brandforge

AI-powered brand content operating system for any brand — blogs, LinkedIn, Twitter/X, Instagram, Facebook, email sequences, PPTX decks, XLSX trackers, and a full multi-channel content pipeline.

What it does

• Brand auto-discovery — 6 questions → web crawl → voice URL analysis → auto-drafted brand config, ready for your review. No manual forms.
• 8-stage content pipeline — Brief → Framework → Draft → Fact-check → AI SEO pass → Copy audit → Channel adaptation → Publish package
• Multi-channel — Blog, LinkedIn, Twitter/X thread, Instagram carousel, Facebook, email newsletter, email sequence
• Copywriting frameworks — AIDA, PAS, BAB, LEMA, 4-Point, AIDA+R — auto-selected per content type
• AI SEO pass — Structures content for citation by ChatGPT, Perplexity, Google AI Overviews: definition blocks, FAQ items, stat+source rows, summary boxes
• AI humanizer — Detect → Humanize → Voice inject — removes AI tells and injects brand voice
• Deck system — PPTX via pptxgenjs, XLSX via openpyxl+pandas, generative art via p5.js
• Image generation — Google Gemini via nanobanana MCP (optional)
• Content registry — CSV log of all outputs with duplicate detection
• Multi-brand — Clone, sync, and manage multiple brand workspaces

Install

# Global — available across all your projects
npx skills add ast-consulting/brandforge -g

# Project — committed to your repo, shared with your team
npx skills add ast-consulting/brandforge

Manual install (git clone):

# Global
git clone https://github.com/ast-consulting/brandforge \
  ~/.claude/skills/brandforge

# Project-local
git clone https://github.com/ast-consulting/brandforge \
  .claude/skills/brandforge

Compatible with

Claude Code · Cursor · GitHub Copilot · Codex CLI · Gemini CLI · Windsurf · Roo Code

Usage

Just describe what you want in natural language:

"Write a blog post for [brand] about [topic]"
"Create a LinkedIn post for [brand] about [topic]"
"Run full content pipeline for [brand] on [topic]"
"Set up a new brand workspace for [name]"
"Make this [brand] copy sound human: [paste]"
"Create a 3-email nurture sequence for [brand] targeting [goal]"
"Build an investor deck for [brand]"
"SEO audit [brand] content on [topic]"

File structure

brandforge/
├── SKILL.md                        # Main skill — brand config, pipeline, routing, QC
├── README.md                       # This file
├── LICENSE                         # MIT
├── evals/
│   └── evals.json                  # 3 test cases: LinkedIn post, new brand, blog plan
├── scripts/
│   ├── setup_outputs.sh            # Scaffold output directory structure for a brand
│   ├── setup_nanobanana.sh         # Full nanobanana MCP install: uv, .env, .mcp.json, permissions
│   └── overlay-logo.sh             # ffmpeg logo compositor — adds brand logo to any generated image
└── references/
    ├── brand-config.md             # Full brand interview + web crawl + auto-draft protocol
    ├── content-pipeline.md         # Deep 8-stage pipeline with per-stage instructions
    ├── social-playbooks.md         # LinkedIn, Twitter/X, Instagram, Facebook formulas
    ├── email-playbooks.md          # Newsletter, campaign, drip sequence templates
    ├── deck-system.md              # pptxgenjs API, XLSX patterns, p5.js generative art
    ├── agent-architecture.md       # 3-tier agent orchestration setup
    └── image-generation.md        # nanobanana MCP install, Gemini API key, troubleshooting

Optional: Image generation (nanobanana MCP)

Hero images, social banners, and Instagram carousel slides are generated via Google Gemini through the nanobanana MCP server.

One-command setup:

bash scripts/setup_nanobanana.sh

This handles everything: installs uv, saves your GEMINI_API_KEY to .env, creates .mcp.json, and updates .claude/settings.json with the right permissions.

Or pass the key non-interactively:

bash scripts/setup_nanobanana.sh --key AIzaSy...yourkey

After setup — start Claude Code:

set -a && source .env && set +a   # export key in current terminal
claude                             # nanobanana auto-starts

See references/image-generation.md for the full guide and troubleshooting.

Logo overlay (no extra API cost)

After generating images, stamp your brand logo with ffmpeg — no second Gemini call:

# Auto-discovers logo in .claude/skills/{brand}-brand/assets/
bash scripts/overlay-logo.sh outputs/social/linkedin/post.png outputs/social/linkedin/post-branded.png

# Specify logo manually
bash scripts/overlay-logo.sh input.png output.png --logo .claude/skills/mybrand-brand/assets/logo.png

# Options: position (bl/br/tl/tr), width, margin
bash scripts/overlay-logo.sh input.png output.png --logo logo.avif --position br --width 120 --margin 20

Requires: ffmpeg — brew install ffmpeg (macOS) · apt install ffmpeg (Linux)

The skill works fully without nanobanana — it writes [IMAGE NEEDED: description] placeholders and continues.

Multi-brand support

Each brand gets its own workspace cloned from a master template:

cp -r [existing-brand]/ [new-brand]/
# Update all brand-specific values in: [new-brand]/.claude/skills/[brand]-brand/SKILL.md
# Replace: brand name, tagline, colors, voice, proof points, services, URLs

Brand configs live at .claude/skills/{brand}-brand/SKILL.md and are auto-loaded at the start of every content task.

License

MIT — see LICENSE