agentleFS
L-X-TL-X-T/ng-agentic-skills/.agents/skills/ng-refactor/SKILL.md

ng-refactor

Skill13 starsChanged 44 days agoMITView on GitHub

Refactor a dusty / legacy Angular component into a small, modern, signal-based one using a repeatable 7-step blueprint – Analyse → Declutter → Update → Modernize → Type → Refactor → Review – while delegating Angular updates and official migration schematics to ng-migrate. Use when the user wants to refactor, modernize, clean up, tame, or "de-dust" a large, messy, or legacy Angular component, service, directive, or feature.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    No big-bang rewrite: Only small, readable intermediate steps that can be reviewed and verified.

    line 51
  2. Distinctive:

    Preserve meaning: Do not delete existing comments or domain logic.

    line 56
  3. Distinctive:

    Git discipline: Start from a clean working tree.

    line 60
  4. Distinctive:

    After every step, run all four in order on the result, and do not proceed until green (on a deliberately dusty fallback, "green" means …

    line 65
  5. Very specific:

    Read the .ts, .html, .scss, and spec together without changing code.

    line 88
  6. Very specific:

    Read the result top-to-bottom as if onboarding fresh: confirm one clear responsibility and canonical member order, no any, small named …

    line 161
  7. Distinctive:

    Use the repository VERIFICATION.md when present.

    line 193

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

ng-refactor

Over years, Angular components accumulate features, special cases, and workarounds until responsibilities blur between UI, state, and business logic – any spreads, lifecycle hooks multiply, templates fill with logic, and files cross 400, 600, 800 lines. Such "dusty" components carry a high mental load, make every change risky, and are harder for an AI agent to grasp than for a human. Clean names, small functions, and precise types reduce both misunderstandings and hallucinations.

This skill modernizes one component (or a small set) at a time through a repeatable blueprint of seven steps, each over a chosen scope, each verified by the full gauntlet and paused at a clean checkpoint for human review before the next. AI accelerates the rebuild; the architecture decisions stay human.

ng-refactor orchestrates the seven-step refactoring flow. For Angular version updates, automatic update migrations, and official @angular/core migration schematics, invoke ng-migrate from the relevant step and let that skill own the migration workflow.

The blueprint – 7 steps

#StepWhat it does
1AnalyseRead the code, name the one responsibility, catalog symptoms, write a plan
2Declutter (Entschlacken)Delete dead code, unused members, redundant SCSS – pure subtraction
3Update (NG Update)Delegate Angular updates to ng-migrate so modern APIs and migrations exist
4Modernize (Modernisieren)Delegate official schematics; replace lifecycle hooks; default OnPush + zoneless
5Type (Typisierung)Replace any with explicit types – the first lever in legacy code
6RefactorSplit functions, extract shared logic, enforce SRP, reorder members
7ReviewRead top-to-bottom; confirm the gains; human approves every change

Steps 2–6 stack: each assumes the previous ran. Declutter before you update, update before you modernize (you need the new APIs), type before you refactor (types make the cut safe). Full details, per-step prompts, and checklists live in references/blueprint.md.

Guardrails (non-negotiable)

These hold for every step. They are what separate a safe refactoring from "AI slop".

  • Treat the agent like a junior. Review every change before commit. No step is "done" because the agent said so – it is done when a human has read the diff.
  • No big-bang rewrite. Only small, readable intermediate steps that can be reviewed and verified. Never replace a whole component in one unreadable diff.
  • Every change must earn its keep (Umbau-Nutzen): each edit must deliver less complexity, clearer responsibility, or modernization. If it delivers none, don't make it.
  • Preserve meaning. Do not delete existing comments or domain logic. Behavior-preserving refactors must not change behavior – only structure, names, and types.
  • Plan first, then implement (the interactive prompting level: the agent proposes, you confirm, then it acts): in Step 1 produce a plan and ask clarifying questions before touching code.
  • Git discipline. Start from a clean working tree; checkpoint after each step. Never stage, commit, or push on the user's behalf – surface the diff and let them commit.

The gauntlet

After every step, run all four in order on the result, and do not proceed until green (on a deliberately dusty fallback, "green" means green or no worse than the documented red baseline from Step 1):

build → lint → formatting check → test

Resolve concrete commands from package.json scripts (this workspace: pnpm build, pnpm lint, pnpm exec prettier --check ., pnpm test --watch=false). The linter is your checklist: it enforces max-lines: 400, complexity (≤ 20 in TS – the ≤ 10 figure is the template cyclomatic-complexity rule), explicit-function-return-type, sort-imports, prefer-const, and the Angular best-practice rules (prefer-signals, prefer-standalone, prefer-on-push-component-change-detection, prefer-control-flow, prefer-self-closing-tags, …). @typescript-eslint/no-explicit-any is only warn-level in eslint.config.js, so lint stays green with any left in – treat remaining any as a manual check, not something the gauntlet catches. A step that is "done" but red is not done.

Workflow

Each step is summarized to its intent and its checkable _Done when_; full actions, prompt shapes, and checklists live in references/blueprint.md.

