Executing EPIC Workflow

1. Context

• Main Objective: Execute feature development using the EPIC methodology (Explore, Research, Plan, Validate, Implement, Review, Iterate)
• Secondary Objective: Ensure proper delegation to specialized subagents for each phase
• User Input: Feature description, requirements, or task specification
• Workflow: Explore → Research → Plan → Validate Plan → Implement → Review → Iterate (main agent only implements)

CRITICAL: Session Directory Initialization

BEFORE starting any EPIC phase, the main agent MUST:

1. Create session directory: .claude/sessions/[NN]-[session-description]/
   • [NN]: Two-digit sequential number (01, 02, 03, etc.)
   • [session-description]: Short hyphenated description (e.g., user-auth-feature, payment-integration)
2. Store the session directory path for use throughout the workflow
3. ALWAYS instruct ALL subagents to save their reports to this session directory

Example:

• Session directory: .claude/sessions/01-user-auth-feature/
• When delegating to any subagent, ALWAYS include: "Save your report to .claude/sessions/01-user-auth-feature/[required-filename].md"

2. Workflow

Phase 1: Explore

Objective: Gather comprehensive context about the codebase and existing implementations

• T001: Initialize session directory [P0]

  • Determine next sequential number by checking existing .claude/sessions/ directories
  • Create new session directory: .claude/sessions/[NN]-[session-description]/
  • Example: .claude/sessions/01-user-auth-feature/
  • Store this path as SESSION_DIR for use in all subsequent phases

• T002: Delegate exploration to codebase-explorer agent [P0]

  • CRITICAL: Include explicit save instruction in delegation prompt:

    "Please analyze the current project status, identify relevant files and components,
    assess recent changes and technical dependencies, and document the current state
    of related features.
    
    IMPORTANT: Save your complete exploration report to:
    [SESSION_DIR]/codebase-status.md
    
    The report must be saved to this exact location for workflow validation."
    

  • Request analysis of current project status
  • Identify relevant files and components
  • Assess recent changes and technical dependencies
  • Document current state of related features
  • Agent MUST save report to: [SESSION_DIR]/codebase-status.md

• T003: Review exploration findings [P0]

  • Read the generated report: [SESSION_DIR]/codebase-status.md
  • Synthesize discovered information
  • Identify gaps or areas needing clarification
  • Prepare context for research phase

• T004: Validate phase completion [P0]

  • Run: python .claude/skills/epic/scripts/validate-phase.py explore [SESSION_DIR]
  • ITERATIVE COMPLIANCE FLOW:
    • If validation PASSES: Proceed to Phase 2 (Research)
    • If validation FAILS:
      1. Reinvoke codebase-explorer agent with EXPLICIT instruction: "Save your report to [SESSION_DIR]/codebase-status.md"
      2. Re-run validation script
      3. Repeat steps 1-2 until validation passes
    • CRITICAL: Do NOT proceed to next phase until validation passes

Phase 2: Research

Objective: Conduct comprehensive research on complex topics and validate approaches

• T005: Delegate research tasks to research-specialist agent [P0]

  • CRITICAL: Include explicit save instruction in delegation prompt:

    "Please conduct comprehensive research on [specific topics], validate approaches
    across multiple sources, perform deep web investigations if needed, and synthesize
    findings into actionable insights.
    
    IMPORTANT: Save your complete research report to:
    [SESSION_DIR]/research-report.md
    
    The report must be saved to this exact location for workflow validation."
    

  • Conduct research on complex topics
  • Validate approaches across multiple sources
  • Perform deep web investigations if needed
  • Synthesize findings into actionable insights
  • Agent MUST save report to: [SESSION_DIR]/research-report.md

• T006: Review research findings [P0]

  • Read the generated report: [SESSION_DIR]/research-report.md
  • Identify best practices and patterns
  • Document technical recommendations
  • Prepare foundation for planning phase

• T007: Validate phase completion [P0]

  • Run: python .claude/skills/epic/scripts/validate-phase.py research [SESSION_DIR]
  • ITERATIVE COMPLIANCE FLOW:
    • If validation PASSES: Proceed to Phase 3 (Plan)
    • If validation FAILS:
      1. Reinvoke research-specialist agent with EXPLICIT instruction: "Save your report to [SESSION_DIR]/research-report.md"
      2. Re-run validation script
      3. Repeat steps 1-2 until validation passes
    • CRITICAL: Do NOT proceed to next phase until validation passes

