Run a lean marketing agency with a team of AI agents handling maintenance, QA, content, SEO, design, research, reporting, and sales support — so the humans can focus on relationships and growth.
This is an open source reference implementation showing how a small agency (2-3 people) can use Claude Code agents to automate back-office operations and scale from ~10 clients to 30+ without hiring.
What's included: Nine specialized agents with detailed instructions, a daily orchestration workflow, automated client report generation (PowerPoint + email), client onboarding scripts, health monitoring, and a complete sales pipeline — ready to fork and customize for your agency.
Small web/marketing agencies (1-5 people) that:
- Spend more time servicing existing clients than finding new ones
- Can't afford to hire specialists (SEO, QA, content, reporting)
- Want to scale revenue without scaling headcount
- Use WordPress or similar CMS platforms for client sites
┌───────────────────────────────────────────────┐
│ FOUNDER A (Supervisor) │
│ Reviews agent output · Approves production │
│ Handles complex dev · Client escalations │
├───────────────────────────────────────────────┤
│ DAILY ORCHESTRATOR │
│ workflows/daily.sh (cron) │
├───────┬─────┬─────┬───────┬──────┬───────┬──────┬────────┤
│Maint. │ QA │ SEO │Content│Design│Resrch │Report│ Sales │
│ Agent │Agent│Agent│ Agent │Agent │ Agent │Agent │Support │
├───────┴─────┴─────┴───────┴──────┴───────┴──────┴────────┤
│ CLIENT CONFIGS │
│ clients/*/config.yaml — one per client │
├───────────────────────────────────────────────┤
│ FOUNDER B (Prospecting) │
│ Networking · Sales meetings · Relationships │
│ Feeds workflows/prospects-queue.yaml │
└───────────────────────────────────────────────┘
npm install -g @anthropic-ai/claude-codegit clone https://github.com/vinay-lgtm-code/brik-for-agencies.git
cd brik-for-agencies
chmod +x scripts/*.sh workflows/*.sh./scripts/onboard-client.sh my-first-client
# Edit clients/my-first-client/config.yaml with their details./scripts/health-check.sh# Run a specific agent for a specific client
./scripts/run-agent.sh maintenance "Run daily health checks" --client my-first-client
# Run the full daily orchestration
./workflows/daily.sh# Add to crontab: crontab -e
0 7 * * 1-5 cd /path/to/agency-agents && ./workflows/daily.sh >> logs/cron.log 2>&1| Agent | Purpose | Key Output |
|---|---|---|
| Maintenance | Updates, security, backups, uptime | Health logs, update reports |
| QA | Testing, accessibility, performance | Audit reports, bug reports |
| SEO | Rankings, local SEO, technical SEO | SEO reports, keyword research |
| Content | Blog posts, copy, meta tags | Content drafts, calendars |
| Design | Wireframes, mockups, design systems | Page designs, brand guidelines |
| Research | Prospect profiling, competitor analysis | Research briefs, landscape reports |
| Report Delivery | Branded PPTX reports, email drafts | PowerPoint decks, cover emails |
| Reports | Internal summaries, churn detection | Weekly summaries, monthly reports |
| Sales | Proposal drafting, meeting prep | Lead briefs, proposal drafts |
Each agent has its own detailed instruction file in agents/*/CLAUDE.md with procedures, output formats, and safety boundaries. The report-delivery agent also uses skills/client-report/SKILL.md for slide design and layout guidelines.
