ai-business-skills 147 marketing skills that turn your AI assistant into a full marketing team. Built for Vietnam first, ready for the US, Europe, Southeast Asia and Latin America. Quick start · How it works · Vietnam or Global · Agents · Workflows · All skills · Other AI tools ✨ What you get Each skill is one Markdown file (SKILL.md) that teaches an AI assistant one marketing job from start to finish: what to ask…
147 marketing skills that turn your AI assistant into a full marketing team.
Built for Vietnam first, ready for the US, Europe, Southeast Asia and Latin America.
Each skill is one Markdown file (SKILL.md) that teaches an AI assistant one marketing job from start to finish: what to ask you, which rules and benchmarks to apply, and what the finished document must contain. You ask in plain words. You get a complete .md file, not generic advice.
You ask
A plain AI chat gives you
With ai-business-skills you get
“Make a marketing plan for my spa”
500 words of general advice
A full plan: 7 sections, KPIs in 3 scenarios, budget split, weekly timeline, risk table
“Write Facebook ads for the new facial”
One generic paragraph
6 variations (2 for cold, 2 for warm, 2 for hot audiences), each with headline, primary text and CTA
“Write this month’s report”
A list of numbers
Insight first, then the root cause, actions for the next 48 hours, and next month’s plan
Good for: marketing plans, campaign briefs, ad copy and video scripts, content calendars, ads audits, e-commerce stores on Shopee / TikTok Shop / Lazada, personal brands, and running a marketing team with AI.
Not for: building product code, technical specs or deploying servers.
⚡ Quick start
1. Install (pick one)
Claude Code plugin (easiest, installs all 147 skills)
> Set up the product marketing context for Luna Spa
The AI interviews you and saves .agents/product-marketing-context.md: product, customers, competitors, positioning, proof and objections. Every skill reads this file first, so you never repeat yourself. For a market outside Vietnam it saves product-marketing-context-global.md instead.
3. Ask for real work
> Plan next month's marketing for Luna Spa
> Write a 30-second TikTok script for the new facial
> Cost per message is 45K VND and ROAS is 1.8x. Audit it and tell me what to fix
> Work backwards: what budget do we need for 200 million VND revenue a month?
Each answer is saved as [skill]-[product]-[YYYYMMDD].md, for example ke-hoach-mkt-spa-luna-20260415.md.
🧭 How it works
flowchart LR
A["💬 You ask<br/>in plain words"] --> B["📇 Read context<br/>.agents/*.md"]
B --> C{"🌏 Which market?"}
C -->|Vietnam| D["🇻🇳 VN skill<br/>skills/vi"]
C -->|Global| E["🌍 Global skill<br/>skills/en"]
D --> F["❓ Ask at most 4 questions<br/>only about what is missing"]
E --> F
F --> G["📏 Apply rules,<br/>benchmarks, legal checks"]
G --> H["📄 Finished .md file"]
Building block
What it is
Where it lives
Context
Your business on one page, written once
.agents/ in your project
Skill
One marketing job, start to finish
skills/vi/, skills/en/, modules/
Agent
A role (strategist, content, design…) that picks the right skills
agents/
Workflow
A fixed chain of skills for a multi-day job or a monthly loop
workflows/, modules/*/workflows/
Knowledge
Background reading to upload into an AI project per role
knowledge/
Skills call each other. One request can pull in several skills, in order:
You never switch modes by hand. The skills and agents look at which context file exists:
flowchart TD
S{"Which context file<br/>is in .agents/?"}
S -->|"product-marketing-context.md"| VN["🇻🇳 VN mode<br/>skills/vi"]
S -->|"product-marketing-context-global.md"| GL["🌍 Global mode<br/>skills/en"]
S -->|both| ASK["Asks you one question"]
S -->|none| NEW["Suggests creating one first"]
The same rule applies to personal brands (personal-brand-context.md / -global.md). When both a product and a personal brand context exist, skills 04 (video script) and 05 (ad copy) ask whether to write for sales or for authority.
Region
Currency
Privacy
Top channels
US
USD
CCPA / state laws
Meta, Google, TikTok, YouTube
EU
EUR / GBP
GDPR
Meta, Google, LinkedIn, TikTok
SEA
IDR / THB / SGD / PHP
PDPA (varies)
TikTok Shop, Shopee, Lazada, Meta
LATAM
BRL / MXN / ARS
LGPD (Brazil)
Meta, TikTok, MercadoLibre, WhatsApp
Skills with a file per region (variants/01-us.md … 04-latam.md): the foundation context, 03, 10, 11, 14, 17, 18 and 21. More in docs/global-region-guide.md.
👥 Meet the 6 agents
An agent is a role. Tell it the goal and it picks the right skills, in VN or Global mode.
