sdlc-kit
Spec-Driven Development for Claude Code. 11 skills for the daily PR loop — init, PRDs, ADRs, SDD specs (EARS), tasks, git worktrees, code re
Open source Repository Open in the app JSON README (API)
About
Spec-Driven Development for Claude Code. 11 skills for the daily PR loop — init, PRDs, ADRs, SDD specs (EARS), tasks, git worktrees, code reviews (Code Reviewer + AppSec dual-axis), retrospectives. Includes a PostToolUse hook that auto-indexes every .md save into a SQLite tracker. Self-contained HTML dashboard (Kanban + branch graph) included.
Details
- Kind
- Plugins
- Topic
- Version control
- Publisher
- vellus-ai
- Origin
- marketplace
- Category
- ferramentas
- Forks
- 1
- Last push
- 2026-04-20T19:20:35Z
- Repository state
- ativo
- Language
- Python
- License
- MIT
- Added
- 2026-08-30 01:48:58
- Updated
- 2026-08-30 01:48:58
- Origin id
vellus-ai/sdlc-kit/sdlc-kit
README
# SDLC Kit > Spec-Driven Development (SDD) + Domain-Driven Design (DDD) vault for **Claude Code**, versioned inside your Git repo — Obsidian-compatible, LLM-optimized, language-agnostic. [](https://claude.ai/code) [](https://github.com/vellus-ai/sdlc-kit/tree/main/plugins) [](#vault-structure) [](https://www.python.org/) [](LICENSE) [](https://github.com/vellus-ai/sdlc-kit/releases/tag/v0.4.0) > 📖 **Este documento também está disponível em [Português (Brasil)](#português-brasil) abaixo.** --- ## Why it exists When you work with Claude Code on real projects, three problems surface again and again: - **Architectural decisions get lost** between sessions — the LLM has no context on what was decided. - **Requirements drift out of sync** with implementation — the code moves forward, the spec falls behind. - **Every new session starts from zero** — no history of ADRs, epics, domain, incidents. SDLC Kit solves this by maintaining a versioned `.sdlc/` vault inside your own repo, with automatic indexing, **22 skills** across the full lifecycle (planning → architecture → domain → specs → development → design-system → retrospectives), and a self-contained HTML dashboard. --- ## Two composable plugins Starting at **v0.4.0**, SDLC Kit ships as **two plugins** in the same marketplace. Install just the core for the daily PR loop; add the extended pack when you need governance, architecture modeling and post-delivery analysis. | Plugin | Skills | Install | Purpose | |---|---|---|---| | **`sdlc-kit`** (core) | 11 | `/plugin install sdlc-kit@sdlc-kit` | Daily delivery loop: init, sync, status, dash, prd, adr, spec, task, worktree, review, retro. | | **`sdlc-kit-extended`** | 11 (optional) | `/plugin install sdlc-kit-extended@sdlc-kit` | Governance (trd, epic, milestone, steer), architecture & domain (c4, api, domain, design-system), ops/analysis (incident, trace, impact). Requires the core plugin. | --- ## Features - **Canonical 8-phase vault** (`00-steering` → `07-retrospectives`). - **22 skills** invokable via `/sdlc-kit:<name>`, each with a stable `list / scaffold / transition` contract. - **SDD** — `requirements.md` (EARS) + `design.md` + `tasks.md` trio per feature, with approval gates. - **DDD + C4** — aggregates, domain events, context map, ubiquitous language, 3-level diagrams. - **Living governance** — auto-numbered ADRs, cross-cutting TRDs, 4 API contract styles (REST/async/gRPC/webhook), Design System. - **Traceable delivery** — PRDs, epics, milestones with RAG status, task lifecycle workflow, PR review co-signed by Code Reviewer + AppSec. - **Post-delivery observability** — retros, 4-state incidents (open → mitigated → resolved → post-mortem) with auto-populated timestamps. - **Graph analysis** — `sdlc-trace` (PRD↔spec↔task↔review matrix) and `sdlc-impact` (forward/backward BFS on wikilinks). - **Self-contained HTML dashboard** — Kanban, Epics & Milestones, Documents, Domain; no server, via File System Access API. - **Git worktree + PR tracking** in SQLite. - **PostToolUse hook** re-indexes every saved `.md`, rate-limited to 1 signal / 5 s per vault. - **i18n of `_INDEX.md`** — rendered in the language set at `.sdlc-kit/marker.json:locale` (default `pt-br`, alternate `en`). Code and templates stay English; the LLM mirrors your chat language in content. --- ## Install ### Via Claude Code marketplace (recommended) Inside any Git project: ``` /plugin marketplace add vellus-ai/sdlc-kit /plugin install sdlc-kit@sdlc-kit # core — required /plugin install sdlc-kit-extended@sdlc-kit # optional extended skills /reload-plugins ``` Then run `/sdlc-kit:init` to scaffold `.sdlc/`. > Pin a version: `/plugin marketplace add vellus-ai/sdlc-kit@v0.4.0`. ### One-liner installer (alternative — clones locally and registers in settings.json) **Linux / macOS:** ```bash curl -fsSL https://raw.githubusercontent.com/vellus-ai/sdlc-kit/main/install.sh | bash ``` **Windows (PowerShell):** ```powershell irm "https://raw.githubusercontent.com/vellus-ai/sdlc-kit/main/install.ps1?v=$(Get-Random)" | iex ``` The installer checks Python 3.11+, `git` and the `claude` CLI; clones the repo to `~/.claude/plugins/sdlc-kit`; installs the Python package via `pip install -e`; and registers the plugins in `~/.claude/settings.json`. ### Uninstall **Linux / macOS:** ```bash curl -fsSL https://raw.githubusercontent.com/vellus-ai/sdlc-kit/main/uninstall.sh | bash ``` **Windows (PowerShell):** ```powershell irm "https://raw.githubusercontent.com/vellus-ai/sdlc-kit/main/uninstall.ps1?v=$(Get-Random)" | iex ``` `.sdlc/` vaults inside your projects are **preserved**. --- ## Quick start ``` /sdlc-kit:init ``` A short interview auto-detects owner, remote and project type (greenfield vs brownfield), creates `.sdlc/` with the 8 canonical phases, `CLAUDE.md`, `_INDEX.md`, `dashboard.html` and the SQLite tracker. ### Typical feature flow ```text # 1 — Product alignment (01-planning) /sdlc-kit:prd → initiative PRD # 2 — Architectural decisions & requirements (02-architecture) /sdlc-kit:adr → numbered technical decision (ADR-NNNN) /sdlc-kit:trd → cross-cutting requirements (extended) /sdlc-kit:c4 → C4 Context/Container/Component diagrams (extended) /sdlc-kit:api → REST/async/gRPC/webhook contract (extended) # 3 — Domain modeling (03-domain, extended) /sdlc-kit:domain → aggregates, events, ACL contracts, context map # 4 — Feature spec (04-specs) /sdlc-kit:spec → trio requirements.md (EARS) + design.md + tasks.md # 5 — Execution (05-development) /sdlc-kit:worktree → isolated git worktree + vault record /sdlc-kit:task → start / complete / block / reopen a TASK-NNN # 6 — PR review /sdlc-kit:review → Code Reviewer + AppSec checklist, 6 sections, decision # 7 — Cycle close /sdlc-kit:retro → sprint retrospective /sdlc-kit:incident → post-mortem (extended) # Anytime /sdlc-kit:sync → revalidate and regenerate MOCs + _INDEX.md /sdlc-kit:status → vault snapshot /sdlc-kit:trace → PRD ↔ spec ↔ review matrix (extended) /sdlc-kit:impact --seed ADR-0007 → what breaks if I deprecate this? (extended) /sdlc-kit:dash → open the dashboard in the browser ``` --- ## Vault structure ```text .sdlc/ ├── CLAUDE.md # vault doctrine — read by the agent before any write ├── _INDEX.md # living index, regenerated by /sdlc-kit:sync ├── dashboard.html # self-contained dashboard ├── .sdlc-kit/ │ ├── marker.json # vault root + kit version + locale │ └── db.sqlite # SQLite (WAL): notes, tasks, worktrees, events, … ├── 00-steering/ # product / tech / standards ├── 01-planning/ # PRDs, epics, milestones ├── 02-architecture/ # ADRs, TRDs, C4, API designs ├── 03-domain/ # DDD artifacts ├── 04-specs/ # SDD trios per feature ├── 05-development/ # git worktrees & branches ├── 06-design-system/ # tokens, components, patterns └── 07-retrospectives/ # reviews, retros, incidents ``` --- ## Development ```bash git clone https://github.com/vellus-ai/sdlc-kit.git cd sdlc-kit pip install -e ".[dev,yaml]" pytest # full suite pytest tests/test_scanner.py # specific file pytest --cov=core --cov-report=html # coverage ``` ### Conventions - Branches: `feat/<slug>` or `fix/<slug>`. - TDD is the default — test before implementation. - Every skill script supports `--dry-run` and emits a single JSON object to stdout; exit codes `0` / `1` / `2`. - Minimum coverage: 90% on `core/` modules. --- ## Documentation - **Architecture** — [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) - **Contributing** — [`docs/CONTRIBUTING.md`](docs/CONTRIBUTING.md) - **Publishing** — [`docs/PUBLISHING.md`](docs/PUBLISHING.md) - **Testing** — [`docs/TESTING.md`](docs/TESTING.md) - **Privacy policy** — [`PRIVACY.md`](PRIVACY.md) - **Changelog** — [`CHANGELOG.md`](CHANGELOG.md) --- ## Contributing 1. Fork the repository. 2. Create a branch: `feat/<slug>`. 