/nuxt-improve-codebase-architecture

Improve Nuxt Codebase Architecture

Surface architectural friction in a Nuxt codebase and propose deepening opportunities — refactors that turn shallow modules into deep ones, using Nuxt's own extension points as seams. The aim is testability and AI-navigability.

Worktree isolation

Before any edit, read and follow the worktree isolation contract. Keep mutation in a task-owned wt worktree with a live claim. Keep the primary checkout read only.

Vocabulary and principles

Use the terms in LANGUAGE.md exactly — module, interface, implementation, depth, seam, adapter, leverage, locality. Don't drift into "component," "service," "API," or "boundary."

Load-bearing principles (full list in LANGUAGE.md):

• Deletion test: if deleting the module just inlines 1-2 lines into each caller, it was a pass-through.
• The interface is the test surface.
• One adapter = hypothetical seam. Two = real.
• No import-time side effects — critical for shared/, composables, server utils.
• Prefer Nuxt-native seams over hand-rolled equivalents.
• Default feature-local, promote on evidence. New composables/useFoo.ts or components/Foo.vue is a claim of app-wide ownership; the rename blast radius and naming-collision risk make global auto-import the most expensive seam decision. Colocate + _-prefix until ≥2 unrelated features consume it.

This skill is informed by the project's domain model (GLOSSARY.md, docs/arch/, docs/adr/): the domain language names good seams; ADRs record decisions not to re-litigate.

Companion files

• LANGUAGE.md — full vocabulary and principles.
• DEEPENING.md — dependency categories, seam ladder (shared/ → packages/* → Nuxt module/layer), testing strategy.
• NUXT-SEAMS.md — catalogue of Nuxt's framework-native seams and the scope boundaries between nitro / nuxt server / nuxt client.
• NITRO-CONVENTIONS.md — server-side conventions Nitro doesn't ship (validators, policies, presenters, error handling, service providers, event/listener registry, tasks, job dispatch, per-layer schema + migrations).
• VUE-CONVENTIONS.md — framework-agnostic Vue patterns (service composables as factory + provide + use, component thinness).
• NUXT-APP-CONVENTIONS.md — Nuxt-specific app layer ($fetch as port, async resource composables, useState for SSR-safe app-wide state, client policy mirror).
• INTERFACE-DESIGN.md — design-it-twice with parallel sub-agents for chosen candidates.
• code-comments.md — the comment contract for any code this skill writes.

Process

1. Explore

Always run ripast first. Every claim about depth, caller counts, locality, or cross-scope leaks must cite a ripast invocation. Text search misses shadowed identifiers, type-only imports, Vue template refs, and Nuxt auto-imported callers.

Pass --tsconfig .nuxt/tsconfig.json on tree/rename/move/rename-file so auto-imports resolve. Run nuxi prepare first if .nuxt/ is stale. For scan (rg-driven) and when trimming tree, scope with --glob:

--glob 'app/**,server/**,composables/**,plugins/**,modules/**,layers/**,shared/**,nuxt.config.ts'

Adjust per project: drop app/** pre-Nuxt-4 (use pages/**,components/**), drop layers/** if none, add packages/** or a module's src/**.

Opening pass — run before forming any candidate:

Per-candidate, before listing (scan is rg-driven — use --glob here):

Cite numbers when presenting ("scan returns 2 callers, both in server/api/ — single-scope, deletion test fails").

Read the project's domain glossary and any ADRs in the area first. Then orient on the Nuxt shape of the project:

• Is this a Nuxt app, a Nuxt module, or a workspace with layers?
• Which Nuxt extension points are already in use: modules/, layers/, plugins/, composables/, server/api, server/middleware, nitro plugins, runtimeConfig, app:*/nitro:*/build:* hooks, addImports, addServerHandler, addComponent, addPlugin?
• Where does the project hand-roll its own seam (a singleton, a custom plugin pattern, a global registry) instead of using one Nuxt provides?

