OpenClaw — Autonomous Executive Operating System The AI-powered operating system that runs InBharat's entire growth, content, outreach, and intelligence operation.
Version 5.0 | Last updated: 2026-04-04 | github.com/inbharatai/OpenClaw-AI-CMO
┌─────────────────────────────────────────────────────────────────────┐
│ FOUNDER (WhatsApp / Terminal) │
└──────────────────────────┬──────────────────────────────────────────┘
▼
┌─────────────────────────────────────────────────────────────────────┐
│ OPENCLAW RUNTIME │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────────────────┐ │
│ │ Gateway │ │ WhatsApp │ │ Browser Automation │ │
│ │ Port :18789 │→ │ Baileys │ │ Playwright Sessions │ │
│ │ Agent Router │ │ Provider │ │ LinkedIn│X│Insta│Zoho │ │
│ └──────────────┘ └──────────────┘ └──────────────────────────┘ │
│ │ │
│ ┌────────────────────────▼─────────────────────────────────────┐ │
│ │ INBHARAT BOT — Internal Intelligence Layer │ │
│ │ 13 Intelligence Lanes: Media Engine, India Problems, │ │
│ │ AI Gaps, Funding, Compete Monitor, Ecosystem Scanner, │ │
│ │ Community Intel, Outreach, Gov & Tenders, Learning, │ │
│ │ Prototype Builder, Reddit Drafting, Revenue Pipeline │ │
│ └───────────────────────────────────────────────────────────────┘ │
│ │ │
│ ┌────────────────────────▼─────────────────────────────────────┐ │
│ │ CONTENT FACTORY (openclaw-media/) │ │
│ │ │ │
│ │ Generation: Guardrails: Publishing: │ │
│ │ ┌──────────────┐ ┌──────────────────┐ ┌─────────────┐ │ │
│ │ │Ollama qwen3 │ │policy_enforcer.py│ │publish.sh │ │ │
│ │ │Native Pipeline│ │rate-limits.json │ │(CANONICAL) │ │ │
│ │ ├──────────────┤ ├──────────────────┤ ├─────────────┤ │ │
│ │ │DALL-E 3 + │ │sanitize_post.py │ │render_post │ │ │
│ │ │enrich_prompt │ │render_post.py │ │qa-guardrail │ │ │
│ │ ├──────────────┤ ├──────────────────┤ ├─────────────┤ │ │
│ │ │ffmpeg Video │ │qa-guardrail.sh │ │dist-engine │ │ │
│ │ │HeyGen Briefs │ │claim-validator │ │(non-browser) │ │ │
│ │ └──────────────┘ │approval L1-L4 │ └─────────────┘ │ │
│ │ └──────────────────┘ │ │
│ └───────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────┘
Product
What it does
Website
Status
InBharat.ai AI consulting, tools & automation for Bharat
inbharat.ai Live
Sahaayak Multilingual AI assistant for Indian users
sahaayak.ai Development
Sahaayak Seva Public service AI for government schemes
—
Development
TestsPrep AI-powered test preparation for Indian exams
testsprep.in Development
UniAssist University and higher education AI assistant
uniassist.ai Development
Phoring AI-enhanced communication platform
phoring.in Development
CodeIn AI coding assistance and education
codein.pro Experimental
OpenClaw This system — autonomous ops engine
GitHub Development
Agent Arcade AI agent experimentation platform
GitHub Experimental
Sahayak OS Framework for deploying AI assistants
—
Experimental
Product facts: strategy/product-registry.json (10 products)
Product claims: strategy/product-truth/*.md (10 files, with safe/restricted claims)
Brand identity: strategy/brand-knowledge-base.json
Platform rules: strategy/platform-rules/*.md (6 files)
Publishing Pipeline (Canonical Path) Every live post goes through this exact path — no exceptions:
Queue (approved/) → POLICY CHECK → CLAIM VALIDATE → SANITIZE → QA GUARDRAIL → RENDER → POST → RECORD
policy_enforcer claim-validator sanitize_post qa-guardrail render_post post_*.py policy counter
Gate
File
What it checks
Policy policy_enforcer.pyblocked=true? Daily cap reached? Platform allowed?
Claims claim-validator.shFabricated stats? Unverified claims? Credentials leaked?
Sanitize sanitize_post.pyJSON leaks? YAML frontmatter? Template placeholders? Internal metadata?
QA qa-guardrail.shPost too long? Too many hashtags? Banned phrases?
