Agent Skills: Plan Feature - 5-Phase Planning Workflow

Esta skill debe usarse cuando el usuario pide "planear una feature", "plan a feature", "crear un plan de acción", "disenar una implementacion", "design an implementation", "crear un plan de desarrollo", "create a development plan", "quiero planear", "quiero hacer un plan", "quiero disenar", o cuando solicite un feature cuya complejidad lo requiera. Ejecuta workflow de 5 fases: discovery, exploracion del codebase, clarificacion, arquitectura, y escritura del plan. Al aprobar el plan, invoca smart-delegation automaticamente para ejecutar la implementacion.

UncategorizedID: DieGopherLT/claude-kit/plan-feature

Install this agent skill to your local

pnpm dlx add-skill https://github.com/DieGopherLT/dotclaudefiles/tree/HEAD/plugins/smart-plan/skills/plan-feature

Skill Files

Browse the full folder contents for plan-feature.

Download Skill

Loading file tree…

plugins/smart-plan/skills/plan-feature/SKILL.md

Skill Metadata

Name
plan-feature
Description
Esta skill debe usarse cuando el usuario pide "planear una feature", "plan a feature", "crear un plan de acción", "disenar una implementacion", "design an implementation", "crear un plan de desarrollo", "create a development plan", "quiero planear", "quiero hacer un plan", "quiero disenar", o cuando solicite un feature cuya complejidad lo requiera. Ejecuta workflow de 5 fases: discovery, exploracion del codebase, clarificacion, arquitectura, y escritura del plan. Al aprobar el plan, invoca smart-delegation automaticamente para ejecutar la implementacion.

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