Then use subagent_type=Explore to walk the codebase. Friction signals:

• Feature requires bouncing across composables/, plugins/, server/api, runtimeConfig — scan --graph mermaid spanning ≥3 Nuxt scopes = no central seam.
• Shallow modules: composable wrapping a single useState, plugin providing one untyped value, server util that's a one-line $fetch passthrough. Spot via tree --exports exported.
• Pure functions extracted for testability while real bugs hide in plugin order / hook timing / SSR vs client divergence (no locality).
• Cross-cutting concerns (auth, logging, telemetry) smeared across app/, server/, nitro instead of sitting behind one hook/module.
• A cluster of plugins + composables + server handlers that wants to be a Nuxt module or layer with a small options + hooks interface.
• Walk the convention files for missing seams: each "Gap" entry is a candidate when present in the codebase.
• Cross-scope leaks. types//utils//constants/ imported from both server/ and composables/ — move to shared/. Detect with scan --graph mermaid: importer set straddling both = leak.
• Import-time side effects in shared/, composables, or server utils — fire in every consumer, break SSR/treeshaking/tests.

2. Present candidates

Present a numbered list of deepening opportunities. For each candidate:

• Files — which files/modules are involved (include the Nuxt-flavoured paths: modules/foo.ts, layers/billing/, server/api/x.post.ts, plugins/auth.client.ts, composables/useBar.ts, nuxt.config.ts)
• Problem — why the current architecture is causing friction
• Solution — plain English description of what would change, named in Nuxt terms (e.g. "collapse these three plugins and two composables into a single Nuxt module that registers addImports for the public composables and an app:created hook for the singleton wiring")
• Benefits — explained in terms of locality and leverage, and also in how tests would improve (e.g. "the module can be tested with @nuxt/test-utils against a fixture; today the plugin's behaviour can only be observed through a full app boot")

Use GLOSSARY.md for the domain, LANGUAGE.md for the architecture, NUXT-SEAMS.md for the framework seam, and the relevant convention file + section number when the candidate is a convention gap. Example phrasings: "the Order intake module exposed as a Nuxt layer", "establish NITRO-CONVENTIONS.md §1 for the Order create handler" — not "the FooBarHandler", not "the Order service".

Reject before listing. Bias toward fewer, sharper candidates. Validate caller counts with scan --kind identifier-reference,import-specifier scoped to the whole project (no --glob narrowing to the candidate's own layer/dir), and confirm the shape they consume — a uniform interface for the wrapper is a precondition for collapsing it. Cross-layer callers with a different state shape break the candidate. No guesswork.

• Deletion test fails. Thresholds: single caller = adapter, not a seam; Nuxt layer/module needs ≥3 plugins/composables/handlers + shared config or hooks; schema/policy/presenter/validator/async-resource composable earns its keep at ≥2 callers OR real branching/transform; pure value/type mapping (just destructure/rename) needs ≥4 callers.
• No locality win. "Testable in isolation" / "future-proof" without bugs/changes concentrating is relocation, not depth. Includes defensive re-validation of invariants a zod schema, typed event, or DB column already enforces.
• Renames or file reshuffles dressed up as architecture. Note separately or skip.
• Smuggles import-time side effects.
• Already matches a NITRO/VUE/NUXT-APP convention.
• Meta-handler with no work. defineApiHandler earns its keep only when ≥2 of {validation, auth, presenter, domain-error mapping} fire per call.
• Render-factory composable. Mostly h() calls / column defs / template output with little reactive state. Acceptable at ≥4 callers; otherwise inline or slot-scope.
• Promotes to global auto-import without cross-feature consumers. Reject if scan --tsconfig .nuxt/tsconfig.json shows callers in a single feature dir — keep colocated + _-prefixed. Promote only at ≥2 unrelated feature consumers. See NUXT-SEAMS.md §"Internal vs global scope".
• Bundled-concern composable fusing ≥3 concerns (fetch + cache + optimistic + rollback + retry) with no override seams.
• Wrapper-collapse without project-wide caller audit. A "delete this pass-through wrapper, inline into N callers" candidate is invalid until scan <Wrapper> --kind identifier-reference,import-specifier (no --glob) has been run and every caller's consumed-shape verified. Wrappers in most cases have cross-layer callers (admin pages, ad-hoc usages) that don't share the "obvious" caller's state shape; collapsing on a partial census produces a narrower interface that breaks the outliers at typecheck.
• Config threaded through ≥3 layers unchanged = missing seam. Consume at creation layer or place adapter at deepest meaningful seam.
• Single-reader runtimeConfig key. Confirm with scan --kind property,member-access; one hit = fails unless genuinely env-overridable at deploy.
• Hides event behind a global accessor. Server abstractions thread H3Event through — reject module-level singletons, ambient useRuntimeConfig() without event, implicit getRequestEvent().
• Contradicts a current ADR without a load-bearing reason.

