The CLAUDE.md Orchestration File
The CLAUDE.md file lives in your project root. It tells Claude Code how to orchestrate the pipeline and when to stop for your approval.
CLAUDE.md Template
# [Your Brand] Content Pipeline
This repository contains the content pipeline for **[Your Brand/Publication]**.
**Tagline:** *[Your Tagline]*
## Brand Voice
All content produced through this pipeline must align with these core principles:
### [Principle 1]
- [Specific guidance]
- [What to do]
- [What to avoid]
### [Principle 2]
- [Specific guidance]
- [What to do]
- [What to avoid]
### [Principle 3]
- [Specific guidance]
- [What to do]
- [What to avoid]
## Content Guidelines
- **Tone:** [Describe your tone]
- **Audience:** [Who you write for]
- **Perspective:** [Your point of view]
- **Avoid:** [What to never do]
## Pipeline Command
When I say "/blog-pipeline [topic]", run these stages with MANDATORY STOPS:
### Stage 1: Research & Planning (Automated)
1. @content-classifier → classify the input
2. @content-researcher → gather evidence
**AUTO-PROCEED to Stage 2**
---
### Stage 2: First Draft
3. @content-writer → draft the post
4. @image-prompter → generate matching visuals
🛑 **CHECKPOINT: STOP HERE AND WAIT FOR MY FEEDBACK**
Display:
- The full draft
- Image prompts generated
- Ask: "Ready for critique, or do you have feedback on the draft first?"
Do NOT proceed until I explicitly say "continue", "proceed", "looks good", or "run critique"
---
### Stage 3: Critique & Revision
5. @content-reviewer → critique draft + image fit
🛑 **CHECKPOINT: STOP HERE AND WAIT FOR MY FEEDBACK**
Display:
- The critique/revision suggestions
- Ask: "Want me to apply these revisions, or do you have additional notes?"
Do NOT proceed until I explicitly confirm.
---
### Stage 4: Humanization
6. @content-humanizer → break AI patterns, preserve voice
🛑 **CHECKPOINT: STOP AND WAIT**
Display:
- Diff of changes made
- List of patterns fixed
- Ask: "Accept humanization, or adjust?"
---
### Stage 5: Final Prep (Automated)
7. @content-writer → apply approved revisions (if any)
8. @substack-publisher → prepare everything for publication
**END: Present final package for review**
---
## Gate Commands
At any checkpoint, I can say:
- "continue" / "proceed" / "looks good" → move to next stage
- "revise [specific feedback]" → apply my notes, show me again, stay at checkpoint
- "restart from [stage]" → go back to that stage
- "pause" → save state, we'll continue later
- "skip critique" → jump straight to publishing prep
## Important
NEVER auto-proceed past a 🛑 checkpoint. Always wait for explicit confirmation.
---
## Folder Structure
### Drafts
```
drafts/
article-slug/
classified.json # Topic classification
research.json # Research and evidence
draft.md # Working draft
```
### Published
```
published/
YYYY-MM-DD-article-slug/
YYYY-MM-DD-article.md # Main article
YYYY-MM-DD-meta.json # Metadata
YYYY-MM-DD-copy-paste-assets.md # Social media copy
YYYY-MM-DD-summary.md # Publication summary
YYYY-MM-DD-checklist.txt # Publication checklist
```🎯 Customize this: Replace [Your Brand], [Your Tagline], and all the bracketed placeholders with your specific content and principles.
🔵 Note: This file and @agent-name syntax are for Claude Code. Using Antigravity, Cursor, or Windsurf? See Module 5 for the /slash-command approach.