIA, WCAG 2.2 conformance, focus order, contrast, platform guidelines (Apple HIG, Material Design 3) | Skill('ux') (separate skill) | | React | Splitting components, compound / namespace components (Component.List.Item, Component.useComponent), slots, RSC boundaries | rules/stacks/react/components.md | | React | Client-side data fetching, server state, query caches, optimistic updates, cache surgery (create/update/delete), query keys, request waterfalls, TanStack Query / React Query / SWR, Next.js App Router prefetch, HydrationBoundary, dehydrate, initialData, per-request QueryClient | rules/stacks/react/data-fetching.md | | React | Autosave forms, debounce + max-wait flush, on-blur / visibility / pagehide triggers, local-first draft buffer (localStorage / IndexedDB), status indicator with ARIA live regions, ETag / If-Match conflict detection, offline retry with idempotent PATCH (defers to /ux for form fundamentals) | rules/stacks/react/autosave.md | | React | App-wide offline + sync, durable mutation queue / outbox in IndexedDB, foreground queue vs Background Sync, TanStack Query networkMode + persistence + resumePausedMutations, idempotency keys, reconnection flow, conflict resolution, multi-tab coordination (navigator.locks + BroadcastChannel) | rules/stacks/react/offline-sync.md | | Next.js | Server endpoints, Route Handlers / Server Actions, Zod boundary validation, shared schemas / contracts (FE + BE), API error envelope, HTTP status mapping, auth + rate-limit ordering, request IDs, span the handler with semconv attributes (defers to /otel-instrumentation + /otel-semantic-conventions for depth) | rules/stacks/nextjs/endpoints.md |

---

Critical Rules (apply always)

These are non-negotiable because they reflect human cognitive limits, not style preferences.

1. Linear flow beats branching

A function that flows top-to-bottom — guards, then logic, then return — is always easier than one with deep branching. If you have if/else if/else spanning 30 lines, restructure.

2. Names carry the load

Code is read 10× more than it's written. A 30-character name that explains intent saves more time than the keystrokes it costs to type. Conversely, single-letter names are fine for tight scopes (loop indices, math formulas) where the meaning is unambiguous from context.

3. Functions describe one thing

The f

…

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mthines
• Source: mthines/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.