Render render_post.pyTwo-stage: queue file (JSON/MD) → clean human-readable text only
Individual posting scripts (post_linkedin.py, post_x.py, etc.) require --allow-direct-post flag for direct invocation. Without it, they refuse to post and log the attempt. Even with the flag, policy enforcement still runs.
Platform
Engine
Posting Mode
Daily Cap
Session
LinkedIn post_linkedin.pyPlaywright
3/day
~/.openclaw/browser-sessions/linkedin/
X/Twitter post_x.pyPlaywright
3/day
~/.openclaw/browser-sessions/x/
Instagram post_instagram.pyPlaywright (mobile)
1/day
~/.openclaw/browser-sessions/instagram/
Discord post_discord.pyWebhook
3/day
Keychain: discord-webhook
Zoho Mail email_zoho.pyPlaywright (visible)
1/day
~/.openclaw/browser-sessions/zoho/
Reddit Draft only
Manual (L3 always)
1/day
—
Website distribution-engine.shFile copy to staging
3/day
—
HeyGen Brief export only
Founder-gated (T3)
2/day
—
Two Publishers, No Overlap
Script
Handles
Method
publish.shLinkedIn, X, Instagram, Discord
Playwright browser automation + webhook
distribution-engine.shWebsite, email, heygen, medium, substack
File staging + export
Both enforce policy_enforcer.py. Both run sanitize_post.py. Neither can touch the other's platforms.
Content Package → enrich_image_prompt.py → DALL-E 3 → Post Image
(brand colors, product (or placeholder
context, style, negative fallback if offline)
guidance, platform sizing)
Component
What it does
enrich_image_prompt.pyTransforms generic briefs into brand-aware DALL-E prompts
brand-knowledge-base.jsonColors, visual style, prohibited styles per product
process-briefs.shBatch processes all image briefs from content packages
generate-image.shRoutes to DALL-E 3, local Stable Diffusion, or placeholder
placeholder_generate.pyOffline branded card generator (loads brand colors from KB)
Level
Decision
Content Types
L1 Auto-approve
Standard posts, Discord, community updates
L2 Score-gated
Product claims, industry commentary, social posts
L3 Founder review
Reddit, HeyGen, bold claims, partnership, outreach
L4 Hard block
Unverified claims, credentials, PII, legal/political
Risk scoring: 6 dimensions (source confidence 25%, claim sensitivity 25%, brand voice 15%, data safety 15%, duplication 10%, platform risk 10%).
Time
Job
Script
Every 15 min
Self-healing watchdog
watchdog-agent.sh
8:07 AM daily
Full CMO pipeline
daily-pipeline.sh
9:00 AM daily
Auto-content generation
daily-auto-content.sh
Every 6h
Session keepalive
session-keepalive.sh
/Users/reeturajgoswami/Desktop/CMO-10million/
├── CLAUDE.md ← System operating instructions
├── README.md ← THIS FILE
├── assets/brand/ ← Logo files (SVG, PNG)
├── OpenClawData/
│ ├── directives/ ← Operating directives
│ ├── policies/ ← rate-limits.json, channel-policies, brand-voice-rules
│ ├── strategy/
│ │ ├── brand-knowledge-base.json ← Canonical brand identity (10 products)
│ │ ├── product-registry.json ← Product catalog
│ │ ├── product-truth/*.md ← Per-product safe/restricted claims + visual identity
│ │ ├── platform-rules/*.md ← Per-platform format, tone, limits (6 files)
│ │ └── content-templates/*.md ← Reusable campaign templates (10 files)
│ ├── inbharat-bot/ ← Intelligence layer (13 lanes)
│ ├── openclaw-media/
│ │ ├── posting-engine/ ← Platform posters + enforcement layer
│ │ │ ├── publish.sh ← THE canonical publish path
│ │ │ ├── policy_enforcer.py ← Runtime policy enforcement
│ │ │ ├── sanitize_post.py ← Content sanitization
│ │ │ ├── render_post.py ← Two-stage rendering
│ │ │ ├── direct_post_gate.py ← Gate for direct script calls
│ │ │ ├── metadata_fields.py ← Shared metadata constants
│ │ │ ├── post_linkedin.py ← Playwright poster (gated)
│ │ │ ├── post_x.py ← Playwright poster (gated)
│ │ │ ├── post_instagram.py ← Playwright poster (gated)
│ │ │ ├── post_discord.py ← Webhook poster (gated)
