creating-development-plans — скилл для Claude Code: как установить
Скилл формирует структурированный план разработки с делением на фазы, требованиями и QA-чек-листом. Он нужен, когда ассистенту требуется чёткий сценарий действий перед написанием кода.
npx skills add sammcj/agentic-coding --skill creating-development-plans -a claude-codeЧто делает creating-development-plans
Скилл заставляет агента исследовать репозиторий до написания кода. Он проверяет документацию и зависимости проекта, чтобы опереться на принятый стек. Затем модель разбирает требования задачи и возможные архитектурные риски. Итоговый документ сохраняется по пути ./docs/DEVELOPMENT_PLAN.md.
План делится на независимые фазы с чек-боксами для подзадач. Каждая фаза завершается самопроверкой и паузой для ручного ревью. Скилл фиксирует только этапы планирования и сознательно останавливает агента перед кодингом, пока человек не подтвердит структуру. В шаблоне запрещены заглушки и комментарии TODO.
В итоговый план входят стратегия тестирования и чеклист контроля качества. Для текста автор задаёт австралийский вариант английского языка без эмодзи. Агент не добавляет оценки сроков или даты. Это избавляет план от пустых обещаний.
Какие задачи решает
- спланировать этапы разработки новой функциональности
- составить файл
./docs/DEVELOPMENT_PLAN.mdперед написанием кода - разбить сложную задачу на проверяемые фазы с чек-листами
- сформировать чеклист контроля качества и протокол отладки
Примеры запросов агенту
- Составь development plan для рефакторинга модуля авторизации
- Сделай dev plan для добавления интеграции с платёжным шлюзом
- Сформируй план разработки новой фичи в docs/DEVELOPMENT_PLAN.md
Как установить creating-development-plans
1Через Skills CLI (Vercel)
npx skills add sammcj/agentic-coding --skill creating-development-plans -a claude-code2Вручную: положить файлы
~/.claude/skills/creating-development-plans/SKILL.mdПапка скилла целиком; для одного проекта — .claude/skills/creating-development-plans/
1Через Skills CLI (Vercel)
npx skills add sammcj/agentic-coding --skill creating-development-plans -a codex2Вручную: положить файлы
.agents/skills/creating-development-plans/SKILL.mdДля всех проектов — ~/.agents/skills/
1Через Skills CLI (Vercel)
npx skills add sammcj/agentic-coding --skill creating-development-plans -a cursor2Вручную: положить файлы
.cursor/skills/creating-development-plans/SKILL.mdCursor читает и .agents/skills/, и .claude/skills/
1GitHub CLI
gh skill install sammcj/agentic-coding creating-development-plans2Вручную: положить файлы
.github/skills/creating-development-plans/SKILL.md1Через Skills CLI (Vercel)
npx skills add sammcj/agentic-coding --skill creating-development-plans -a gemini-cli2Командой платформы
gemini skills install https://github.com/sammcj/agentic-coding --path Skills/creating-development-plansВручную: положить файлы
.devin/skills/creating-development-plans/SKILL.mdСтарый путь .windsurf/skills/ тоже читается
Через Skills CLI (Vercel)
npx skills add sammcj/agentic-coding --skill creating-development-plans -a opencodeClaude.ai
Упакуйте папку в zip и загрузите в Settings → Capabilities → Skills
По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов
Вопросы о creating-development-plans
Как установить creating-development-plans в Claude Code?
Выполните команду npx skills add sammcj/agentic-coding --skill creating-development-plans -a claude-code.
Что делает creating-development-plans?
Скилл анализирует контекст репозитория и генерирует файл ./docs/DEVELOPMENT_PLAN.md. В нём задача разбивается на независимые фазы с критериями самопроверки и контрольными точками для ревью.
Почему агент останавливается после каждой фазы плана?
Шаблон требует обязательной контрольной точки для подтверждения человеком в конце каждого этапа. Это защищает проект от неконтролируемого кодинга без согласования архитектурных решений.
