Dev Kit Agent Skill

Full-stack Development Kit theo chuẩn Agent Skills — tương thích Claude Code, Cursor, VS Code, Kiro, Gemini CLI, và 40+ agentic clients khác.

Cài đặt

Claude Code

Cách 1 — Global (dùng được ở mọi project):

git clone git@github.com:SoftwareOneHN/dev-kit.git ~/.claude/skills/dev-kit

Cách 2 — Project-level (chỉ áp dụng cho 1 project):

git clone git@github.com:SoftwareOneHN/dev-kit.git .claude/skills/dev-kit

Cách 3 — Symlink (dev workflow, dễ pull updates):

git clone git@github.com:SoftwareOneHN/dev-kit.git ~/Projects/dev-kit
ln -s ~/Projects/dev-kit ~/.claude/skills/dev-kit

Global vs Project-level: ~/.claude/skills/ là global — skill tự động available trong mọi project. .claude/skills/ (trong repo) là project-level — chỉ hoạt động khi bạn đang ở repo đó.

Verify: ls ~/.claude/skills/dev-kit/SKILL.md — thấy file là OK.

Cursor / VS Code

Copy thư mục dev-kit/ vào .cursor/skills/ hoặc .vscode/skills/.

Kiro

Copy thư mục dev-kit/ vào .kiro/skills/ hoặc cài từ Power registry.

Bất kỳ Agent Skills client nào

Đặt thư mục dev-kit/ vào vị trí skills mà client hỗ trợ.

Cấu trúc

dev-kit/
├── SKILL.md              # Required: metadata + core instructions
├── README.md             # File này
├── LICENSE               # MIT
└── references/           # Detailed docs (loaded on-demand)
    ├── _shared/
    │   └── workflow-artifacts.md
    ├── languages/
    │   ├── java/
    │   │   ├── core-rules.md
    │   │   ├── spring-boot.md
    │   │   ├── clean-architecture.md
    │   │   └── testing.md
    │   └── typescript/
    │       ├── core-rules.md
    │       ├── react-nextjs.md
    │       ├── vue-nuxt.md
    │       ├── api-patterns.md
    │       └── forms.md
    ├── brainstorm/
    │   └── SKILL.md
    ├── spec/
    │   └── SKILL.md
    ├── plan/
    │   ├── SKILL.md
    │   ├── archive-workflow.md
    │   ├── codebase-understanding.md
    │   ├── output-standards.md
    │   ├── plan-organization.md
    │   ├── red-team-personas.md
    │   ├── red-team-workflow.md
    │   ├── research-phase.md
    │   ├── solution-design.md
    │   ├── task-management.md
    │   ├── validate-question-framework.md
    │   ├── validate-workflow.md
    │   └── workflow-modes.md
    ├── cook/
    │   ├── SKILL.md
    │   ├── intent-detection.md
    │   ├── review-cycle.md
    │   ├── subagent-patterns.md
    │   └── workflow-steps.md
    ├── fix/
    │   ├── SKILL.md
    │   ├── complexity-assessment.md
    │   ├── diagnosis-protocol.md
    │   ├── mode-selection.md
    │   ├── parallel-exploration.md
    │   ├── prevention-gate.md
    │   ├── review-cycle.md
    │   ├── skill-activation-matrix.md
    │   ├── task-orchestration.md
    │   ├── workflow-ci.md
    │   ├── workflow-deep.md
    │   ├── workflow-logs.md
    │   ├── workflow-quick.md
    │   ├── workflow-standard.md
    │   ├── workflow-test.md
    │   ├── workflow-types.md
    │   └── workflow-ui.md
    ├── debug/
    │   ├── SKILL.md
    │   ├── defense-in-depth.md
    │   ├── frontend-verification.md
    │   ├── investigation-methodology.md
    │   ├── log-and-ci-analysis.md
    │   ├── performance-diagnostics.md
    │   ├── reporting-standards.md
    │   ├── root-cause-tracing.md
    │   ├── systematic-debugging.md
    │   ├── task-management-debugging.md
    │   └── verification.md
    ├── test/
    │   ├── SKILL.md
    │   ├── report-format.md
    │   ├── test-execution-workflow.md
    │   └── ui-testing-workflow.md
    ├── scout/
    │   ├── SKILL.md
    │   ├── external-scouting.md
    │   ├── internal-scouting.md
    │   └── task-management-scouting.md
    └── code-review/
        ├── SKILL.md
        ├── adversarial-review.md
        ├── checklist-workflow.md
        ├── checklists/
        │   ├── api.md
        │   ├── base.md
        │   └── web-app.md
        ├── code-review-reception.md
        ├── codebase-scan-workflow.md
        ├── edge-case-scouting.md
        ├── input-mode-resolution.md
        ├── parallel-review-workflow.md
        ├── requesting-code-review.md
        ├── spec-compliance-review.md
        ├── task-management-reviews.md
        └── verification-before-completion.md

Progressive Disclosure

Theo chuẩn Agent Skills, skill hoạt động theo 3 giai đoạn:

1. Discovery (~100 tokens): Chỉ load name + description từ frontmatter
2. Activation (~4000 tokens): Load toàn bộ SKILL.md body khi task match
3. Execution (on-demand): Load reference files cụ thể khi cần chi tiết step

Sử dụng

Sau khi cài, nói tự nhiên:

Bạn nói	Agent làm
"Brainstorm cách implement auth"	Scout → exact requirements → trade-off analysis → design doc
"Plan feature user profile"	Research → codebase analysis → solution design → phase plan
"Implement story login từ PRD"	Scout → plan → code → test → review → finalize
"Fix lỗi 500 ở endpoint /users"	Scout → diagnose root cause → fix → verify+prevent
"Debug tại sao test flaky"	Systematic debugging → root cause trace → report
"Run tests cho module auth"	Execute test suite → coverage → QA report
"Review PR #42"	Fetch diff → spec compliance → quality → red-team
"Scout codebase cho payment module"	Parallel agents → file discovery → structured report

Key Features

• Lifecycle-driven: brainstorm → plan → implement → test → review → ship
• Scout-first: always scan codebase before asking questions or making changes
• Evidence-based: no guessing, no "probably" — verify before claiming
• Adversarial: red-team review on every code change
• Side-effect-free: mandatory verification that changes don't break existing behavior
• Multi-mode: interactive, auto, fast, parallel workflows for different needs
• Language-agnostic: works with any language, framework, or tech stack
• Task-managed: native Claude Tasks for tracking multi-step workflows

License

MIT