│ │ │ └── email_zoho.py ← Playwright poster
│ │ ├── image-engine/
│ │ │ ├── enrich_image_prompt.py ← Brand-aware prompt enrichment
│ │ │ ├── dalle_generate.py ← DALL-E 3 backend
│ │ │ ├── placeholder_generate.py ← Offline fallback (brand colors from KB)
│ │ │ ├── generate-image.sh ← Backend router
│ │ │ └── process-briefs.sh ← Batch processor with enrichment
│ │ ├── native-pipeline/ ← Content generation (Ollama)
│ │ ├── analytics/ ← Post logs, error screenshots
│ │ └── generated-images/ ← DALL-E 3 output
│ ├── queues/ ← Per-platform content queues
│ │ └── {platform}/pending|approved|posted|rejected/
│ ├── scripts/
│ │ ├── daily-pipeline.sh ← Master pipeline orchestrator
│ │ ├── watchdog-agent.sh ← Self-healing agent (every 15 min)
│ │ ├── distribution-engine.sh ← Non-browser distribution (policy-enforced)
│ │ ├── qa-guardrail.sh ← Pre-publish QA gate
│ │ ├── approval-engine.sh ← L1-L4 approval scoring
│ │ ├── session-keepalive.sh ← Browser session refresh (every 6h)
│ │ └── reporting-engine-v2.sh ← Daily/weekly reports
│ ├── skills/ ← 76 skill templates
│ ├── security/ ← Claim validator
│ ├── tests/ ← test_sanitize_post.py, test_enrich_image_prompt.py
│ ├── reports/ ← Generated reports
│ └── logs/ ← Execution + policy enforcement logs
└── MarketingToolData/ ← Staged website content + research
Component
Technology
Cost
LLM (content gen)
Ollama qwen3:8b (local)
Free
LLM (coding)
Ollama qwen2.5-coder:7b (local)
Free
Images
DALL-E 3 (OpenAI API)
~$0.04/image, 10/day cap
Video (local)
ffmpeg v7.0 at ~/local/bin/ffmpeg
Free
Video (avatar)
HeyGen
Paid, founder-gated
Browser posting
Playwright persistent sessions
Free
Discord posting
Webhook (curl)
Free
Secrets
macOS Keychain
Free
Scheduling
crontab
Free
CI/CD
GitHub Actions
Free
Estimated monthly cost : ~$12-15 (DALL-E at max usage). Everything else is free/local.
Component
Status
Policy enforcement Enforced at runtime — policy_enforcer.py loads rate-limits.json, blocks platforms, enforces caps
Content sanitization Active in all posting scripts — strips JSON, metadata, placeholders
Two-stage rendering Active in publish.sh — queue files rendered to clean text before posting
QA guardrails Active in publish.sh — length, hashtags, banned phrases checked
Direct post gating Active — scripts require --allow-direct-post, log all attempts
Brand knowledge base Complete — 10 products with visual identity, colors, image rules
Image prompt enrichment Active — generic briefs enriched with product context and brand colors
LinkedIn posting Playwright, policy-enforced, 3/day cap
X/Twitter posting Playwright with overlay dismissal, dropdown clearing, upload waits
Instagram posting Playwright with multi-selector caption targeting, Stories detection
Discord posting Webhook, fully operational
Daily pipeline All stages functional — intake through reporting
Approval engine L1-L4 routing with 6-dimension risk scoring
Self-healing watchdog Active every 15 min — 9 checks, auto-restart Ollama/Gateway, session refresh, zombie cleanup
32 unit tests Passing — sanitization (20) + image enrichment (12)
# 1. Ensure external HD is connected# 2. Start Ollama&
# 3. Check policy status# 4. Check publishing status# 5. Run full pipeline (dry-run first)# 6. Publish approved content (live)# 7. Run tests
Issue
Fix
Ollama not running
ollama serve &
External HD not found
Plug in, check /Volumes/Expansion/
Session expired
python3 post_<platform>.py --login
DALL-E budget exceeded
Wait for next day or --force
Platform blocked by policy
Edit policies/rate-limits.json — set blocked: false
Direct post refused
Add --allow-direct-post flag (still enforces policy)
ffmpeg not found
Already at ~/local/bin/ffmpeg
Zoho headless blocked
Runs visible mode only (by design)
Built by Reeturaj Goswami | InBharat.ai