Исходный текст
SKILL.md · Apache-2.0
--- name: creating-development-plans description: Creates structured development plans with phased task breakdowns, requirements, and QA checklists. Use when the user explicitly asks to create a dev plan, development plan, or document development requirements. disable-model-invocation: true --- # Development Planning Skill <role> ## Core Principles - **Planning occurs before code**: Thoroughly understand project context and requirements first - **Context gathering is critical**: Always start by understanding the existing codebase and documentation - **Phased approach**: Break work into discrete, manageable phases with human review checkpoints - **Simplicity over complexity**: Keep solutions free of unnecessary abstractions - **Actionable output**: The plan must be clear enough for another senior AI agent to execute independently </role> <instructions-part-one> ## Planning Process ### Step 1: Context Gathering If there is existing code in the project: 1. Read all relevant files in the project directory 2. Examine existing documentation (README.md, docs/, CONTRIBUTING.md, etc.) 3. Analyse codebase structure, architecture, and dependencies 4. Identify coding conventions, patterns, and standards used 5. Review existing tests to understand expected behaviour 6. Note package versions and technology stack choices ### Step 2: Requirements Analysis Based on your conversation with the user: 1. Identify the core goal and objectives 2. List hard requirements explicitly stated 3. Document any unknowns or assumptions 4. Consider edge cases and architectural implications 5. Evaluate multiple implementation approaches and trade-offs (performance, maintainability, complexity) 6. Identify integration points with existing code 7. Clarify any ambiguous requirements with the user before proceeding ### Step 3: Task Breakdown Organise development into phases: - Each phase should be independently testable and reviewable - Break down complex tasks into sub-tasks (use nested checkboxes) - Identify dependencies between tasks - Order tasks logically within each phase - Each phase MUST end with: - A self-review checkpoint - A "STOP and wait for human review" checkpoint ### Step 4: Quality Assurance Planning Build a concise QA checklist that includes (if applicable): - Standard items (listed below) - Project-specific requirements gathered from conversation - Technology-specific checks (e.g., "Go vet passes" for Go projects, "ESLint clean" for JavaScript) - Security considerations mentioned - Any other quality gates discussed with the user ### Step 5: Deep Review Before finalising: 1. Use "ultrathink" to deeply consider: - Implementation approach soundness - Potential architectural issues - Constraint satisfaction - Alignment to requirements - Missing considerations - Complexity mitigations - Over-engineering risks 2. Make necessary adjustments to the plan ## Development Plan Structure Create a new file called `./docs/DEVELOPMENT_PLAN.md` with this structure: </instructions-part-one> <plan-template> ```markdown # Development Plan for [PROJECT_NAME] ## Project Purpose and Goals [Clear statement of what this project aims to achieve and why] ## Context and Background [Important background information, architectural context, constraints, research findings, and design decisions made during discussion] ## Development Tasks ### Phase 1: [Phase Name] - [ ] Task 1 - [ ] Sub-task 1.1 (if needed) - [ ] Sub-task 1.2 (if needed) - [ ] Task 2 - [ ] Task 3 - [ ] Perform a self-review of your code, once you're certain it's 100% complete to the requirements in this phase mark the task as done. <!-- - [ ] STOP and wait for human review # (Unless the user has asked you to complete the entire implementation) (only add this to the end of phases requested)--> ### Phase 2: [Phase Name] - [ ] Task 1 - [ ] Task 2 - [ ] Perform a self-review of your code, once you're certain it's 100% complete to the requirements in this phase mark the task as done. [Additional phases as needed] ## Important Considerations & Requirements - [ ] Do not over-engineer the solution, simple solutions are preferred - [ ] If a change means increasing complexity, weigh up options to reduce or mitigate that complexity - [ ] Do not add placeholder or TODO code - [ ] [Additional requirements from conversation] - [ ] [Architectural constraints] - [ ] [Integration requirements] ## Technical Decisions [Document any key technical decisions, trade-offs considered, and rationale for chosen approaches explicitly agreed upon with the user] ## Testing Strategy [Describe testing approach - should be lightweight, fast, and run without external dependencies] ## Debugging Protocol If issues arise during implementation: - **Tests fail**: Analyse failure reason and fix root cause, do not work around - **Performance issues**: Profile and optimise critical paths - **Integration issues**: Check dependencies and interfaces - **Unclear requirements**: Stop and seek clarification ## QA Checklist - [ ] All user instructions followed - [ ] All requirements implemented and tested - [ ] No critical code smell warnings - [ ] Australian English spelling used throughout - [ ] No smart formatting, non-ascii characters or emojis - [ ] Code follows project conventions and standards - [ ] Documentation is updated and accurate if needed - [ ] Security considerations addressed - [ ] Integration points verified (if applicable) - [ ] [Project-specific QA criteria based on technology stack] - [ ] [Additional QA criteria from user requirements] ``` </plan-template> <instructions-part-two> ## Writing Guidelines - Use dashes with single spaces for markdown lists: `- [ ] Task` - Do not include dates or time estimates - Be clear, concise, and actionable - Write in Australian English - Use technical terminology consistently - Avoid vague language - be specific about what needs to be done ## Quality Gates Adjust based on project risk tolerance: - **High-risk production systems**: Strict QA, extensive testing, security audits - **Internal tools/local development**: Lighter QA, focus on functionality - **Open source contributions**: Follow project's contribution guidelines precisely - **Prototypes/experiments**: Minimal QA, emphasis on learning and iteration ## Testing Philosophy - Lightweight and fast - No external dependencies required - Tests should run in isolation - When possible behavioural tests should be written before the code - Cover critical paths and edge cases - Integration tests for key workflows (if applicable) ## Final Steps 1. Write the complete `docs/DEVELOPMENT_PLAN.md` file 2. Apply deep thinking to review the plan thoroughly 3. Make any necessary adjustments including those to reduce complexity and over-engineering 4. Present the plan to the user 5. **STOP** and wait for user review ## Remember - This is a **planning document**, not implementation - The user will review and potentially iterate on this plan - Another AI agent (or you, in a future session) will execute this plan - Clarity and completeness are paramount but keep it concise - When in doubt about requirements, ask the user for clarification </instructions-part-two>
© sammcj · Apache-2.0 · sammcj/agentic-coding