Create clear, actionable technical execution plans with mandatory quality review before display.

Phase 1: Planning (REQUIRED FIRST)

ALWAYS create a comprehensive plan BEFORE taking any action.

Best Practices:

•Analyze requirements, constraints, existing systems
•Decompose into specific, actionable steps
•Identify sequential vs parallel work
•Anticipate edge cases and failure modes
•Define testable completion conditions

Plan Structure

1. High-Level Overview (Always Visible)

Must include:

•Visual Diagram: Mermaid diagram (flowchart, sequence, component)
•System Summary: 2-3 sentences
•Key Components: List major components
•Data Flow: How data moves through system

Example:

code

🏗️ System Architecture Overview

[Mermaid Flowchart Here]

**System Summary**: Chrome extension that captures webpage elements and saves them to a persistent canvas using Shadow DOM isolation and chrome.storage for cross-context state management.

**Key Components**:
- Background Worker (service worker, manages storage)
- Content Script (injected into pages, captures elements)
- Canvas UI (React app, displays saved elements)
- Storage Layer (chrome.storage.local, source of truth)

**Data Flow**: User triggers capture → Content script extracts element → Background worker persists to storage → Canvas UI reacts to storage change → UI updates

2. Executive Summary (Always Visible)

Must include:

•Problem Statement: What and why
•Proposed Solution: High-level approach (2-3 sentences)
•Key Technical Decisions: Critical architectural choices
•Estimated Complexity: Rough time/effort
•Dependencies: External libraries, APIs, systems

3. Prerequisites & Context (Collapsible)

•Files/directories to examine
•Existing code patterns to understand
•Dependencies or blockers
•Environment setup requirements

4. Step-by-Step Implementation (Always Visible)

•Numbered steps in logical order
•Specific and actionable
•Include file paths and function names
•Mark dependencies
•Indicate parallel steps
•Brief technical notes where needed

Example Step Format:

html

<li>
  <strong>Step 3:</strong> Create <code>src/content/ElementSelector.tsx</code> - React component for hover overlay

  <div class="indent-1">
    <strong>Key functions</strong>:
    - <code>handleMouseMove(e)</code>: Update overlay position based on cursor
    - <code>handleClick(e)</code>: Capture element HTML and metadata
    - <code>injectStyles()</code>: Add styles to Shadow DOM
  </div>

  <div class="indent-1 muted">
    File: src/content/ElementSelector.tsx
  </div>

  <div class="tech-note">
    <strong>Technical Note:</strong> Use absolute positioning with z-index: 2147483647 to ensure overlay appears above all page content. Inject styles into Shadow DOM to prevent page CSS conflicts.
  </div>
</li>

5. Testing & Validation (Collapsible)

•How to verify each major step
•Test cases (happy path and edge cases)
•Expected outcomes
•Common failures and debugging

6. Potential Issues & Mitigations (Collapsible)

•Edge cases
•Common pitfalls
•Fallback strategies
•Performance considerations
•Security considerations

7. Post-Implementation (Collapsible)

•Documentation needed
•Follow-up tasks
•Maintenance considerations
•Monitoring/observability

Phase 2: MANDATORY PLAN REVIEW

CRITICAL: Do NOT write HTML until AFTER review.

Review Flow

•Draft plan (in memory)
•Launch Codex MCP review (preferred) OR Task agent (fallback)
•Wait for review results
•Incorporate feedback
•Write HTML file

Using Codex MCP for Review (Preferred):

typescript

mcp__codex__codex({
  prompt: `Review the following technical implementation plan and provide critical analysis:

[PLAN SUMMARY - Include key sections: Overview, Executive Summary, Implementation Steps]

Analyze:
1. **Technical Accuracy**: Are the proposed solutions technically sound?
   - Check for API misuse, incorrect algorithms, architecture anti-patterns

2. **Implementation Completeness**: Are there missing steps or edge cases?
   - Look for unstated assumptions, missing error handling, incomplete workflows

3. **Architecture**: Is this the best architectural choice?
   - Consider scalability, maintainability, testability tradeoffs

4. **Security Considerations**: Any security issues with the approach?
   - Check for injection vulnerabilities, authentication gaps, data exposure

5. **Code Quality**: Review code snippets for correctness and best practices
   - Check syntax, idioms, performance implications

6. **Testing Strategy**: Is the testing approach comprehensive enough?
   - Verify test coverage of happy path, edge cases, error conditions

7. **Risk Assessment**: What are the biggest risks?
   - Identify show-stoppers, performance bottlenecks, user impact

Provide specific, actionable recommendations for improvement.`,
  cwd: "[current working directory]",
});

Fallback: Task Agent Review (if Codex unavailable):

typescript

Task({
  subagent_type: "general-purpose",
  description: "Review and critique plan",
  prompt: `Review the following technical implementation plan:

[PLAN SUMMARY]

Provide:
1. **Strengths**: Well-thought-out aspects
2. **Potential Issues**: Problems or edge cases missed
3. **Suggestions**: How to improve the plan
4. **Risk Assessment**: Biggest technical risks
5. **Alternative Approaches**: Better ways to accomplish this

Be critical. Look for:
- Missing error handling
- Overlooked dependencies
- Performance/scalability concerns
- Security vulnerabilities
- Testing gaps
- Unclear technical specifications`,
});

Detection Logic:

javascript

// Check if Codex MCP is available
if (typeof mcp__codex__codex === 'function') {
  // Use Codex MCP review (preferred for technical accuracy)
  await mcp__codex__codex({...});
} else {
  // Fallback to Task agent
  await Task({subagent_type: "general-purpose", ...});
}

Incorporating Feedback

•Analyze critique - identify valid concerns
•Update plan - address major issues
•Document changes
•
Add "Review Summary" section:
- •Key feedback received
- •Changes made
- •Risks acknowledged but accepted

Phase 3: HTML Output (After Review)

CRITICAL: Use plan template at ${CLAUDE_PLUGIN_ROOT}/templates/html/plan-template.html

Workflow

•Read plan template
•Replace {{TITLE}}
•Fill content in 
•Use Mermaid for diagrams
•Critical info visible, details collapsed
•Save to claude_html/ and open

Phase 4: File Handling

•Create directory: mkdir -p claude_html
•Write to: claude_html/plan_[task_description].html
•Open: open claude_html/plan_[task_description].html
•Inform user

Phase 5: User Confirmation

•Present plan (includes review summary)
•Ask: "Proceed with implementation / Make revisions / Clarify decisions?"
•Wait for confirmation
•Update plan HTML if changes requested

Principles

•NEVER skip planning
•ALWAYS review before display
•Diagram first
•Be specific (file paths, function names)
•Anticipate issues
•Wait for review results
•Incorporate feedback
•Use plan template
•Progressive disclosure
•User approval required

Workflow

code

User Request
  → Draft Plan (memory, with diagram)
  → Review (Codex MCP or Task agent)
  → Incorporate Feedback
  → Read Plan Template
  → Write HTML
  → Open File
  → User Reviews
  → User Approves
  → Execute