Gerador de Portfólio E2E de Testes a partir de PRD
CharmYue/claude-skill-e2e-scenario-design
Transforma PRD em cenários E2E priorizados com asserções frontend+backend
Ver no GitHubSkill Claude que transforma um PRD em um portfólio priorizado de cenários de teste end-to-end. Seleciona automaticamente entre filosofia promise-based (10-15 cenários) ou feature-grid (30-80+) conforme a forma do produto, aplicando disciplina de fidelidade de sinal e fan-out multi-ator para detectar bugs de consistência de estado.
Instalação
npx convez-ai CharmYue/claude-skill-e2e-scenario-design1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Gerar plano de testes de regressão para SaaS antes do launch
- Validar cobertura de cenários em produtos com múltiplos atores/permissões
- Estruturar portfólio de testes com matriz de ferramentas (Playwright/Claude em Chrome)
Veredito

Skill maduro e validado em projetos open-source reais (Cal.com). Abordagem principiada (fidelidade de sinal, seleção de filosofia, fan-out de atores) reduz cenários inúteis e captura bugs de estado. Excelente para times que precisam dimensionar planos de teste sem guesswork, mas demanda PRD bem estruturada como entrada.
README
A tradução para o português deste README ainda está na fila.
e2e-scenario-design-from-prd
English | 简体中文
A Claude Skill that turns a PRD (or any product doc) into a prioritized E2E test scenario portfolio — 10-15 scenarios for promise-based products, 30-80+ for feature-grid SaaS, philosophy auto-selected by product shape.
Three load-bearing ideas, earned through three validation rounds against real open-source projects:
- Signal fidelity. Every UI assertion is paired with a backend assertion. A scenario that asserts only on "✓" badges passes through embarrassing production bugs.
- Philosophy selection. Small compliance product → promise-based (10-15). Mature SaaS → feature-grid (30-80+). Don't mix them up.
- Actor fan-out. Multi-role products need per-actor happy paths, cross-actor permission tests, and asymmetric workflows — all verified in single scenarios so state-consistency bugs surface.
See docs/philosophy.md for the narrative version.
What it produces
A single Markdown portfolio with:
- Feature inventory (≤25 clusters or raw items).
- 10-15 or 30-80+ scenarios (philosophy-dependent) grouped as P0 / P1 / P2.
- Dropped candidates — 3-5 items with one-line rejection reasons.
- Tool matrix — scenario × primary tool × secondary, with Playwright / Claude in Chrome / hybrid labeling.
Each scenario has: Priority, Category, Goal (user-voice), Preconditions, Steps, Assertions table (≥2 rows, ≥1 Backend), Tool, Known risks, Out of scope.
A real example generated by the skill for Cal.com lives at
examples/sample-portfolio.md.
Install
Claude Code plugin marketplace
/plugin marketplace add CharmYue/claude-skill-e2e-scenario-design
/plugin install e2e-testing@kylo-testing-toolkit
Manual (flat skill)
git clone https://github.com/CharmYue/claude-skill-e2e-scenario-design
mkdir -p ~/.claude/skills/e2e-scenario-design-from-prd
cp claude-skill-e2e-scenario-design/skill/SKILL.md \
~/.claude/skills/e2e-scenario-design-from-prd/SKILL.md
The flat-skill path (skill/SKILL.md) also works with OpenCode and Codex
skill loaders.
Keeping two copies in sync
Both the plugin manifest and the flat skill point at the same SKILL.md.
If you edit the skill locally under ~/.claude/skills/, run:
./scripts/sync-skill.sh
Usage
Invoke the skill with one of:
/e2e-scenario-design-from-prd- A PRD / spec / design doc pasted in + a question like "design E2E scenarios for this product"
- Phrases like "测试策略"、"E2E 测试设计"、"scenario 拆分"
- "My SaaS / feature-grid / CRUD product needs a regression-depth test plan"
The skill will:
- Gather inputs across 5 layers (README → docs sidebar → pricing → settings → API/webhooks).
- Build a feature inventory and decide the philosophy.
- Draft 10-15 or 30-80+ scenarios depending on that decision.
- Run a sanity checklist (coverage completeness, within-cluster audit, partial-heavy classification, actor fan-out, four falsification probes).
- Output a portfolio and a tool matrix.
Validation history
This skill was not designed in a vacuum. Three rounds:
| Round | Target | Scenarios | Score | What it taught |
|---|---|---|---|---|
| v1 | Compliance-sensitive data pipeline | 12 | — | Signal fidelity; UI-lying bugs |
| v2 | Documenso (e-signature) | 13 | 6.5 / 10 | Philosophy selection; actor fan-out |
| v3 | Cal.com (scheduling SaaS) | 52 | 8.0 / 10 | Within-cluster granularity; partial classification |
Full write-ups:
docs/reference-project.md— the audit that produced the original principles.docs/validation-documenso.md— round 2.docs/validation-calcom.md— round 3.
Repo layout
.
├── .claude-plugin/
│ └── marketplace.json # Claude Code marketplace manifest
├── plugins/e2e-testing/
│ ├── .claude-plugin/plugin.json
│ └── skills/e2e-scenario-design-from-prd/SKILL.md
├── skill/SKILL.md # Flat copy for OpenCode / Codex
├── docs/
│ ├── philosophy.md
│ ├── reference-project.md
│ ├── validation-documenso.md
│ └── validation-calcom.md
├── examples/sample-portfolio.md
├── scripts/sync-skill.sh
├── CHANGELOG.md
├── LICENSE
└── README.md
License
Contributing
The skill is version-controlled. If you extend or correct it, please:
- Run the skill on a real product (not a toy PRD).
- Compare your portfolio to the product's real E2E suite.
- Write a
docs/validation-<product>.mdbefore changingSKILL.md.
Two validation rounds (Documenso v2, Cal.com v3) produced 8 of the 11 rules added after v1 — 5 from Documenso, 3 from Cal.com. Evidence-first beats opinion-first.