Skip to content

jbrahy/meta-agent-teams

Agent Teams

An open framework for building AI agent teams that improve themselves through structured human feedback.

Instead of hand-crafting individual agents, you define a team — the roles, the rules, the relationships — and the framework gives you a complete system: specialist agents, a meta-agent that evolves them based on your feedback, an independent auditor that keeps evolution honest, and a git-backed history of every change and why it was made.

┌─────────────────────────────────────────────────┐
│                   Human (You)                    │
│         Execute, evaluate, provide feedback      │
└──────────┬──────────────────────┬────────────────┘
           │ feedback             │ review audits
           ▼                     ▼
┌─────────────────┐    ┌─────────────────┐
│   Meta-Agent    │◄───│  Auditor Agent  │
│  Evolves agents │    │  Checks for     │
│  based on your  │    │  drift, bias,   │
│  feedback       │    │  & regression   │
└────────┬────────┘    └─────────────────┘
         │
         ▼
┌─────────────────────────────────────────────────┐
│              Your Agent Team                     │
│  Specialists that advise — you decide & execute  │
└─────────────────────────────────────────────────┘

Why This Exists

Most AI agent setups are static. You write a prompt, it works okay, you tweak it when it doesn't, and you lose track of what you changed and why. There's no structured way to improve agents over time, no accountability for changes, and no safeguards against the agents drifting from their purpose.

This framework treats agent development as an iterative, auditable process. Agents start imperfect and get better through cycles of use, feedback, and constrained evolution. Every change is documented. Every modification has a rationale. An independent auditor prevents the system from optimizing itself into something you didn't intend.

The result is agents that earn trust through demonstrated improvement — not agents you have to trust because you can't see what they're doing.

What's in This Repo

meta-agent-teams/
├── teams/                      # 230 pre-built agent teams
│   ├── marketing/              # Full marketing team (8 specialist agents)
│   ├── devops/                 # DevOps/SRE team
│   ├── customer-success-team/  # Customer success team
│   ├── product-management-team/
│   └── ... (226 more teams)
│       ├── agents/             # Specialist agent system prompts
│       ├── meta-agent/         # Feedback processor and agent evolver
│       ├── auditor/            # Independent reviewer
│       ├── shared/             # Constitution and glossary
│       ├── feedback/           # Structured feedback templates
│       └── evals/              # Performance tracking
│
├── skill/                      # Team-builder skill for Claude Code (optional)
│   ├── SKILL.md                # Main skill definition
│   └── references/             # Architecture docs, domain constitutions
│
├── prompt/
│   └── agent-team-builder.md   # Portable prompt (works with any LLM)
│
└── docs/
    ├── architecture.md         # System design philosophy
    ├── getting-started.md      # First-run walkthrough
    └── domain-guide.md         # Domain-specific patterns

Three Ways to Use It

1. Use an existing team. Clone the repo, pick a team from teams/, and run agents with ./bin/run-agent.sh. Works with Claude, Ollama, OpenAI-compatible APIs, or paste system prompts directly into any LLM. All 230 teams are ready to go.

2. Build a new team with the skill. If you use Claude Code, install the team-builder skill from skill/. Then say "build me a DevOps team" and it generates the full repo structure, tailored to your domain.

3. Build a new team with the prompt. Copy the prompt from prompt/agent-team-builder.md and paste it into any LLM. Describe your team, and it produces the same output. Works with Claude, GPT, Gemini, Llama, or anything else that can handle a long system prompt.

Getting Started

Running an existing agent

# Clone the repo
git clone https://github.com/jbrahy/meta-agent-teams.git
cd meta-agent-teams

# Run an agent (uses your configured provider — defaults to Claude)
./bin/run-agent.sh marketing sdr

# One-shot prompt
./bin/run-agent.sh marketing sdr "Draft a cold outreach sequence for SaaS CTOs"

# Or paste the system prompt directly into any LLM conversation
# teams/marketing/agents/sdr/system-prompt.md

Configuring your provider

# Copy the example config
cp .agent-teams.env.example .agent-teams.env

# Edit to choose your provider
# AGENT_PROVIDER=ollama
# AGENT_MODEL=llama3.2

All agents work with Claude (default), Ollama (local), OpenAI-compatible APIs, or any LLM via Simon Willison's llm tool.

