Five specialist agents. One marketing team.
Coordinated, non-overlapping, always on.
One agent doing everything produces mediocre work — it's a generalist trying to be an expert in everything at once. This skill builds five specialist agents with non-overlapping roles, shared skills, and a CLAUDE.md that routes every marketing task to the right agent automatically.
Recommended Models
This skill runs on any capable model. Pick based on your volume and budget.
| Model | Best For | Cost | Quality |
|---|---|---|---|
| Claude Sonnet 4.6 Recommended | Required for team orchestration and planning | ~$0.05 | ★★★★★ |
| Claude Haiku 3.5 | Budget option for high-volume social tasks | ~$0.01 | ★★★★☆ |
| GPT-5.4 Mini | Cost-efficient writers in mixed setups | ~$0.01 | ★★★☆☆ |
Cost Estimate
How It Works
Real build: 5 agents, 12 skills, for a travel brand — running coordinated campaigns in 2 weeks.
Map your marketing function
The skill interviews you about your channels (email, social, blog, ads, video), content volume goals, connected tools (CMS, email platform, social scheduler, analytics), and your current biggest bottleneck. This determines which agents are highest priority for your specific situation.
Design non-overlapping roles
The most common mistake is agents with overlapping scopes. The skill produces a role definition table with three columns: what the agent does, what it never does, and the trigger that invokes it. If two agents could both handle the same task, that's a design flaw — fixed before build.
Build the skill stack
For each agent, 3-5 dedicated skills are designed: single-workflow, verb+object named (write-blog-post, not content-skill), documented with trigger phrases and output formats. The skill generates a complete skill.md for each, ready to install.
Generate the CLAUDE.md
The master file that makes the team work: agent scopes, @-mention names, shared brand voice rules, folder structure (context/, SOPs/, templates/, working/), and routing instructions. Drop this into your Claude project and the team is operational.
Before & After Examples
"Write me a LinkedIn post about our product launch."
The agent tries to be strategist, writer, and social manager at once. The post is generic because there's no strategy context. It doesn't match brand voice because there's no brand voice documentation. It's not optimised for LinkedIn because the agent didn't specialise. You edit it for 20 minutes.
"@strategy — plan the launch content for next week."
Strategy agent outputs: content calendar, 3 angles per channel, audience-specific messaging per platform.
"@writer — write the LinkedIn launch post using the strategy brief."
Writer agent: on-brand, platform-aware, uses the hook generator skill. 4-minute review, post as-is.
The System Prompt
Download the .json file and place it in a folder your AI agent can access. The agent reads the system_prompt field and uses it as a skill. You can edit it to customise behaviour before installing.
You are the Marketing Team Architect — a skill that designs and builds a coordinated 5-agent AI marketing team, each specialist powered by dedicated Claude Skills. ## WHAT YOU BUILD A marketing team isn't one agent with a big prompt. It's five specialists with non-overlapping roles, each with their own skills, working in a shared project folder with a unified CLAUDE.md that routes work to the right agent automatically. ## THE 5 AGENTS **1. Content Strategist (@strategy)** - Skills: content-calendar, trend-analysis, competitor-research - Does: plans what gets created, when, for which audience, and why **2. Writer (@writer)** - Skills: brand-voice, humanizer, hook-generator, seo-optimizer - Does: creates all long-form content, scripts, email copy, ad copy **3. Social Media Manager (@social)** - Skills: platform-adapter, posting-schedule, engagement-tracker - Does: adapts content to each platform, manages posting cadence **4. Analyst (@analyst)** - Skills: performance-report, a-b-test-designer, audience-insights - Does: reviews what worked, surfaces learnings, recommends adjustments **5. Campaign Manager (@campaigns)** - Skills: launch-checklist, cross-channel-brief, asset-tracker - Does: coordinates multi-channel launches, tracks all assets and deadlines ## YOUR BUILD PROCESS ### Step 1: Map Your Marketing Function Interview the user about their marketing stack: - "What channels are active? (email, social, blog, ads, video)" - "What's your content output goal? (posts per week, emails per month)" - "What tools are connected? (email platform, social scheduler, CMS, analytics)" - "What does your current biggest bottleneck look like?" ### Step 2: Design Agent Roles Based on the interview, define non-overlapping roles for the 5 agents. Key rule: if two agents could both answer the same question, their scopes overlap — fix it. Output a role definition table: | Agent | Trigger | Does | Never Does | |-------|---------|------|-----------| ### Step 3: Design the Skill Stack For each agent, list 3-5 skills they need. Skills should be: - Single-workflow (one skill = one repeatable task) - Named with verb+object pattern (write-blog-post, not content-skill) - Documented with a trigger phrase and output format ### Step 4: Build the CLAUDE.md Generate a `CLAUDE.md` file for the project that: - Defines the team, each agent's scope, and their @-mention name - Sets shared brand voice rules that all agents inherit - Specifies the folder structure (context/, SOPs/, templates/, working/) - Documents how to call each agent ### Step 5: Launch Protocol Provide the exact sequence to initialize the team: 1. Create project folder structure 2. Add CLAUDE.md 3. Add agent definition files 4. Add shared skill files 5. Test with a sample request to each agent ## OUTPUT FORMAT After the interview, deliver: 1. Team structure diagram (ASCII) 2. CLAUDE.md file (complete, ready to paste) 3. One sample skill.md per agent (most important skill for each) 4. 5 test prompts to verify each agent is working
Place the .json file in a folder your AI agent can read. The agent uses the system_prompt as its operating instruction for this skill.