frontend

How the @nuclia/core SDK is structured and how to use it correctly in this monorepo. Covers SDK architecture (Nuclia class, rest/db/auth layers, KnowledgeBox, RetrievalAgent), adding new API calls, wrapping SDK calls in Angular services in @flaps/core, error handling patterns, and URL routing (global vs regional backends). Activate this skill whenever a task involves calling the Nuclia API, adding a new endpoint, extending KnowledgeBox or RetrievalAgent, creating an Angular service that fetches data, working with SDKService, or handling HTTP errors. Do not wait for the user to say "SDK" — if they ask for "a service that fetches X from the API" or "add support for endpoint Y", this skill applies.

Domain knowledge about the Nuclia / Progress Agentic RAG platform. Use this skill when discussing product features, platform capabilities, API behaviour, UX decisions, or how the platform works end-to-end. Also use when planning a new feature to validate it against the platform's existing capabilities and constraints before building.

Angular 21 patterns as used in the Nuclia frontend monorepo — covering OnPush change detection, inject() dependency injection, NgModule-based components (default), signal-based state, RxJS↔Signal bridges, functional guards, and lazy routing. Activate this skill for ANY Angular task in this repo: creating or modifying components, writing services, managing state, adding routes or guards, refactoring legacy code to modern patterns, debugging change detection issues, or deciding which state management tier to use. Do not wait to be asked about "Angular patterns" specifically — if the task involves TypeScript files in apps/ or Angular libs/, this skill applies. Also use when migrating from @Input/@Output decorators or constructor injection to the modern Angular 21 style.

Keeps all AI-agent documentation in sync with the codebase: root AGENTS.md, per-project AGENTS.md files in apps/ and libs/, and knowledge embedded in .github/skills/ and .github/agents/. Activate this skill when the user mentions syncing docs after a PR or commit, asks "what docs need updating?", says "update knowledge", references a big feature that just landed ("there was a big update in commit X"), or runs the staleness check script and sees stale files. Also activate proactively when you notice an AGENTS.md file is missing for a project that was just created, or when a skill references a pattern that no longer matches the source code. Do NOT wait for the user to explicitly say "knowledge sync" — if code has changed and docs might be stale, this skill applies immediately.

Reviews and refactors AGENTS.md files in the Nuclia frontend monorepo for completeness, accuracy, and AI-agent usability. Invoke this skill whenever you are working with an AGENTS.md file in apps/ or libs/, when asked to review or improve agent documentation, when creating a new AGENTS.md for a project, or when you notice that an AGENTS.md seems outdated, thin, or missing. Proactively suggest a review even when the user only asks an unrelated question about a project if no AGENTS.md exists for it.

Deeply scans code for runtime bugs, logic errors, and correctness defects in the Nuclia frontend monorepo. Use this skill whenever code needs to be audited for actual bugs rather than style issues — including during code review, when a user reports unexpected behaviour, when writing or reviewing a PR, or when asked to "find bugs", "check for bugs", "is this correct?", or "what could go wrong here?". This skill is intentionally narrow: it reports only things that are broken or will likely break at runtime, not style, architecture, or best-practice nudges (those belong in code-review). Designed to be invoked by automated agents as well as interactively.

Performs a focused code review of changed files or PR diffs in the Nuclia frontend monorepo. Activates when the user explicitly says "review", "code review", "review this PR", "review this diff", "review these changes", or "review this file". Checks changed code against all repo standards: Angular 21 patterns, RxJS best practices, design system usage, SDK patterns, module boundaries, test coverage, and general TypeScript hygiene. Returns a flat list of findings grouped by file with a severity label for each.

Expert knowledge of the Nuclia design system — @nuclia/sistema and @guillotinaweb/pastanaga-angular. Use this skill whenever a task involves ANY of the following: creating or modifying Angular UI components in dashboard, rao, manager-v2, or nucliadb-admin; using or overriding design tokens (colours, spacing, typography); adding icons, modals, toasts, dropdowns, tables, tabs, form fields, or any pa-/nsi- prefixed component; styling with SCSS tokens or utility classes; implementing settings pages, configuration rows, cards, badges, labels, progress indicators, or navigation elements. Don't wait to be asked explicitly about design system — if the task is about frontend UI in this monorepo, this skill applies.

