cm-content-factory

CM Content Factory v2.0 — AI Content Machine Platform

Config-driven, self-improving content factory with real-time dashboard, multi-agent independence, and token management. Gets smarter with use through memory + reward system.

Architecture

┌─────────────────────────────────────────────┐
│         🌐 DASHBOARD (localhost:5050)        │
│  Pipeline │ Tasks │ Tokens │ Logs │ Landing  │
└───────────┬─────────────────────────────────┘
            │ SSE / Polling
┌───────────┴─────────────────────────────────┐
│            🏭 PIPELINE ENGINE               │
│  ┌──────────┐ ┌──────────┐ ┌────────────┐  │
│  │  State   │ │  Token   │ │   Agent    │  │
│  │ Manager  │ │ Manager  │ │ Dispatcher │  │
│  └──────────┘ └──────────┘ └────────────┘  │
│                    │                         │
│  EXTRACT → PLAN → WRITE → AUDIT → SEO → PUB│
│                    │         │              │
│              📊 SCOREBOARD (reward/penalty)  │
│                    │                         │
│              🧠 MEMORY (3-layer learning)   │
└─────────────────────────────────────────────┘

Config file: content-factory.config.json at project root. Schema: config.schema.json.

🚀 Quick Start

# New project (interactive wizard)
python3 scripts/wizard.py

# Full pipeline WITH dashboard
python3 scripts/pipeline.py --dashboard

# Pipeline with budget limit
python3 scripts/pipeline.py --dashboard --budget 5.0

# Dashboard only (standalone)
python3 scripts/dashboard_server.py

Phase 0: Discovery (MANDATORY)

AI MUST ask 5 question groups in order:

| # | Group | Key Questions | |---|-------|---------------| | Q1 | Niche Info | Industry, brand, address, phone, USP | | Q2 | Reference & Avoid | Reference websites, styles to avoid, tone | | Q3 | Data Sources | Existing files, URLs to extract, images | | Q4 | Content Goals | Number of articles, keywords, language, region | | Q5 | Deploy | Cloudflare account, domain, milestone |

Phase 0.5: Confirm

Display summary table → WAIT for user OK → then proceed.

Operating Modes (12)

| Mode | Script | Purpose | |------|--------|---------| | 📦 EXTRACT | extract.py | Source docs → JSON knowledge-base | | 📋 PLAN | plan.py | Knowledge → topic queue | | ✍️ WRITE | write.py | AI content generation (batch/single) | | 🔍 AUDIT | audit.py | Quality check + auto-fix | | 🔎 SEO | seo.py | Metadata optimization | | 🚀 PUBLISH | publish.py | Build + deploy | | 🧠 LEARN | scoreboard.py + memory.py | Extract patterns from feedback | | 🔬 RESEARCH | research.py | Auto-research new topics | | 💰 REVIEW | monetize.py | Monetization scoring | | 🏭 PIPELINE | pipeline.py | Full automated A→Z | | 📊 DASHBOARD | dashboard_server.py | Real-time web dashboard | | 🎯 LANDING | landing_generator.py | Persona-based landing pages |

All scripts: python3 scripts/<script> --config content-factory.config.json

Dashboard (NEW in v2.0)

Real-time web dashboard at http://localhost:5050:

• Pipeline Progress: Visual 6-phase timeline with progress bars
• Task Queue: Active/queued/completed/failed task cards
• Token Tracker: Cost by provider, budget progress bar
• Event Log: Filterable real-time log viewer
• Error Panel: Highlighted error details

# Auto-start with pipeline
python3 scripts/pipeline.py --dashboard --dashboard-port 5050

# Standalone
python3 scripts/dashboard_server.py --port 5050

Multi-Agent Support (NEW in v2.0)

Multiple agents can work independently on the same pipeline via file-based task queue.

from agent_dispatcher import AgentDispatcher
d = AgentDispatcher()

# Enqueue tasks
d.enqueue("write-article-1", "write", {"topic": "SEO Tips"}, priority=3)
d.enqueue_batch([{"id": "w-2", "type": "write"}, {"id": "w-3", "type": "write"}])

# Agent claims next task
task = d.claim_next("gemini-agent-1")
d.heartbeat("gemini-agent-1", task["id"])  # Keep alive

# Complete or fail
d.complete(task["id"], "gemini-agent-1", {"result": "ok"})
d.fail(task["id"], "gemini-agent-1", "API timeout")  # Auto-retry up to 3x

Features: priority ordering, stale lock detection (10min), auto-retry (3x), heartbeat.