1. Analyse

Read the .ts, .html, .scss, and spec together without changing code. Name the single responsibility, catalog the symptoms (LoC; lifecycle hooks; any / weak types; old patterns like *ngIf, @Input(), @Output(), ngClass, ngStyle; template logic; duplication; unclear names), and map the contract. Establish a behavior safety net – an e2e or characterization test (see create-e2e-tests) plus a recorded gauntlet baseline – then write the plan and ask clarifying questions before touching code.

Done when you hold a written plan, a baseline that is green or explicitly documented as red, and a behavior safety net; the user has answered open questions. No production code changed.

2. Declutter (Entschlacken)

Pure subtraction, lowest risk first. Delete dead code, unused imports/members/inputs, commented-out blocks (but keep meaningful comments and domain logic), and redundant SCSS.

Done when only live code remains, behavior is unchanged, and the gauntlet is green.

3. Update (NG Update)

A workspace-level step – do it once, not per component. Delegate Angular package updates and automatic version migrations to ng-migrate (it owns the update command, clean-tree preflight, migrations, gauntlet, and checkpoint); bring non-Angular packages along when the plan needs them, and address security (CSP, Trusted Types, 3rd-party packages) while versions move. Cadence and the automatic-migration list: references/modern-angular.md.

Done when Angular is current, automatic migrations have run, and the gauntlet is green.

4. Modernize (Modernisieren)

Delegate the official schematics to ng-migrate – one migration per checkpoint (standalone, control flow, inject(), signal input()/output()/queries, self-closing tags, ngClass/ngStyle → bindings, lazy routes, unused-import cleanup) – then do the modernization the schematics don't cover. Replace lifecycle hooks (prefer none at all): move setup into field initializers / the constructor, use takeUntilDestroyed() for manual subscriptions, and reach for effect() / afterNextRender() / afterEveryRender() / afterRenderEffect() instead of ngAfterViewInit / ngOnChanges.

Rely on the default OnPush + zoneless (v22+: drop explicit changeDetection: settings and any Zone.js remnants), prefer computed() over manually-maintained derived state and signal-based binding (input/output/model); SignalStore only if you truly need it (gate and patterns in ng-signal-store), Signal Forms via ng-forms.

Mapping table and rule-by-rule guidance: references/modern-angular.md.

Done when the chosen migrations are applied (each gauntlet-green and checkpointed), no lifecycle hooks remain that signals/render hooks can replace, and the component carries no explicit changeDetection (default OnPush).

5. Type (Typisierung)

Strict typing is the first lever in legacy code: give every return and parameter type an explicit type ("everything has a type"), prefer unknown over any and type over interface, and extract small, precisely-named types – each non-trivial one in its own file (Rule of One) – so typed in-/outputs make the component readable.

Done when no any remains and explicit-function-return-type is clean (no-explicit-any is only warn-level, so confirm the remaining any by hand), and the gauntlet is green.

6. Refactor

The structural work in small reviewable diffs: shrink big event handlers into small, intent-revealing functions (buildRequest, validateInput, updateSelection); lift shared logic into services/directives/pipes/utils (DRY); split oversized components under SRP + Rule of One (each in its own file, ≤ ~400 LoC); reorder members to the canonical structure; thin the template by moving logic into computed()/methods (template cyclomatic complexity ≤ 10); prefer private, readonly, const. Member ordering, naming, and extraction patterns: references/clean-code-and-structure.md.

Done when each unit has one responsibility, no file exceeds ~400 LoC, functions are small and named, the template is logic-light, and every diff was gauntlet-green and checkpointed.

7. Review

Read the result top-to-bottom as if onboarding fresh: confirm one clear responsibility and canonical member order, no any, small named functions, no file > ~400 LoC, no explicit changeDetection (default OnPush), no leftover lifecycle hooks, intact comments and domain logic, and tests that are still green and still meaningful – every change earning a real gain. Summarize what changed and why (the Umbau-Nutzen) and list follow-ups. The general review practice (scopes, two passes, verifying findings) lives in the code-review skill; run code-review for an independent reviewer, or hand a large scope to ng-review-architecture. The human approves every change before commit – this skill never commits for you.

Done when the human has reviewed the full diff, the gauntlet is green, and the gains are written down.

Setup this skill assumes

The workshop's guardrails make AI refactoring reproducible – they should already be in place (see Lab 01): Prettier + ESLint (the shared quality frame), an AGENTS.md with the project's rules, the Angular MCP for current docs, and per-tool secret-deny rules (.claude/settings.json, .codex/config.toml) so the agent never reads what it shouldn't. If they're missing, set them up first – the gauntlet and the linter-as-checklist depend on them.

Pairs with

  • ng-migrate – owns Angular package updates, automatic update migrations, and official @angular/core migration schematics for Steps 3 and 4.
  • ng-review-architecture – deeper review in Steps 1 & 7.
  • ng-forms · ng-security · ng-performance · ng-accessibility – focused passes when a refactoring touches forms, security, performance, or a11y.
  • create-e2e-tests – build the safety net in Step 1.

Verification contract

Use the repository VERIFICATION.md when present. Run checks independently on the final files so a known-red baseline cannot suppress later results. Autofix and formatting writes belong to implementation; reverify after them. Compare exact baseline diagnostics and run the relevant behavior tests explicitly.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.