Providing feedback

After reviewing an agent's output:

# Copy the feedback template
cp teams/marketing/feedback/template.md teams/marketing/feedback/$(date +%Y-%m)/$(date +%Y-%m-%d).md

# Edit with your feedback, then run the full cycle:
./bin/run-cycle.sh marketing

# Or run steps individually:
# Meta-agent processes feedback → auditor reviews → you approve → git commit

Building a new team

# Option A: Using Claude Code with the skill installed
claude "Build me a customer success team for a B2B SaaS product"

# Option B: Using the portable prompt with any LLM
# Copy prompt/agent-team-builder.md into your LLM of choice
# Then describe the team you want

Core Principles

  1. Agents advise, humans execute. No agent takes autonomous action without explicit human approval. Output is always a suggestion.

  2. Everything in git. Every agent modification is a commit with a documented rationale. You can diff, bisect, revert, and branch agent evolution just like code.

  3. Feedback-driven improvement. Agents only change based on structured human feedback. The meta-agent never self-directs optimization.

  4. Constrained evolution. The meta-agent operates within a constitution it cannot modify. Changes are incremental (max 30% per cycle), require documented rationale, and must pass auditor review.

  5. No single point of control. The auditor independently reviews the meta-agent's decisions. The human reviews audit findings. No entity in the system has unchecked authority.

Architecture

The system has four layers:

Specialist agents do the domain work. They produce advisory output in their area of expertise (content writing, analytics, prospecting, etc). Each has a system prompt defining capabilities, output standards, guardrails, and evaluation criteria.

The meta-agent processes human feedback and proposes modifications to specialist agents. It follows a four-step pipeline: categorize feedback → diagnose root cause → propose changes → document rationale. It cannot modify itself, the auditor, or the constitution.

The auditor independently reviews every meta-agent proposal across six dimensions: constitutional compliance, feedback fidelity, drift detection, regression risk, cross-agent coherence, and change magnitude. It has no authority to make changes — only to approve, flag, or recommend rejection.

The constitution defines inviolable constraints. Every domain has ethical and regulatory boundaries that no agent should cross regardless of feedback. Only the human operator can amend the constitution.

For full architectural detail, see docs/architecture.md.

Pre-Built Teams

230 teams are ready to use. Browse by category:

Engineering & Technology

Team Description
architecture System design, architecture decisions, and technical planning
back-end-engineering-team Backend development, APIs, and server-side systems
front-end-engineering-team Frontend development and UI implementation
full-stack-engineering-team Full-stack feature development
devops Infrastructure, CI/CD, and deployment operations
sre-team Site reliability and production operations
data-engineering-team Data pipelines, warehousing, and infrastructure
data-science-team Data analysis, modeling, and experimentation
ml-engineering-team Machine learning systems and MLOps
mobile-engineering-team iOS/Android and cross-platform mobile development
platform-engineering-team Developer platforms and internal tooling
security-engineering-team Application security and vulnerability management
design-systems-team UI component libraries and design standards
internal-tools Internal tooling and workflow automation
implementation Feature implementation and technical delivery
qa Quality assurance and test strategy
testing-team Test automation and coverage
game-design Game design and development
open-source-maintenance Open source project maintenance

Marketing

Team Description
marketing Full marketing team — campaign orchestration, content, SDR, SEO, analytics, lead scoring, audience, creative
brand-strategy-team Brand positioning, voice, and identity strategy
competitive-intelligence-team Competitive landscape research and analysis
competitive-mention-team Monitoring and responding to competitive mentions
content-design-team Content design and visual communication
content-gap-team Identifying and prioritizing content gaps
content-idea-team Content ideation and editorial planning
content-marketing-team Content strategy and production
content-repurposing-team Repurposing content across formats and channels
crm-marketing-team CRM-driven marketing and lifecycle campaigns
demand-generation-team Demand generation programs and pipeline
growth-marketing-team Growth experiments and acquisition optimization
seo-strategy-team SEO strategy, content, and technical optimization
social-media-team Social media content and community management
social-post-advisor-team Social post review and improvement
pr-and-communications-team PR strategy and media communications
product-marketing-team Product positioning and go-to-market
thought-leadership-team Executive and brand thought leadership
narrative-team Brand and product storytelling