Forbidden patterns — always flag, never propose. Full lists in NITRO-CONVENTIONS.md and NUXT-APP-CONVENTIONS.md "Forbidden patterns". Headline: never wrap an authenticated route in defineCachedEventHandler — the cache wrapper strips request headers, breaking session/auth. Hoist auth to an outer defineEventHandler and cache the data fetch via defineCachedFunction keyed on the resource.

ADR conflicts: if a candidate contradicts an existing ADR, only surface it when the friction is real enough to warrant revisiting the ADR. Mark it clearly (e.g. "contradicts ADR-0007 — but worth reopening because…"). Don't list every theoretical refactor an ADR forbids.

Do NOT propose interfaces yet. Ask the user: "Which of these would you like to explore?"

3. Grilling loop

Once the user picks a candidate, drop into a grilling conversation. Walk the design tree with them — constraints, dependencies, the shape of the deepened module, what sits behind the seam, what tests survive. For Nuxt candidates, also walk: which Nuxt hook fires when, SSR vs client behaviour, build-time vs runtime, where runtimeConfig ends and component state begins.

Side effects happen inline as decisions crystallize:

• Naming a deepened module after a concept not in GLOSSARY.md? Read the installed Brundlefly glossary Skill and add the term there: the term, one sentence of meaning in this codebase, and where it lives.

• Sharpening a fuzzy term during the conversation? Update GLOSSARY.md right there.

• User rejects the candidate with a load-bearing reason? Offer an ADR, framed as: "Want me to record this as an ADR so future architecture reviews don't re-suggest it?" Only offer when the reason would actually be needed by a future explorer to avoid re-suggesting the same thing — skip ephemeral reasons ("not worth it right now") and self-evident ones. Write it to docs/adr/NNNN-slug.md with context, decision, and consequences.

• Want to explore alternative interfaces for the deepened module? See INTERFACE-DESIGN.md. Sub-agents are pre-seeded with Nuxt-native shapes (module + hooks, layer, plugin + composable, nitro plugin) so the design space is grounded in what Nuxt already offers.

• Need to know the true blast radius of a rename/move before committing? pnpm dlx @ripast/cli scan <symbol> (counts) or pnpm dlx @ripast/cli scan <symbol> --graph mermaid (importer graph). Quote numbers before promising scope.

• Decision crystallized into a concrete refactor? Execute through ripast (--tsconfig .nuxt/tsconfig.json rewrites auto-imported callers too):

  Refactor	Command
  Rename symbol	pnpm dlx @ripast/cli rename <from> <to> --tsconfig .nuxt/tsconfig.json --apply (--scope <file> if multi-declared)
  Move exported declaration	pnpm dlx @ripast/cli move <symbol> --from <a> --to <b> --tsconfig .nuxt/tsconfig.json --apply
  Move a file	pnpm dlx @ripast/cli rename-file <old> <new> --tsconfig .nuxt/tsconfig.json --apply
  Design-token / class migration	pnpm dlx @ripast/cli css-class-rename --map tokens.json --apply (seed via css-class-scan)

  All mutating commands default to dry-run; preview, then --apply. --verify blocks on new type diagnostics — fix them, never --no-verify. Edit is only correct for single-file or <5-match changes.