Agent skill

plan-feature

Stars 1
Forks 0

Install this agent skill to your Project

npx add-skill https://github.com/DieGopherLT/dotclaudefiles/tree/main/plugins/smart-plan/skills/plan-feature

SKILL.md

Plan Feature - 5-Phase Planning Workflow

Operate as the Smart Plan orchestrator. Orchestrate the planning of a feature through 5 structured phases, delegating work to specialized agents and coordinating their outputs. The final output is a self-contained implementation plan approved by the user.

Feature request: $ARGUMENTS


Phase 1: Discovery

Goal: Understand the feature request and establish tracking.

  1. Create tasks (TaskCreate) for all 5 phases:
    • Phase 1: Discovery
    • Phase 2: Codebase Exploration
    • Phase 3: Smart Interview
    • Phase 4: Architecture Design
    • Phase 5: Plan Mode
  2. Set up task dependencies: each phase is blocked by the previous one
  3. Mark Phase 1 as in_progress
  4. Summarize your understanding of the feature to the user and confirm it is correct
  5. Mark Phase 1 as completed

Phase 2: Codebase Exploration

Goal: Build a comprehensive understanding of the codebase relevant to this feature.

  1. Mark Phase 2 as in_progress

  2. Launch 3 agents in parallel using the Task tool:

    Agent 1 - code-explorer (textual, broad):

    Explore the codebase to understand architecture, patterns, and features similar to: <actual feature request from $ARGUMENTS>.
    Focus on: project structure, naming conventions, error handling, testing patterns, configuration.
    Do NOT produce a formatted report. Only list the file paths you identified as essential and a one-line note per file explaining its relevance.
    

    Agent 2 - code-explorer (textual, feature-focused):

    Find and trace complete execution flows of features most similar to: <actual feature request from $ARGUMENTS>.
    For each similar feature, trace from entry point through all layers to data storage/external calls.
    Do NOT produce a formatted report. Only list every file path involved and which layer it belongs to.
    

    Agent 3 - code-indexer (semantic/LSP):

    Build a semantic index of the codebase areas most relevant to: <actual feature request from $ARGUMENTS>.
    Focus on: type dependencies, interface contracts, call hierarchy, shared symbols.
    Do NOT produce a formatted report. Only list file paths, key symbols, and type relationships discovered.
    
  3. After all 3 agents complete, read all essential files they identified (Read tool) to have them in your own context

  4. Consolidate findings internally. Present a brief summary to the user (key files, patterns detected, similar features found)

  5. Mark Phase 2 as completed


Phase 3: Smart Interview

Goal: Elicit quantifiable requirements, business rules, and flows before designing architecture.

  1. Mark Phase 3 as in_progress
  2. Invoke the Skill tool with smart-plan:smart-interview
    • The skill will identify gaps, conduct structured interview rounds, and consolidate confirmed results
    • The results will be structured and ready for the ## Requirements section of the plan
  3. Mark Phase 3 as completed

Phase 4: Architecture Design

Goal: Design the implementation approach with multiple perspectives.

  1. Mark Phase 4 as in_progress

  2. Launch 2-3 code-architect agents in parallel, each with a different approach:

    Architect 1 - Minimal Changes:

    Design the architecture for: <actual feature request from $ARGUMENTS>
    Approach: MINIMAL CHANGES - maximum reuse of existing code, minimum new files.
    
    Context from exploration:
    <insert consolidated findings from Phase 2>
    
    Decisions made:
    <insert all decisions documented in Phase 3>
    
    Produce the full Architecture Blueprint as specified in your instructions.
    

    Architect 2 - Clean Architecture:

    Design the architecture for: <actual feature request from $ARGUMENTS>
    Approach: CLEAN ARCHITECTURE - proper separation of concerns, new abstractions where they add value.
    
    Context from exploration:
    <insert consolidated findings from Phase 2>
    
    Decisions made:
    <insert all decisions documented in Phase 3>
    
    Produce the full Architecture Blueprint as specified in your instructions.
    

    Architect 3 - Pragmatic Balance (optional, use when the feature spans 3+ new files or introduces a new architectural layer):

    Design the architecture for: <actual feature request from $ARGUMENTS>
    Approach: PRAGMATIC BALANCE - reuse where sensible, abstract where it adds clear value.
    
    Context from exploration:
    <insert consolidated findings from Phase 2>
    
    Decisions made:
    <insert all decisions documented in Phase 3>
    
    Produce the full Architecture Blueprint as specified in your instructions.
    
  3. Review all architecture proposals

  4. Form your own recommendation with reasoning

  5. Present all approaches to the user with your recommendation highlighted

  6. Use AskUserQuestion to let the user choose the approach

  7. Mark Phase 4 as completed


Phase 5: Plan Mode

Goal: Write a comprehensive, self-contained implementation plan using the plan template.