3. Implement with TDD. 4. Open a PR describing **what** and **why**. Bugs or feature requests: [open an issue](https://github.com/vellus-ai/sdlc-kit/issues). --- ## License [MIT](LICENSE) © 2026 [Vellus](https://vellus.tech/) --- --- ## Português (Brasil) > Vault de **Spec-Driven Development (SDD)** + **Domain-Driven Design (DDD)** para o **Claude Code**, versionado dentro do seu repositório Git — compatível com Obsidian, otimizado para LLMs, agnóstico de linguagem. ### Problema que resolve Ao trabalhar com Claude Code em projetos reais, três problemas surgem constantemente: - **Decisões arquiteturais se perdem** entre sessões — o LLM não tem contexto do que foi decidido. - **Requisitos saem de sincronia** com a implementação — o código avança, a spec fica para trás. - **Cada nova sessão começa do zero** — sem histórico de ADRs, épicos, domínio, incidentes. O SDLC Kit resolve isso mantendo um vault `.sdlc/` versionado no seu próprio repo, com indexação automática, **22 skills** cobrindo o ciclo completo (planning → architecture → domain → specs → development → design-system → retrospectives) e um dashboard HTML autocontido. ### Dois plugins compostos A partir da **v0.4.0**, o SDLC Kit é distribuído como **dois plugins** no mesmo marketplace. Instale só o core para o ciclo de PR diário; adicione o extended quando precisar de governança, modelagem arquitetural e análise pós-entrega. | Plugin | Skills | Instalação | Propósito | |---|---|---|---| | **`sdlc-kit`** (core) | 11 | `/plugin install sdlc-kit@sdlc-kit` | Ciclo de entrega diário: init, sync, status, dash, prd, adr, spec, task, worktree, review, retro. | | **`sdlc-kit-extended`** | 11 (opcional) | `/plugin install sdlc-kit-extended@sdlc-kit` | Governança (trd, epic, milestone, steer), arquitetura & domínio (c4, api, domain, design-system), ops/análise (incident, trace, impact). Requer o plugin core. | ### Instalação Dentro de qualquer projeto Git no Claude Code: ``` /plugin marketplace add vellus-ai/sdlc-kit /plugin install sdlc-kit@sdlc-kit # core — obrigatório /plugin install sdlc-kit-extended@sdlc-kit # skills extended (opcional) /reload-plugins ``` Depois execute `/sdlc-kit:init` para criar o vault `.sdlc/`. > Pinar uma versão: `/plugin marketplace add vellus-ai/sdlc-kit@v0.4.0`. ### Uso rápido Execute `/sdlc-kit:init` dentro do seu projeto. Uma entrevista curta auto-detecta dono, remoto e tipo (greenfield / brownfield) e cria as 8 fases canônicas, o `CLAUDE.md`, o `_INDEX.md`, o `dashboard.html` e o SQLite tracker. ### Funcionalidades - Vault estruturado em **8 fases canônicas** (`00-steering` → `07-retrospectives`). - **22 skills** invocáveis via `/sdlc-kit:<nome>`, cada uma com contrato `list / scaffold / transition` estável. - **SDD** — trio `requirements.md` (EARS) + `design.md` + `tasks.md` por feature, com portas de aprovação. - **DDD + C4** — aggregates, domain events, context map, ubiquitous language, diagramas em 3 níveis. - **Governança viva** — ADRs numerados, TRDs cross-cutting, 4 tipos de contratos de API (REST/async/gRPC/webhook), Design System. - **Entrega rastreável** — PRDs, épicos, milestones com RAG status, task lifecycle workflow, review de PR co-assinado por Code Reviewer + AppSec. - **Observabilidade pós-entrega** — retros, incidentes 4-estado com timestamps auto-preenchidos. - **Análise de grafo** — `sdlc-trace` (matriz PRD↔spec↔task↔review) e `sdlc-impact` (BFS forward/backward). - **Dashboard HTML autocontido** — Kanban, Épicos & Milestones, Documentos, Domínio; sem servidor. - **Rastreamento de git worktrees e PRs** em SQLite. - **Hook PostToolUse** que indexa cada `.md` salvo, rate-limited a 1 sinal / 5 s por vault. - **i18n do `_INDEX.md`** — renderizado no idioma definido em `.sdlc-kit/marker.json:locale` (default `pt-br`, alternativo `en`). Código e templates em inglês; o LLM espelha o idioma da conversa no conteúdo. ### Desenvolvimento ```bash git clone https://github.com/vellus-ai/sdlc-kit.git cd sdlc-kit pip install -e ".[dev,yaml]" pytest ``` **Convenções:** branches `feat/<slug>` ou `fix/<slug>`; TDD por padrão; cada script emite JSON em stdout com exit codes `0` / `1` / `2`; cobertura mínima 90% em `core/`. ### Documentação - **Arquitetura** — [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) - **Contribuição** — [`docs/CONTRIBUTING.md`](docs/CONTRIBUTING.md) - **Publicação** — [`docs/PUBLISHING.md`](docs/PUBLISHING.md) - **Testes** — [`docs/TESTING.md`](docs/TESTING.md) - **Política de privacidade** — [`PRIVACY.md`](PRIVACY.md) - **Changelog** — [`CHANGELOG.md`](CHANGELOG.md) ### Licença [MIT](LICENSE) © 2026 [Vellus](https://vellus.tech/)