Phase 3: Plan

Objective: Develop comprehensive implementation strategy

• T008: Delegate strategic planning to strategic-planner agent [P0]

  • CRITICAL: Include explicit save instruction in delegation prompt:

    "Please analyze the problem comprehensively using the exploration and research
    findings from [SESSION_DIR]/codebase-status.md and [SESSION_DIR]/research-report.md.
    Devise optimal solution approaches, identify multiple implementation paths, and
    evaluate trade-offs and risks.
    
    IMPORTANT: Save your complete implementation plan to:
    [SESSION_DIR]/implementation-plan.md
    
    The plan must be saved to this exact location for workflow validation."
    

  • Provide paths to exploration and research reports for context
  • Analyze problem comprehensively using exploration and research findings
  • Devise optimal solution approaches
  • Identify multiple implementation paths
  • Evaluate trade-offs and risks
  • Agent MUST save report to: [SESSION_DIR]/implementation-plan.md

• T009: Review and consolidate plan [P0]

  • Read the generated plan: [SESSION_DIR]/implementation-plan.md
  • Integrate insights from exploration and research
  • Verify comprehensive implementation roadmap exists
  • Confirm success criteria and validation steps are defined

• T010: Validate phase completion [P0]

  • Run: python .claude/skills/epic/scripts/validate-phase.py plan [SESSION_DIR]
  • ITERATIVE COMPLIANCE FLOW:
    • If validation PASSES: Proceed to Phase 4 (Validate Plan)
    • If validation FAILS:
      1. Reinvoke strategic-planner agent with EXPLICIT instruction: "Save your plan to [SESSION_DIR]/implementation-plan.md"
      2. Re-run validation script
      3. Repeat steps 1-2 until validation passes
    • CRITICAL: Do NOT proceed to next phase until validation passes

Phase 4: Validate Plan

Objective: Critical review and validation of proposed approach

• T011: Delegate plan validation to consulting-expert agent [P0]

  • CRITICAL: Include explicit save instruction in delegation prompt:

    "Please review the implementation plan at [SESSION_DIR]/implementation-plan.md
    objectively. Identify potential risks and over-complications, suggest pragmatic
    alternatives, and validate alignment with best practices.
    
    IMPORTANT: Save your complete validation feedback to:
    [SESSION_DIR]/validation-feedback.md
    
    The feedback must be saved to this exact location for workflow validation."
    

  • Provide path to implementation plan for review
  • Review proposed approaches objectively
  • Identify potential risks and over-complications
  • Suggest pragmatic alternatives
  • Validate alignment with best practices
  • Agent MUST save report to: [SESSION_DIR]/validation-feedback.md

• T012: Refine plan based on validation feedback [P0]

  • Read the validation feedback: [SESSION_DIR]/validation-feedback.md
  • Address identified concerns
  • Simplify over-complicated approaches
  • Update [SESSION_DIR]/implementation-plan.md with refinements if needed
  • Finalize implementation strategy

• T013: Validate phase completion [P0]

  • Run: python .claude/skills/epic/scripts/validate-phase.py validate [SESSION_DIR]
  • ITERATIVE COMPLIANCE FLOW:
    • If validation PASSES: Proceed to Phase 5 (Implement)
    • If validation FAILS:
      1. Reinvoke consulting-expert agent with EXPLICIT instruction: "Save your feedback to [SESSION_DIR]/validation-feedback.md"
      2. Re-run validation script
      3. Repeat steps 1-2 until validation passes
    • CRITICAL: Do NOT proceed to next phase until validation passes

Phase 5: Implement

Objective: Execute the implementation directly as the main agent

• T014: Set up implementation tracking with TodoWrite tool [P0]

  • Read the finalized plan: [SESSION_DIR]/implementation-plan.md
  • Break down plan into actionable tasks
  • Create todo list with clear status tracking
  • Mark tasks as in_progress when working on them

• T015: Execute implementation following the plan [P0]

  • Write code according to specifications from [SESSION_DIR]/implementation-plan.md
  • Follow established patterns from [SESSION_DIR]/codebase-status.md
  • Implement one task at a time
  • Mark todos as completed immediately after finishing

• T016: Handle errors and blockers [P1]

  • Keep tasks as in_progress if encountering issues
  • Create new tasks for blockers that need resolution
  • Never mark incomplete work as completed

• T017: Document implementation completion [P0]

  • Add inline comments where logic isn't self-evident
  • Update relevant documentation files if needed
  • Note any deviations from original plan
  • CRITICAL: Main agent creates implementation summary:

    Save a summary of what was implemented, key decisions made, and any
    deviations from the plan to:
    [SESSION_DIR]/implementation-complete.md
    
    This file is required for workflow validation.
    

