kikita-labs/kikita-create-nestjs-app

kikita-create-nestjs-app 是做什麼的？

Bootstraps a new NestJS project and generates its AGENTS.md / .agents/ documentation tree so any AI agent (Claude Code, Codex, etc.) working in the project afterwards has a complete, self-maintaining source of truth.

0. Mode detection (always do this first)

Check for .agents/.kikita-scaffold.json in the target directory.

• Missing → fresh init. Continue with section 1 below, then plan.md.
• Present → this project was already scaffolded by this skill. Follow update.md instead — do not re-run the questionnaire or re-scaffold. update.md handles pulling the latest template, diffing since the recorded commit, and merging changes into the project's (possibly customized) .agents/ files.

0.5 Preconditions (fresh init only)

• Confirm the working directory is empty or the user explicitly wants to init here.
• Never run this against a directory that already has an unrelated project without asking first.

1. Questionnaire (always ask before touching the filesystem)

Follow plan.md step by step from here. Do not skip the questionnaire. When done, run through checklist.md before telling the user it's finished.

Ask in two stages, not as 12 questions dumped in one shot — the first answer decides which of the later questions are even relevant, so asking them all upfront means presenting a bot-only user with an "auth for your REST API" question that doesn't apply to them.

Stage 1 — the branching question, alone:

1. Application type: REST API, bot, or both in the same app?
   • If bot (or both): which platform? Telegram (nestjs-telegraf) or Discord — and for Discord specifically, necord is the recommended default (least glue code to satisfy the generic event→handler→service pattern via native Nest DI/guards/pipes) but not mandatory: if the user wants raw discord.js instead (existing team expertise, a discord.js feature necord doesn't wrap yet), that's a legitimate choice — treat it the same as "another platform" below, adapting the generic pattern by hand instead of getting it for free.
   • Any other platform/library (name it — the generic bot transport pattern in architecture/transport-adapter.md adapts to it, but the concrete adapter code needs to be written by hand for anything outside nestjs-telegraf/necord).

Stage 2 — everything else, batched (question 3 only asked if stage 1 answered REST or "both"; every other question below applies regardless of application type):

2. Tests: none, or unit / e2e (Supertest) / both? "Both" also wires Testcontainers for integration tests that hit a real Postgres instance instead of mocks.
3. Auth (skip entirely if stage 1 was bot-only): needed or not? A "yes" scaffolds one fixed battle-tested pattern, not a menu — see core/auth.md's scaffold block: short-lived access token (5–15 min, never in a cookie) + refresh token in an httpOnly cookie scoped to /auth/refresh + rotation (single-use, hashed in DB) + csrf-csrf on the refresh endpoint + argon2id password hashing + a RolesGuard. Do not offer alternatives (sessions, bare JWT without rotation) — this is the one default.
4. Background jobs: BullMQ (Redis-backed queues) or not?
5. Caching: @nestjs/cache-manager + @keyv/redis (cache-aside) or not? Default recommendation is no — add it when there's a real performance need, not speculatively on every CRUD endpoint.
6. File uploads: needed or not? If yes, no further sub-question — the storage vendor (local disk in dev, any S3-compatible bucket in prod: AWS S3, MinIO, R2, Spaces) is an env variable, not a scaffold-time choice.
7. Async messaging / inter-service events: needed or not? If yes, no broker sub-question either — RabbitMQ is the fixed default (task queues, routing, DLQ cover the overwhelming majority of cases). architecture/messaging.md documents Kafka as a later migration path for event-streaming/replay/extreme-throughput needs, not a scaffold-time option.
8. i18n: does this project need multiple locales — translated REST error/validation messages and/or bot replies in more than one language? If yes, no further sub-question — nestjs-i18n is the fixed library; locale resolution differs by transport (REST: Accept-Language header; bot: the platform's own per-user locale field, never Accept-Language) and both are covered in core/i18n.md. If no, skip — a project that will only ever run in one language doesn't need this.
9. JSDoc/TSDoc on public API: enforce mandatory doc comments on every exported symbol, or skip it? Drives whether .agents/agent-surface.md is generated (recommended default: yes).
10. Git policy: may the agent commit and push without asking each time, or must every commit/push be confirmed?
11. Package manager: npm, pnpm, or yarn? Default recommendation: pnpm.
12. Git remote: does the user already have a repo URL to push to? If yes, record it — git remote add origin <url> runs right after git init (see plan.md). If no URL is given, skip this; the user wires the remote later themselves.

Record every answer — they drive both scaffolding and which doc files get generated. Never silently assume a default beyond what's explicitly fixed above; if the user skips a question, ask again for that one. This skill targets a single NestJS project, not a monorepo/Nx workspace and not a distributed microservices topology (multiple deployable services/repos) — if the user wants either, say this skill doesn't cover that and stop rather than improvising. A message broker (see question 7) only wires a hybrid app (app.connectMicroservice() inside the same single app) — it never spins up a second service.