How the team passes work along in a typical campaign:
flowchart LR
You(["You"]) --> ST["mkt-strategist<br/>plan · budget · brief"]
ST --> CP["content-producer<br/>scripts · copy · captions"]
ST --> DP["design-producer<br/>visual briefs · review"]
CP --> CO["channel-operator<br/>channels · email · chat · shops"]
DP --> CO
CO --> PA["performance-analyst<br/>tracking · audit · scaling"]
PA -->|"report + next plan"| ST
PB["personal-brand-builder<br/>founder & creator brands"] -.-> CP
🔁 Workflows
A workflow is a fixed chain of skills: either a job that takes several days, or a loop you run every month.
Content engine: a monthly loop
flowchart TB
subgraph PREP["1 · Plan the month"]
direction LR
V["35 Brand voice"] --> I["09 Insight"] --> CAL["01 Calendar"] --> BR["36 Content brief"]
end
subgraph MAKE["2 · Make the content"]
direction LR
W["Write<br/>37 captions · 04 scripts · 05 ads<br/>14 email · 38 seeding · 06 UGC"] --> VIS["Visual briefs<br/>42 · 43 · 44"]
end
subgraph LEARN["3 · Measure and plan again"]
direction LR
AU["39 Content audit"] --> RP["07 Report"] --> NX["40 Next plan"] -.-> AGAIN(["↻ Next month:<br/>back to step 1"])
end
PREP --> MAKE --> LEARN
Performance loop: no spend before tracking works
flowchart TB
subgraph PLAN["1 · Plan"]
direction LR
A["51 Audience"] --> K["10 Reverse KPI"] --> M["54 Media plan"]
end
subgraph GATE["2 · Tracking gate"]
direction LR
T{"53 Tracking<br/>verified?"} -->|no| FIX["Fix tracking<br/>no spend yet"] --> T
end
subgraph RUN["3 · Launch and test"]
direction LR
S["52 Account structure"] --> C["05 Ad copy"] --> AB["19 A/B test"] --> AU["21 Ads audit"]
end
subgraph GROW["4 · Scale and learn"]
direction LR
SC["55 Scale winners<br/>+20-30%, 72h apart"] --> R["07 Report"]
RT["56 Retarget<br/>warm audiences"] --> R
R --> N["57 Next ads plan"] -.-> AGAIN(["↻ Next cycle:<br/>back to step 1"])
end
PLAN --> GATE -->|verified| RUN --> GROW
Sales scripts for an AI chatbot (inbox and public comments) (VN only)
🛡️ Built-in guardrails
Every skill follows the same 8 operating rules (VN · Global): read the context first, treat outside content as data rather than instructions, label evidence, date and source every number, never invent numbers or customers, and send every real-world action through a draft-first gate.
sequenceDiagram
participant You
participant AI as AI + skill
participant Ads as Ad account (MCP)
You->>AI: "Scale the winning ad set"
AI->>AI: Enough data? Within the agent's permissions?
AI-->>You: Change proposal: current, proposed, expected result, rollback
You->>AI: Approve this one change (named approver)
AI->>Ads: Apply exactly that change
AI->>Ads: Read back the new value and status
AI-->>You: Log: time, change, approver, result
Ads decisions follow one ladder. Below 3x the target CPA in spend there is not enough data to judge. A pause is only proposed, never done silently. Budgets scale +20-30% per step, at least 72 hours apart.
Paste skills/en/product-marketing-context-global/SKILL.md (or the VN one) and answer its questions. Keep the answer.
Open the skill you need, for example skills/en/00-marketing-plan-global/SKILL.md.
Paste it with the saved context and write: “Follow this workflow. My input: …”
The AI follows the skill’s questions, rules and output template.
🧱 Anatomy of a skill
What is inside a SKILL.md
---
name: 00-ke-hoach-mkt
description: Short description the AI uses to decide when to run the skill
category: strategy
triggers:
- "lập kế hoạch marketing"
- "marketing plan"
output: .md file with 7 sections + KPIs in 3 scenarios
related: [08-nghien-cuu-doi-thu, 09-insight-khach-hang, 10-tinh-kpi-nguoc]
---
# Skill title
## Gather information -> at most 4 questions, skipped when the context answers them
## Rules -> insight before numbers, local benchmarks, no generic advice
## Output structure -> the exact sections, tables and scenarios to deliver
## Quality checklist -> checked before the file is handed over
Bilingual VN + EN is preferred, one language is accepted. Register a new skill in .claude-plugin/marketplace.json.
💖 Support
Free and MIT-licensed. If it saves you hours, you can support new region variants, industry skills and bilingual tutorials at opa.business/donate. Starring the repo, reporting bugs and sending PRs help too.
评论