• T018: Validate phase completion [P0]

  • Run: python .claude/skills/epic/scripts/validate-phase.py implement [SESSION_DIR]
  • ITERATIVE COMPLIANCE FLOW:
    • If validation PASSES: Proceed to Phase 6 (Review)
    • If validation FAILS (missing implementation-complete.md):
      1. Main agent creates [SESSION_DIR]/implementation-complete.md with summary
      2. Re-run validation script
      3. Repeat until validation passes
    • CRITICAL: Do NOT proceed to next phase until validation passes

Phase 6: Review

Objective: Validate implementation quality through specialized review

• T019: Delegate code review to appropriate review agent [P0]

  • CRITICAL: Include explicit save instruction in delegation prompt:

    "Please review the implementation comprehensively. Review the code changes,
    check for code quality, performance, and security issues. Reference the
    implementation summary at [SESSION_DIR]/implementation-complete.md.
    
    IMPORTANT: Save your complete quality review to:
    [SESSION_DIR]/quality-report.md
    
    The report must be saved to this exact location for workflow validation.
    Include all findings, recommendations, and severity levels."
    

  • For code quality: Use general review processes
  • For security: Consider security-expert agent
  • Request comprehensive feedback on implementation
  • Agent MUST save report to: [SESSION_DIR]/quality-report.md

• T020: Delegate testing validation to test-engineer agent if tests exist [P1]

  • CRITICAL: Include save instruction in delegation prompt
  • Ensure tests pass for critical business logic
  • Validate test coverage
  • Review test data and mocking patterns
  • Results should be appended to: [SESSION_DIR]/quality-report.md

• T021: Compile review findings [P0]

  • Read the quality report: [SESSION_DIR]/quality-report.md
  • Document all feedback from review agents
  • Prioritize issues by severity
  • Prepare for iteration phase if needed

• T022: Validate phase completion [P0]

  • Run: python .claude/skills/epic/scripts/validate-phase.py review [SESSION_DIR]
  • ITERATIVE COMPLIANCE FLOW:
    • If validation PASSES: Proceed to Phase 7 (Iterate) if issues found, or complete workflow if no issues
    • If validation FAILS:
      1. Reinvoke review agents with EXPLICIT instruction: "Save your report to [SESSION_DIR]/quality-report.md"
      2. Re-run validation script
      3. Repeat steps 1-2 until validation passes
    • CRITICAL: Do NOT proceed to next phase until validation passes

Phase 7: Iterate

Objective: Address feedback and resolve issues until quality standards are met

• T023: Delegate troubleshooting to troubleshooter agent if issues found [P0]

  • CRITICAL: Include explicit save instruction in delegation prompt if issues exist:

    "Please diagnose and resolve the errors/bugs/build failures found in the
    quality report at [SESSION_DIR]/quality-report.md.
    
    OPTIONAL: If troubleshooting is performed, save your troubleshooting report to:
    [SESSION_DIR]/troubleshooting-report.md
    
    This file is optional but helpful for documentation."
    

  • Diagnose errors, bugs, or build failures
  • Resolve runtime exceptions
  • Fix configuration problems
  • Agent MAY save report to: [SESSION_DIR]/troubleshooting-report.md (optional)

• T024: Address review feedback (main agent) [P0]

  • Read the quality report: [SESSION_DIR]/quality-report.md
  • Implement suggested improvements
  • Fix identified issues
  • Update [SESSION_DIR]/implementation-complete.md with changes made

• T025: Repeat review phase if significant changes made [P1]

  • If major modifications were implemented, return to Phase 6
  • Ensure changes don't introduce regressions
  • This creates an iteration loop: Review → Iterate → Review until quality passes

• T026: Final verification [P0]

  • Confirm all success criteria from [SESSION_DIR]/implementation-plan.md are met
  • Verify no regressions introduced
  • CRITICAL: Main agent creates final verification summary:

    Save a summary confirming all criteria are met, no regressions exist,
    and the implementation is complete to:
    [SESSION_DIR]/final-verification.md
    
    This file is required for workflow validation.
    