| Time | Agent | Task | Frequency |
|---|---|---|---|
| 07:00 | Maintenance | Health checks (all clients) | Daily |
| 07:30 | QA | Post-update regression checks | Daily (if updates) |
| 08:00 | Sales Support | Process prospect queue | Daily |
| Monday AM | Maintenance | Weekly update audit | Weekly |
| Tuesday AM | SEO + Research | Keyword & competitor checks | Weekly |
| Wednesday AM | QA | Weekly site audits | Weekly |
| Thursday AM | Content | Draft weekly content | Weekly |
| Friday PM | Reports | Weekly internal summary | Weekly |
| 1st of month | All | Monthly deep audits & reports | Monthly |
| 5th of month | Report Delivery | Branded PPTX client reports | Monthly |
agency-agents/
├── README.md ← You are here
├── CLAUDE.md ← Master ground rules for all agents
├── LICENSE
├── CONTRIBUTING.md
│
├── agents/ ← Agent-specific instructions
│ ├── maintenance/CLAUDE.md
│ ├── qa/CLAUDE.md
│ ├── content/CLAUDE.md
│ ├── seo/CLAUDE.md
│ ├── design/CLAUDE.md
│ ├── research/CLAUDE.md
│ ├── report-delivery/CLAUDE.md ← Branded PPTX + email
│ ├── reports/CLAUDE.md
│ └── sales/CLAUDE.md
│
├── skills/ ← Skill definitions for agent tasks
│ └── client-report/SKILL.md ← Report design guide
│
├── security/ ← Security policy and data classification
│ └── SECURITY-POLICY.md
│
├── scripts/ ← Utility scripts
│ ├── run-agent.sh ← Run individual agents
│ ├── health-check.sh ← Automated health checker
│ ├── onboard-client.sh ← New client setup
│ ├── secrets-check.sh ← Scan for leaked credentials
│ └── audit-logger.sh ← Tamper-evident audit trail
│
├── workflows/ ← Multi-agent workflows
│ ├── daily.sh ← Daily orchestration (cron job)
│ ├── client-report.sh ← Monthly PPTX report generation
│ └── prospects-queue.yaml ← Sales pipeline
│
├── docs/ ← Strategy and reference
│ ├── STRATEGY.md ← Scaling strategy and revenue model
│ └── PACKAGES.md ← Service tier pricing (example)
│
├── examples/ ← Example configurations
│ └── sheffield-dental/
│ └── config.yaml ← Example client config
│
├── clients/ ← Your clients (created by onboard script)
│ └── {client-slug}/
│ ├── config.yaml
│ ├── health-log.md
│ ├── reports/
│ └── content/
│
├── logs/ ← Daily action logs (auto-created)
│ └── audit/ ← Hash-chained audit trail
└── templates/ ← Report and email templates
This repo is built around a Sheffield (UK) web agency as a worked example. To adapt it for your agency:
- Update
CLAUDE.md— change the ground rules to match your workflows - Edit
agents/*/CLAUDE.md— adjust agent instructions for your services, tools, and local market - Edit
docs/PACKAGES.md— define your own service tiers and pricing - Edit
docs/STRATEGY.md— set your own revenue targets and growth plan - Edit
workflows/prospects-queue.yaml— customize for your sales process - Update local references — replace Sheffield-specific content with your own market
Every agent has hard safety limits (no production changes, no client contact, no credential storage, no deleting, staging-first, log everything). See security/SECURITY-POLICY.md for the full policy including data classification, agent access controls, and incident response.
Scan for leaked credentials before committing:
./scripts/secrets-check.sh # Full project scan
./scripts/secrets-check.sh --staged # Git pre-commit (staged files only)Detects API keys, passwords, tokens, private keys, cloud credentials, PII (emails, phone numbers), and sensitive file types. Set it as a git pre-commit hook:
ln -s ../../scripts/secrets-check.sh .git/hooks/pre-commitTamper-evident logging with SHA-256 hash chains — any modification to previous entries is detectable:
./scripts/audit-logger.sh log maintenance "updated-plugins" "3 plugins on staging"
./scripts/audit-logger.sh security qa "access-violation" "Attempted to read .env"
./scripts/audit-logger.sh verify 2025-02-08 # Check integrity
./scripts/audit-logger.sh summary # Daily summaryAll data is classified into four levels (CRITICAL → SENSITIVE → INTERNAL → PUBLIC) with handling rules for each. Client credentials must never be stored in this repo — use credential references pointing to your password manager instead.
| Item | Monthly |
|---|---|
| Claude Code / API usage | $200–$500 |
| Hosting tools (uptime monitoring, etc.) | $25–$60 |
| Total agent team | $225–$560 |
The agent team pays for itself with a single new client.
The examples/ and docs/ folders contain a complete worked example for a 2-person web agency in Sheffield, UK, including:
- A real client configuration (dental practice)
- Three-tier pricing model (Basic / Standard / Premium)
- Six-month scaling strategy from 10 to 28+ clients
- Revenue projections and churn reduction playbook
- Local SEO strategy and prospecting intelligence
See docs/STRATEGY.md for the full business case.
See CONTRIBUTING.md for guidelines on adding new agents, improving existing ones, or sharing your agency's adaptations.
MIT — see LICENSE for details.
Built with Claude Code. Designed for small agencies that want to punch above their weight.