Token Management (NEW in v2.0)

Track token usage, costs, rate limits, and budget across all providers.

from token_manager import TokenManager
tm = TokenManager(budget_usd=5.0)

# Record usage
tm.record_usage("gemini", input_tokens=1000, output_tokens=500, task_id="w-1")

# Check budget
if not tm.check_budget():
    print("Budget exceeded!")

# Rate limiting
tm.wait_if_rate_limited("gemini")

# Circuit breaker (auto-stop after 5 consecutive failures)
if tm.is_circuit_open("gemini"):
    print("Provider down, switching...")

python3 scripts/token_manager.py status

Landing Pages (NEW in v2.0)

Generate persona-based landing pages using Content Mastery SB7 framework.

Add personas array to config:

{
  "personas": [{
    "name": "Economic Buyer",
    "headline": "Save 50% on Marketing Costs",
    "subheadline": "AI creates professional content, 10x faster",
    "pain_points": ["High marketing costs", "Lack of content staff"],
    "benefits": [{"title": "Cost Savings", "description": "Reduce costs by 50%"}],
    "social_proof": [{"number": "2,347", "label": "Businesses trust us"}],
    "steps": [{"title": "Configure", "description": "Enter your business information"}],
    "cta_text": "Try It Free"
  }]
}

python3 scripts/landing_generator.py --config content-factory.config.json
python3 scripts/landing_generator.py --config content-factory.config.json --list

Pipeline Execution

1. INIT     → wizard.py + scaffold.py → Astro project + config
2. RESEARCH → pipeline:research → 30+ topics
3. EXPAND   → expand-topics.py → 30 → target (100/200/...)
4. WRITE    → pipeline:write → 3 workers, 8s sleep, ~3 articles/min
5. MILESTONE → audit → build → deploy (at 50%/100%)
6. SHIP     → Final audit → deploy → notify user

🛡️ Golden Rules (Score 95+)

• Performance: Font preload, critical CSS inline, preconnect, img width/height, GTM defer
• Accessibility: WCAG AA contrast (#555+ on white), semantic HTML, h1-h3 hierarchy
• Security: CSP/HSTS/XFO headers via public/_headers, cache immutable for /_astro/*
• SEO: robots.txt → sitemap, proper meta tags

Self-Learning System

Memory (3 layers)

| Layer | Path | Purpose | |-------|------|---------| | Semantic | memory/semantic/ | Long-term patterns, style, SEO rules | | Episodic | memory/episodic/ | Per-session experiences + outcomes | | Working | memory/working/ | Current session context |

Scoreboard

| Event | Points | |-------|--------| | User praise | +10 | | Engagement (share/bookmark) | +5 | | Article passes audit first try | +3 | | User edits article | -5 | | User deletes article | -10 | | Audit fail | -3 |

Scripts Reference

Core Pipeline

| Script | Purpose | |--------|---------| | pipeline.py | Master orchestrator (6-phase) + dashboard integration | | extract.py | Source extraction | | plan.py | Topic planning | | write.py | AI content writer | | audit.py | Quality audit + fixer | | seo.py | SEO optimization | | validate.py | Content validation | | publish.py | Build + deploy | | deploy.py | Multi-platform deploy |

Platform Layer (NEW v2.0)

| Script | Purpose | |--------|---------| | state_manager.py | Central state management (JSON + JSONL events) | | token_manager.py | Token tracking, cost, rate limits, circuit breaker | | dashboard_server.py | HTTP server + SSE for dashboard | | agent_dispatcher.py | Multi-agent task queue with file locking | | landing_generator.py | Persona → Landing page generator |

Intelligence Layer

| Script | Purpose | |--------|---------| | memory.py | 3-layer memory engine | | scoreboard.py | Reward/penalty system | | research.py | Auto-research engine