• T027: Validate phase completion [P0]

  • Run: python .claude/skills/epic/scripts/validate-phase.py iterate [SESSION_DIR]
  • ITERATIVE COMPLIANCE FLOW:
    • If validation PASSES: EPIC workflow complete ✅
    • If validation FAILS:
      1. Main agent creates [SESSION_DIR]/final-verification.md with summary
      2. Re-run validation script
      3. Repeat until validation passes
    • CRITICAL: Do NOT mark workflow complete until validation passes

3. Implementation Strategy

Agent Delegation Pattern

CRITICAL: ALL delegation prompts MUST include explicit save instructions to SESSION_DIR

Phase 1 - Explore:

• Use Task tool with subagent_type="codebase-explorer"
• MUST include in prompt: "Save your report to [SESSION_DIR]/codebase-status.md"
• Provide specific investigation goals
• Request comprehensive analysis of current state

Phase 2 - Research:

• Use Task tool with subagent_type="research-specialist"
• MUST include in prompt: "Save your report to [SESSION_DIR]/research-report.md"
• Conduct research on complex topics
• Validate approaches across multiple sources

Phase 3 - Plan:

• Use Task tool with subagent_type="strategic-planner"
• MUST include in prompt: "Save your plan to [SESSION_DIR]/implementation-plan.md"
• Provide paths to previous reports for context
• Analyze problem using exploration and research findings
• Devise optimal solution approaches

Phase 4 - Validate Plan:

• Use Task tool with subagent_type="consulting-expert"
• MUST include in prompt: "Save your feedback to [SESSION_DIR]/validation-feedback.md"
• Provide path to implementation plan for review
• Review proposed approaches objectively
• Identify risks and suggest alternatives

Phase 5 - Implement:

• Main agent executes directly (NO delegation)
• Read plan from [SESSION_DIR]/implementation-plan.md
• Use TodoWrite tool to track progress
• Follow plan strictly without deviation
• Main agent MUST save: [SESSION_DIR]/implementation-complete.md
• Ask clarifying questions via AskUserQuestion tool when needed

Phase 6 - Review:

• Use Task tool with appropriate review agent for code review
• MUST include in prompt: "Save your report to [SESSION_DIR]/quality-report.md"
• Provide path to implementation summary
• Use Task tool with subagent_type="test-engineer" for testing validation
• Compile all feedback from review agents

Phase 7 - Iterate:

• Use Task tool with subagent_type="troubleshooter" if issues found (optional)
• Provide path to quality report
• Address review feedback (main agent)
• Return to Phase 6 if significant changes made
• Main agent MUST save: [SESSION_DIR]/final-verification.md
• Final verification when all checks pass

Progressive Task Tracking

• Create todo list at start of implementation phase
• Update task status in real-time
• Exactly ONE task in_progress at any time
• Mark tasks completed immediately upon finishing
• Never batch completion updates

Iterative Compliance Validation

CRITICAL Pattern: Validate → If Fail → Reinvoke → Repeat

After EVERY phase, run the validation script:

• python .claude/skills/epic/scripts/validate-phase.py <phase> <session-dir>

If validation FAILS:

1. Identify which required file(s) are missing
2. Reinvoke the responsible subagent with explicit instruction to create missing file(s)
3. Re-run validation script
4. Repeat steps 1-3 until validation PASSES
5. Do NOT proceed to next phase until validation passes

This iterative flow ensures:

• Subagents complete their assigned tasks
• All required documentation is created
• Compliance is enforced at every phase
• No phase can be skipped or incomplete

Session Directory Structure

All subagent reports MUST be saved to: .claude/sessions/[NN]-[session-description]/

Required files by phase:

• Phase 1 (Explore): codebase-status.md
• Phase 2 (Research): research-report.md
• Phase 3 (Plan): implementation-plan.md
• Phase 4 (Validate): validation-feedback.md
• Phase 5 (Implement): implementation-complete.md
• Phase 6 (Review): quality-report.md
• Phase 7 (Iterate): final-verification.md