Sales

Team Description
account-executive Account executive workflow support
channel-sales-team Channel and partner sales programs
sales-development-team SDR outreach and pipeline development
sales-call-review-team Sales call analysis and coaching
sales-closing-team Closing strategies and deal negotiation
sales-script-evolution-team Iterating and improving sales scripts
objection-analysis-team Analyzing and responding to sales objections
discovery-question-team Developing discovery questions for sales conversations
lead-quality-team Lead scoring and qualification
offer-refinement-team Refining and improving offers
pipeline-insight-team Sales pipeline analysis and forecasting
proposal-improvement-team Proposal writing and improvement
win-loss-insight-team Win/loss analysis and sales learning
partnership-fit-team Partnership evaluation and development
solutions-engineering-team Pre-sales technical support

Customer Success & Support

Team Description
customer-success-team Customer success management and retention
customer-support-team Customer support operations
customer-onboarding-team Customer onboarding workflows
customer-journey-friction-team Identifying and reducing customer friction
customer-persona-refinement-team Refining customer personas and segments
customer-signal-team Monitoring customer health signals
concierge-team High-touch customer concierge service
retention-and-renewals-team Retention programs and renewal management
retention-clue-team Early warning signals for churn risk
follow-up-coach-team Follow-up strategy and execution coaching
onboarding-insight-team Onboarding analytics and improvement
support-trend-team Support ticket trend analysis

Product

Team Description
product-management-team Product strategy, roadmap, and prioritization
product-discovery-team User research and product discovery
product-feedback-team Synthesizing and acting on product feedback
product-confusion-team Identifying and resolving product confusion
product-positioning-team Product messaging and positioning
feature-prioritization-team Feature ranking and roadmap decisions
beta-feedback-team Beta program feedback collection and synthesis
roadmap-sanity-check-team Stress-testing product roadmaps
experiment-review-team Reviewing and learning from experiments
ux-research-team UX research and usability testing
user-research-synthesis-team Synthesizing user research findings

Operations & Strategy

Team Description
business-operations-team Business operations and process management
strategy-operations-team Strategy execution and operational alignment
automation-team Workflow automation and tooling
operations-clarity-team Clarifying operational processes
program-management Program and portfolio management
pmo-team Project management office operations
dispatch-team Work dispatch and task routing
resource-allocation-team Resource planning and allocation
workflow-simplification-team Process simplification and waste reduction
process-drift-team Detecting and correcting process drift
sop-improvement-team Standard operating procedure improvement
exception-analysis-team Analyzing exceptions and edge cases
bottleneck-finder-team Identifying operational bottlenecks
weekly-operations-review-team Weekly ops review and action tracking

Finance & Legal

Team Description
accounting-operations-team Accounting workflows and financial operations
fp-a-team Financial planning, analysis, and forecasting
revenue-cycle-management-team Revenue cycle operations
revenue-operations-team Revenue operations and GTM alignment
compliance-and-risk-team Compliance programs and risk management
contracts-team Contract review and management
legal-operations Legal operations and workflow management
m-a-team Mergers and acquisitions support
corporate-development Corporate strategy and business development
investor-relations-team Investor relations and communications
esg-reporting-team ESG reporting and sustainability initiatives

People & Learning

Team Description
people-operations-team HR and people operations
recruiting-team Talent acquisition and hiring
learning-and-development-team L&D programs and training
curriculum-design-team Curriculum and learning experience design
instructional-design-team Instructional design and course development
learning-loop-team Continuous learning and improvement cycles
habit-reinforcement-team Building and reinforcing productive habits