Fixed defaults — never ask about these, they're locked by design:

• ORM/DB: Prisma + Postgres. No TypeORM (not recommended for new projects), no Drizzle (niche perf/edge pick), no NoSQL/MongoDB branch in v1 of this skill.
• Validation: class-validator + class-transformer on the HTTP layer, global ValidationPipe wired in main.ts unconditionally with whitelist, forbidNonWhitelisted, forbidUnknownValues, and transform all true. Zod for env/config validation (@nestjs/config + validate), enforced by an ESLint rule blocking direct process.env reads outside the schema file — see testing-and-quality.md's "Mechanically Enforced Rules". Nested DTO fields require both @ValidateNested() and @Type(() => NestedDto) — see code-style/dto-and-validation.md. DTO reuse via PartialType/OmitType/PickType imported from @nestjs/swagger, never @nestjs/mapped-types — ESLint-blocked, not just documented (same section).
• Response shape: a global ClassSerializerInterceptor + @Exclude() on DTO fields is the one fixed serialization mechanism — never a raw Prisma entity returned, never a second serialization approach introduced without an ADR.
• Swagger/OpenAPI: always wired for the REST branch.
• CORS: always configured in main.ts via an env-driven origin allowlist, never *.
• API versioning: Nest URI Versioning (/v1/...), fixed in architecture/transport-adapter.md.
• Rate limiting: @nestjs/throttler always installed for REST and/or bot, IP-keyed for REST routes, user/chat-id-keyed for bot handlers — never skipped as "not needed yet".
• Logging: nestjs-pino, always — structured JSON logs, no plain Logger option offered.
• Health checks: @nestjs/terminus, always wired, not questionnaire-gated — two separate routes, GET /health/live (process-only, no external checks) and GET /health/ready (Prisma + any chosen Redis/RabbitMQ dependency), never merged into one. See core/health.md.
• Graceful shutdown: app.enableShutdownHooks() always called in main.ts — without it, Prisma's OnModuleDestroy connection-close hook never fires on container restart.
• Prisma error mapping: a global PrismaExceptionFilter (APP_FILTER) maps PrismaClientKnownRequestError codes to the matching Nest HTTP exception — a Prisma error never surfaces as a bare unhandled 500.
• Prisma client generation: package.json always gets "postinstall": "prisma generate" — the generated client is gitignored, so a fresh clone/CI needs this to compile at all.
• docker-compose.yml: always generated (dev/test only, never referenced by prod deploy) — Postgres always present; Redis added only if BullMQ and/or caching was chosen (one shared instance for both); RabbitMQ added only if messaging was chosen.

2. Generate

Follow plan.md. Copy files from templates/ into the target project, including the dotfiles (.gitignore, .gitattributes, .editorconfig, .prettierrc, .prettierignore, .nvmrc, .vscode/extensions.json, .env.example, docker-compose.yml).

Two different things happen with questionnaire answers, don't conflate them:

• Text placeholders — find-and-replace every {{TOKEN}} with the real value, leave none behind: {{PROJECT_NAME}}, {{APP_TYPE}}, {{BOT_PLATFORM}}, {{TESTS}}, {{GIT_POLICY}}, {{PACKAGE_MANAGER}}, {{NODE_VERSION}}, {{DATE}}.
• Inclusion gates — application type, bot platform, tests, auth, queue, cache, storage, messaging, i18n, and mandatory-TSDoc answers don't fill a placeholder; they decide whether a whole file (or a <!-- SCAFFOLD -->-marked block inside one) is copied at all. A "no" answer means the file/block is deleted, not filled with an empty string. Gated files must still be linked from AGENTS.md / the relevant README when kept, and their links removed when skipped.

3. Verify

Run checklist.md in full before reporting success.

Notes on documentation structure (read once, then follow templates/ literally)

• CLAUDE.md is always a one-line stub pointing to AGENTS.md.
• AGENTS.md is the mandatory entry point: a short "Must Read" list plus non-negotiable rules, at the project root.
• .agents/README.md is a flat index of everything under .agents/ — keep it in sync whenever a conditional file (auth, queue, cache, storage, messaging, i18n, agent-surface) is added or skipped.
• Topics that are genuinely one short doc stay flat in .agents/*.md (workflow, git-policy, documentation, testing-and-quality, agent-surface, refactoring, progress).
• Topics that fan out into several docs, or per-feature registries, get a subfolder with its own README.md hub: .agents/code-style/, .agents/architecture/, .agents/shared/, .agents/core/, .agents/decisions/.
• .agents/shared/README.md registers src/common/ (framework-agnostic utilities — zero @nestjs/* imports — plus generic pipes/filters/interceptors/decorators). .agents/core/README.md registers src/core/ app-wide singletons (auth, Prisma client provider, logger, queue, cache, storage, i18n — each with its own conditional doc file when that feature was chosen).
• .agents/decisions/README.md explains when a short ADR is required (layer direction, message broker migration, versioning strategy change) — always generated, starts with no ADR files.
• All tracked file content — including TSDoc — is English only. No Cyrillic, no mojibake.
• All docs in .agents/ must read like kikita-create-angular-app's templates: imperative, short, example-backed, ending in a review/verification checklist.