Strict Plan Adherence

• Follow the plan exactly as designed
• Do not implement beyond defined scope
• Do not improvise or add unplanned features
• Stop and ask if uncertain about any task
• Simple/lean approach over complex solutions

4. Constraints

• CRITICAL: Main agent ONLY implements - MUST delegate all other phases (Explore, Research, Plan, Validate, Review, Iterate)
• CRITICAL: Follow EXACT phase sequence - Explore → Research → Plan → Validate Plan → Implement → Review → Iterate
• CRITICAL: Run validation script after EVERY phase - do NOT proceed until validation PASSES
• CRITICAL: If validation fails, MUST reinvoke subagents until they create required files
• CRITICAL: Iterative compliance flow is MANDATORY - cannot skip or bypass validation
• CRITICAL: All subagent reports MUST be saved to session directory: .claude/sessions/[NN]-[session-description]/
• CRITICAL: Use TodoWrite tool throughout implementation to track progress
• CRITICAL: Follow plan strictly - no deviation or improvisation
• CRITICAL: Do not implement tasks beyond the defined scope
• CRITICAL: Mark exactly ONE task as in_progress at a time
• CRITICAL: Complete tasks immediately upon finishing (no batching)
• Do NOT skip exploration phase - comprehensive context gathering is required
• Do NOT skip research phase - research must happen BEFORE planning
• Do NOT skip planning phase - thorough strategy development is mandatory
• Do NOT skip validation phase - plan must be validated before implementation
• Do NOT skip review phase - validation and code review are essential
• Do NOT skip iteration phase - feedback must be addressed
• Do NOT proceed to next phase if validation fails - reinvoke subagents iteratively
• Do NOT create documentation files unless explicitly requested (except required phase reports)
• Do NOT add emojis unless user explicitly requests them
• Do NOT over-engineer - build for MVP with simple/lean approach
• Do NOT implement features, refactoring, or improvements beyond what was asked
• Only add comments where logic isn't self-evident
• Only validate at system boundaries (user input, external APIs)
• Delete unused code completely - no backwards-compatibility hacks
• Ask clarifying questions via AskUserQuestion tool when uncertain

5. Success Criteria

Phase 1 - Exploration Success

• Comprehensive codebase context gathered via codebase-explorer agent
• Relevant files and components identified
• Current state and dependencies documented
• Exploration findings reviewed and synthesized

Phase 2 - Research Success

• Research conducted via research-specialist agent
• Complex topics investigated thoroughly
• Approaches validated across multiple sources
• Best practices and patterns identified
• Technical recommendations documented

Phase 3 - Planning Success

• Strategic plan developed via strategic-planner agent
• Problem analyzed comprehensively using exploration and research
• Multiple implementation paths identified
• Trade-offs and risks evaluated
• Consolidated implementation roadmap created
• Success criteria defined

Phase 4 - Validation Success

• Plan reviewed via consulting-expert agent
• Potential risks and over-complications identified
• Pragmatic alternatives suggested where needed
• Plan refined based on validation feedback
• Final implementation strategy approved

Phase 5 - Implementation Success

• Todo list created with all implementation tasks
• All planned tasks executed following specifications
• Code follows established patterns from exploration
• Todos marked as completed immediately after finishing
• No tasks left as in_progress if actually completed
• Implementation decisions documented appropriately
• No features or improvements added beyond scope

Phase 6 - Review Success

• Code review completed via appropriate review agent
• Testing validation performed if tests exist
• All review findings documented
• Issues prioritized by severity
• Feedback compiled for iteration phase

Phase 7 - Iteration Success

• Troubleshooting completed via troubleshooter agent (if needed)
• All review feedback addressed and implemented
• Issues and bugs resolved
• Re-review performed if significant changes made
• No regressions introduced
• Final verification confirms completion

Overall Success

• All seven EPIC phases completed in correct sequence
• Proper delegation to specialized agents for all phases except Implementation
• Main agent handled implementation phase directly
• Research completed BEFORE planning
• Plan validated BEFORE implementation
• Plan followed strictly without deviation
• User requirements fully satisfied
• Code quality validated through review and iteration
• Documentation updated where required