  1. Mark Phase 5 as in_progress
  2. Invoke the Skill tool with smart-plan:smart-delegation — it will provide delegation guidelines to embed in the plan (task grouping, parallelization, model recommendations)
  3. Resolve branch context for every repository involved in the plan:
    • If the initial prompt specified the repository/repositories and base branch(es): use them directly
    • If not: use AskUserQuestion to ask for each repository: which base branch to use, and confirm the feat/<name> branch name
  4. Read the plan template from references/plan-template.md (relative to this skill's directory) for structure guidance
  5. Call EnterPlanMode directly (do NOT launch a Plan sub-agent; operate as the planner directly)
  6. Write the plan using the template structure as a base:
    • Fill ## Repository Context with the resolved branch information
    • Incorporate the delegation guidelines from step 2
    • Fill the ## Requirements section with the requirements, rules, and flows produced by Phase 3 (smart-interview)
  7. Ensure the plan is fully self-contained: anyone reading it must be able to execute it without additional context
  8. Call ExitPlanMode to request user approval
  9. If the user requests changes, modify the plan and re-submit
  10. Mark Phase 5 as completed after approval

Orchestration Rules

  • Always track progress: Update tasks (TaskUpdate) as you start and complete each phase
  • Never skip phases: Even if a phase seems unnecessary, execute it (it may reveal something)
  • Respect dependencies: Do not start a phase until the previous one is completed
  • Consolidate agent outputs: After agents return, synthesize their findings before presenting to user
  • User approval at key points: Phases 1 (understanding), 4 (architecture choice), 5 (plan approval)
  • Plan completeness: The plan must be self-contained with all instructions needed for execution
  • Fail gracefully: If an agent fails or returns poor results, inform user and offer to retry or adjust
  • Be transparent: Show the user what is happening at each phase; do not work silently

Expand your agent's capabilities with these related and highly-rated skills.

DieGopherLT/dotclaudefiles

smart-interview

Esta skill debe usarse cuando el usuario pregunta "¿Preguntas?", "¿Tienes dudas?", "¿Tienes preguntas?", "¿Quieres aclarar algo?", "¿Necesitas aclarar algo?", "aclara lo que necesites", o cuando quiere aterrizar requerimientos antes de planear. Tambien la invoca plan-feature en Phase 3. Ejecuta entrevista estructurada para obtener requerimientos cuantificables, reglas de negocio traducibles a codigo, y flujos del sistema; luego anota los resultados en el plan.

1 0
Explore
DieGopherLT/dotclaudefiles

post-implementation

Esta skill debe usarse cuando el usuario pide "haz review de los cambios", "review the changes", "aplica el post-implementacion", "apply post-implementation", "run post-implementation", o cuando el plan aprobado instruye "invoke /smart-plan:post-implementation". Ejecuta workflow post-implementacion despues de que el codigo este completo con quality review (3 reviewers paralelos), auto-fix de issues (confianza >= 80%), documentacion del feature, y commit opcional. Puede usarse despues de cualquier implementacion, no solo smart-plan.

1 0
Explore
DieGopherLT/dotclaudefiles

smart-delegation

Esta skill debe usarse cuando Claude detecta que una implementacion es lo suficientemente grande como para dividirla en sub-agentes (5+ archivos, dependencias entre cambios, nuevas abstracciones que otros archivos consumen), cuando el usuario pide "delega la implementacion", "orquesta los implementadores", "usa sub-agentes", "delegate to sub-agents", "orchestrate implementers", o cuando plan-feature invoca la delegacion tras aprobar un plan.

1 0
Explore
DieGopherLT/dotclaudefiles

tdd-workflow

Esta skill debe usarse cuando el usuario pide "aplicar TDD", "desarrollo guiado por tests", "test-driven development", "red green refactor", "ciclo RGR", "escribir tests primero", "nueva feature con TDD", "mejorar testabilidad", "filosofia TDD", o menciona Iron Laws, TDD philosophy, o quiere entender el proceso TDD. NO usar cuando el usuario pide agregar tests a un proyecto existente o setup de testing — eso es testing:add-testing.

1 0
Explore
DieGopherLT/dotclaudefiles

add-testing

Esta skill debe usarse cuando el usuario pide "agregar tests a proyecto existente", "add testing to existing project", "setup de testing", "configurar infraestructura de tests", "pipeline completo de testing", "quiero tests para este modulo", "agregar cobertura de tests", "add test coverage", "integrar testing al proyecto". Ejecuta pipeline completo de 7 pasos: auditoria de testabilidad, adaptacion condicional de codigo, investigacion de dependencias, implementacion de tests, auditoria de calidad de inputs, reporte de cobertura y generacion de reglas del proyecto.

1 0
Explore
DieGopherLT/dotclaudefiles

team-setup

Esta skill debe usarse cuando el usuario pide 'configura un team', 'prepara equipo de agentes', 'crea un team de agentes', 'setup del team', 'team de agentes para', 'orquesta agentes', 'team-setup', 'arma un equipo', 'necesito un team para', o cuando quiere coordinar multiples agentes trabajando en repos/directorios distintos. Genera el prompt estructurado para el TeamLead incluyendo roles, workspaces, contexto inicial y reglas universales del equipo. Detecta automaticamente si el objetivo es negociacion de contratos (endpoints, sockets, SSEs) y activa el flujo especializado de debate balanceado.

1 0
Explore

Didn't find tool you were looking for?

Be as detailed as possible for better results