Analytics & Research

Team Description
analytics Analytics strategy and data analysis
bi-team Business intelligence and reporting
insight-mining-team Extracting insights from data and feedback
pattern-recognition-team Identifying patterns and trends
research-digest-team Summarizing and applying research
scientific-literature-review-team Academic and scientific literature review
adoption-insight-team Product and behavior adoption analysis

Healthcare & Life Sciences

Team Description
care-coordination-team Patient care coordination
clinical-documentation-improvement-team Clinical documentation quality
clinical-trial-operations-team Clinical trial operations and management
medical-coding-team Medical coding and billing
medical-information-team Medical information services
pharmacy Pharmacy operations and medication management
biotech-research-operations-team Biotech research operations
regulatory-affairs-team Regulatory submissions and compliance
lab-management-team Laboratory operations and management

Education & Nonprofits

Team Description
academic-advising-team Student academic advising
admissions Admissions review and communications
enrollment-team Enrollment management
student-success-team Student success and retention
grant-writing-team Grant research and proposal writing
nonprofit-program-operations-team Nonprofit program management
policy-research-team Policy research and analysis
public-affairs Public affairs and government relations

Content, Media & Communications

Team Description
editorial-team Editorial planning and content quality
newsroom Newsroom operations and editorial workflow
podcast-production-team Podcast production and distribution
video-production-team Video production and content
publishing-operations-team Publishing workflow and operations
newsletter-insight-team Newsletter strategy and performance
writing-coach-team Writing improvement and coaching
articles-into-practical-advice Transforming articles into actionable takeaways
localization Translation and localization operations
translation-team Translation services and quality
ux-writing UX copy and interface writing
voice-consistency-team Brand voice and tone consistency
executive-communications-team Executive communications and messaging
communication-clarity-team Improving communication clarity

Leadership & Productivity

Team Description
chief-of-staff Chief of staff support and strategic coordination
chief-of-staff-lite-team Lightweight chief of staff functions
executive-assistant-advisory-team Executive assistant operations and advisory
calendar-optimization-team Calendar management and time optimization
daily-priorities-team Daily priority setting and focus
focus-coach-team Focus, distraction management, and deep work
inbox-advisory-team Email triage and inbox management
time-audit-team Time use analysis and optimization
weekly-planning-team Weekly review and planning
delegation-advisor-team Delegation strategy and skill building
overcommitment-detector-team Identifying overcommitment and scope creep
meeting-prep-team Meeting preparation and briefing
meeting-debrief-team Meeting debrief and action tracking
personal-kpi-team Personal performance metrics and tracking
personal-power-team Personal effectiveness and energy management
personal-board-of-advisors-team Building and leveraging an advisory network
founder-reflection-team Founder reflection and decision review

Strategy & Analysis

Team Description
strategic-question-team Strategic question generation and framing
scenario-planning-team Scenario planning and strategic foresight
decision-journal-team Decision tracking and retrospective learning
blind-spot-finder-team Identifying strategic blind spots
assumption-tracker-team Tracking and validating key assumptions
red-team-team Adversarial thinking and challenge testing
second-opinion-team Structured second opinions on key decisions
tradeoff-advisor-team Trade-off analysis and decision support
perspective-shift-team Reframing problems from new perspectives
constraint-clarifier-team Clarifying constraints and boundaries
opportunity-spotting-team Identifying strategic opportunities
risk-check-team Risk identification and assessment
leverage-finder-team Finding high-leverage actions and investments
long-term-compounding-team Long-term compounding decisions and habits

Community & Advocacy

Team Description
community-team Community building and engagement
community-management-team Community operations and moderation
advocacy-team Customer and user advocacy programs
audience-questions-team Surfacing and answering audience questions
faq-discovery-team FAQ research and content development
testimonial-mining-team Collecting and leveraging testimonials

