feature

Agent Skills › di-sukharev/vibe › feature

feature

GitHub

端到端脚手架并实现新功能，涵盖契约、数据库、后端、前端及测试。适用于新增实体、界面或API的全栈开发任务。

.agents/skills/feature/SKILL.md di-sukharev/vibe

Trigger Scenarios

用户请求创建新的产品功能（如实体、屏幕或API）需要从零开始生成全栈代码结构并实现业务逻辑

Install

npx skills add di-sukharev/vibe --skill feature -g -y

More Options

Non-standard path

npx skills add https://github.com/di-sukharev/vibe/tree/master/.agents/skills/feature -g -y

Use without installing

npx skills use di-sukharev/vibe@feature

指定 Agent (Claude Code)

npx skills add di-sukharev/vibe --skill feature -a claude-code -g -y

安装 repo 全部 skill

npx skills add di-sukharev/vibe --all -g -y

预览 repo 内 skill

npx skills add di-sukharev/vibe --list

SKILL.md

Frontmatter

{
    "name": "feature",
    "description": "Scaffolds and implements a new full-stack product capability (an entity, a screen, or an API) end to end, from the contract and Prisma schema through the backend, the webapp, checks, and the CHECKLIST registry row. Use when the user asks for a new capability shaped like an owned resource; for cross-cutting work, follow the auth slice by hand instead."
}

Steps

Confirm scope. CHECKLIST defines product scope; a capability without a registry row is absent. Build only what was requested. Ask only the product questions you cannot infer.
Read the "New feature" section of ARCHITECTURE and the users reference slice (backend/src/modules/users, webapp/src/features/users).
Run bun run scaffold:feature -- <plural-kebab-name>. Run it with --dry-run first, check the printed singular, model, table, and route, and rerun with --singular <word> when one is wrong. ARCHITECTURE describes what it generates, its options, and when it refuses. Do not use it for cross-cutting work (no single owner, or it touches many existing features) — follow the auth slice by hand instead.
Shape the contract in packages/contracts. Finish the generated Prisma model in backend/prisma/schema/<name>.prisma, then run bun run --cwd backend prisma:migrate -- --name add_<name>. State compatibility, risk, and rollout order for the schema change.
Work test-first, as AGENTS.md requires: for each real rule (a permission, a validation, a state change), write a failing HTTP integration test, run it, confirm it fails for the expected reason, then implement the service and repository until it passes.
Build the webapp side:
- the API client and queries, with keys and invalidation;
- UI with loading, empty, error-with-retry, and success states, plus pending and result states for every mutation;
- model logic in pure functions with unit tests.
Run the checks: bun run typecheck:backend, bun run typecheck:webapp, bun run lint, bun run architecture:check, and the new tests. Then run bun run screens -- -g "<route>" and follow the ui-review skill.
Update the capability's row in CHECKLIST and any owning docs whose behavior changed.
Report as AGENTS.md Workflow step 8 requires: what changed, the checks and their results, risks, and any user action.