7.0 KiB
Banatie Content Repository
Overview
This is a content repository for Banatie blog and website. Content is created by 10 Claude Desktop agents. You (Claude Code) manage files and structure.
Core principle: One markdown file = one article. Files move between stage folders like kanban cards.
Your Role: Repository Manager
You do NOT write articles — agents do that. You manage files, validate structure, check consistency.
Repository Structure
banatie-content/
├── CLAUDE.md ← You are here
├── content-framework.md ← System architecture documentation
├── human-editing-checklist.md ← Human editing guide
├── batch-processing.md ← Intensive workflow guide
│
├── project-knowledge/ ← Static context (for Claude Desktop Projects)
│ ├── project-soul.md
│ ├── banatie-product.md
│ ├── target-audience.md
│ ├── competitors.md
│ └── research-tools-guide.md
│
├── shared/ ← Dynamic context (for operational updates)
│ └── (empty by default)
│
├── desktop-agents/ ← Agent configs (10 agents)
│ ├── 000-spy/
│ ├── 001-strategist/
│ ├── 002-architect/
│ ├── 003-writer/
│ ├── 004-editor/
│ ├── 005-seo/
│ ├── 006-image-gen/
│ ├── 007-style-guide-creator/
│ ├── 008-webmaster/
│ └── 009-validator/
│
├── style-guides/ ← Author personas
│ ├── AUTHORS.md
│ └── {author}.md
│
├── assets/ ← Static assets
│ └── avatars/ ← Author avatar images
│
├── research/ ← @spy output
├── pages/ ← @webmaster output (landing pages)
│
├── 0-inbox/ ← Ideas
├── 1-planning/ ← Briefs
├── 2-outline/ ← Structures + Validation
├── 3-drafting/ ← Drafts + Revisions
├── 4-human-review/ ← Human editing
├── 5-seo/ ← SEO optimization
├── 6-ready/ ← Ready to publish (+ image specs)
└── 7-published/ ← Archive
Context Architecture
project-knowledge/
Static context files. Added to Claude Desktop Project Knowledge. Rarely changes. Contains: product info, audience, competitors, research tools guide.
shared/
Dynamic context. Agents read via MCP at /init. Used for: operational updates, experiment results, temporary priorities. Priority: shared/ overrides project-knowledge/
File Format
Each article is ONE file with frontmatter:
---
slug: my-article
title: "Article Title"
author: henry
status: drafting
created: 2024-12-22
updated: 2024-12-23
content_type: tutorial
primary_keyword: "main keyword"
---
# Idea
(raw idea, discovery source)
---
# Brief
(from @strategist)
---
# Outline
(from @architect)
---
# Validation Request
(from @architect — claims to verify)
---
# Validation Results
(from @validator — verification verdicts)
---
# Text
(final text — renamed from Draft after PASS)
---
# SEO Optimization
(from @seo)
---
# Image Specs
(from @image-gen)
---
# Review Chat
(comments from @strategist, @architect, @editor during final review)
Status Values
| Status | Folder | Meaning |
|---|---|---|
| inbox | 0-inbox/ | Raw idea |
| planning | 1-planning/ | Brief in progress |
| outline | 2-outline/ | Structure + validation in progress |
| drafting | 3-drafting/ | Writing |
| revision | 3-drafting/ | Revision after critique |
| review | 4-human-review/ | Human editing |
| seo | 5-seo/ | SEO optimization |
| ready | 6-ready/ | Ready to publish |
| published | 7-published/ | Archived |
Commands
/status
Show pipeline overview:
- Count files in each stage
- Flag issues (missing author, wrong status)
- List files needing attention
/validate {slug}
Check specific article:
- Frontmatter complete?
- Required sections for stage?
- Author style guide exists?
/move {slug} {stage}
Move article to stage (validate first):
- Check transition is allowed
- Update status in frontmatter
- Move file
/check-consistency
Validate repository structure and cross-references:
Checks:
- Agent folders (000-spy through 009-validator exist with required files)
- Stage folders (0-inbox through 7-published match documentation)
- Project knowledge files (all referenced files exist)
- Cross-references (paths in system-prompt.md, CLAUDE.md, content-framework.md are valid)
- Status values (frontmatter statuses match folder structure)
Output format:
Repository Consistency Check
============================
Agent Folders:
✓ 000-spy (system-prompt.md, agent-guide.md)
✓ 001-strategist (system-prompt.md, agent-guide.md)
...
✓ 009-validator (system-prompt.md, agent-guide.md)
Stage Folders:
✓ 0-inbox/
✓ 1-planning/
...
Project Knowledge:
✓ project-soul.md
✓ banatie-product.md
✗ Missing: some-file.md (referenced in X)
Cross-References:
✓ All paths valid
or
✗ Invalid path in 001-strategist/system-prompt.md: /wrong/path
Summary: {N} issues found
Files to fix: {list}
Language
- Files: English
- Communication with user: Russian
- Reports: Russian
The 10 Agents
| # | Agent | Role | Special Tools |
|---|---|---|---|
| 000 | @spy | Research, competitive intelligence | DataForSEO, Brave Search, Perplexity |
| 001 | @strategist | Topic planning, briefs | DataForSEO, Perplexity |
| 002 | @architect | Article structure, validation requests | — |
| 003 | @writer | Draft writing | — |
| 004 | @editor | Quality review | — |
| 005 | @seo | SEO optimization | DataForSEO, Brave Search |
| 006 | @image-gen | Visual asset specs | — |
| 007 | @style-guide-creator | Author personas | — |
| 008 | @webmaster | Landing pages, web content | Brave Search, Perplexity |
| 009 | @validator | Fact verification | Brave Search, Perplexity |
Pipeline Overview
@spy → @strategist → @architect → @validator → @writer → @editor → human → @seo → @image-gen → publish
Validation gate: After @architect, @validator must PASS before @writer starts. This prevents publishing unverified claims.
Review Chat System
After human editing, articles go through final review:
- @strategist, @architect, @editor review with
/reviewcommand - Each adds comment to Review Chat section
- All three must say "APPROVED" before publishing
What You Do NOT Do
❌ Write article content ❌ Create briefs or outlines ❌ Make editorial decisions ❌ Generate images ❌ Verify facts
These are done by Claude Desktop agents.
Key Documentation
content-framework.md— Full system architecture and design decisionshuman-editing-checklist.md— What human does after AI reviewbatch-processing.md— Workflow for content intensives