Error handling patterns for the Nuclia frontend monorepo — covering IErrorResponse discrimination, SisToastService usage, catchError categories (rethrow / convert / swallow), the retry429Config and uploadRetryConfig helpers, the AuthInterceptor contract, error state in Svelte search-widget stores, and error handling in the React rao-widget. Activate this skill ANY TIME a task involves catchError, handling SDK responses, showing error toasts, writing error callbacks in subscriptions, defining retry logic, or surfacing errors in the UI. Do not wait to be asked about "error handling" specifically — if a service is calling the SDK and there is no error path, or if an error toast is being triggered with a raw string, this skill applies immediately.

Internationalisation patterns for the Nuclia frontend monorepo — covering ngx-translate usage in Angular components and services, the BabelEdit .babel project file workflow for adding/ renaming/deleting translation keys, flat JSON key conventions, interpolation syntax, the Pastanaga Angular translation bridge, and the independent i18n system inside search-widget (Svelte 5). Activate this skill ANY TIME a task involves adding, renaming, or deleting a translation key, adding a new locale file, using the translate pipe, calling TranslateService, wiring up i18n in a new component or lib, or when the user mentions BabelEdit. Do not wait to be asked about "i18n" specifically — if a template string is being hardcoded that should be translatable, this skill applies.

Nx monorepo patterns as used in the Nuclia frontend workspace — covering module boundary enforcement, generator usage, project.json targets & executors, adding new apps/libs with correct tags, affected commands, and cache configuration. Activate this skill for ANY Nx task in this repo: running/debugging builds, adding a new library or app, wiring up a new target, debugging affected runs, understanding project graph dependencies, or fixing ESLint module-boundary violations. Do not wait for the user to say "Nx" explicitly — if the task touches project.json, nx.json, tsconfig.base.json paths, or the build/test/serve pipeline, this skill applies. Also use when a user asks which project name to pass to nx commands, or when they want to know which tags or constraints are in play.

Performance patterns for the Nuclia frontend monorepo — covering OnPush change detection with signals, computed() memoisation, track in @for loops, lazy routing, shareReplay(1) placement, debounce/auditTime conventions, CDK virtual scroll, bundle budgets, and the known shareReplay() anti-pattern. Activate this skill when reviewing a component for performance, adding @for loops, choosing between signal() and computed(), setting up lazy routes, writing service-level streams, or when the user asks about bundle size, change detection, or rendering performance. Also use when a code review finds markForCheck(), shareReplay() without a buffer argument, $index in track expressions, or eager route loading — these are all actionable problems this skill covers.

RxJS patterns as used in the Nuclia frontend monorepo — covering Subject type selection, flattening operator decisions (switchMap vs concatMap vs mergeMap), combination operators (combineLatest vs forkJoin vs withLatestFrom), error handling, shareReplay placement, cleanup strategies, debounce conventions, Angular signal interop, and import hygiene. Activate this skill for ANY task that touches observable pipelines, service state streams, RxJS subscriptions, or operator choices in Angular apps or libs. Do not wait to be asked about "RxJS patterns" specifically — if the task writes or modifies a .pipe(), creates a Subject, adds a subscription in a service or component, uses catchError/switchMap/forkJoin, or asks "which operator should I use here?", this skill applies. Also use when debugging streams that never emit, emit multiple times unexpectedly, or cause memory leaks. This skill covers Angular libs and apps only; it does not cover search-widget (Svelte) or rao-widget (React).

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, update or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

Testing patterns for the Nuclia frontend monorepo — covering Jest + jest-preset-angular for Angular apps/libs and Vitest for search-widget (Svelte 5) and rao-widget (React 19). Use this skill whenever you are writing or modifying a spec file, debugging a failing test, setting up a new TestBed configuration, mocking Angular services, working with OnPush components in tests, or using Vitest with Svelte/React. Do not wait to be asked about "testing patterns" specifically — any task involving *.spec.ts files, ng-mocks, TestBed, vitest, or fixture.detectChanges() should activate this skill.