Specialized Industries

Team Description
construction-estimating-team Construction cost estimating
e-commerce-merchandising-team E-commerce merchandising and catalog management
energy-management-team Energy operations and optimization
event-planning-and-production-team Event planning and production management
hospitality-guest-experience-team Hospitality and guest experience
logistics Logistics and supply chain operations
manufacturing-process-improvement-team Manufacturing process optimization
manufacturing-quality-team Manufacturing quality assurance
marketplace-operations-team Marketplace operations and seller management
property-management-team Property management operations
real-estate-acquisitions-team Real estate deal sourcing and analysis
restaurant-operations-team Restaurant operations management
supply-chain-planning-team Supply chain planning and optimization
procurement Procurement strategy and vendor management
sourcing-team Supplier sourcing and evaluation
sustainability Sustainability programs and reporting
travel-planning Corporate and executive travel planning
partner Partner program operations

Documentation & Knowledge

Team Description
documentation-team Technical documentation writing and maintenance
documentation-gap-team Identifying documentation gaps
knowledge-management Knowledge base operations and findability
handoff-quality-team Handoff quality and knowledge transfer
reading-notes-synthesis-team Synthesizing reading notes into insights

Feedback & Improvement

Team Description
feedback-digestion-team Synthesizing and prioritizing feedback
feedback-digestion-team Feedback collection and analysis
quality-review-team Output quality review
simplicity-team Reducing complexity across systems
sop-improvement-team SOP review and improvement
repeated-mistake-team Identifying and eliminating recurring mistakes
momentum-tracking-team Tracking progress and maintaining momentum
goal-alignment-team Aligning team goals and priorities
conflict-early-warning-team Early detection of organizational conflicts
stakeholder-radar-team Stakeholder mapping and relationship management
idea-capture-team Capturing and organizing ideas
slack-pattern-team Analyzing patterns in Slack communications
story-mining-team Mining stories from data and customer interactions
trend-interpretation-team Interpreting emerging trends
ai-evaluation Evaluating AI outputs and model performance
trust-and-safety-team Trust, safety, and content moderation
pricing-perception-team Price sensitivity and perception research
relationship-health-team Monitoring and improving key relationships
lifecycle Customer and product lifecycle management
professional-services-team Professional services delivery
project-design-team Project scoping and design
creative-design-team Creative design and visual production
crisis-management-team Crisis planning and response
it-help-desk-team IT help desk and user support

Contributing

We'd love your help making this framework better. There are several ways to contribute:

Share a team you've built

If you've used the framework to build a team for a new domain, consider contributing it. Community-contributed teams help everyone and demonstrate the framework's versatility.

  1. Fork the repo
  2. Add your team to teams/<domain-name>/
  3. Ensure it includes all required files (see the structure above)
  4. Submit a PR using the team contribution template

Improve the framework

The meta-agent architecture, auditor evaluation dimensions, constitution patterns, and feedback loop can all be improved. If you've found a better pattern through real-world use:

  1. Open an issue describing what you learned and why it's better
  2. Fork, implement, and submit a PR
  3. Include before/after examples if possible

Add domain-specific constitution patterns

The skill/references/domain-constitutions.md file contains ethical and regulatory constraint patterns for various domains. If you have domain expertise and can improve or add to these patterns, that's a high-value contribution — these constraints directly impact agent safety.

Report issues

If you find that the framework produces poor results for a specific domain, or if an agent's guardrails are insufficient, open an issue. Safety-related issues are especially welcome.

Contribution guidelines

  • Keep it open. All contributions must be compatible with the AGPL-3.0 license.
  • Document your reasoning. The framework is built on the principle that every change has a documented rationale. Contributions should follow the same principle. Explain what you changed and why in your PR.
  • Test with real use. If contributing a team, it should have been run through at least a few feedback cycles. Untested teams are accepted as drafts if clearly labeled.
  • Respect the architecture. The meta-agent/auditor/constitution pattern is the core of the framework. Contributions should work within this pattern, not around it. If you think the pattern itself should change, open an issue for discussion first.
  • Be thoughtful about ethics. Constitution patterns and agent guardrails exist for a reason. PRs that weaken safety constraints need strong justification.

See CONTRIBUTING.md for full details.

License

This project is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).

This means:

  • You can use, modify, and distribute this freely
  • If you modify it, you must share your modifications under the same license
  • If you run a modified version as a service (even over a network), you must make the source available to users of that service
  • The framework will always remain open source

See LICENSE for the full text.

Why AGPL-3.0

This project exists to make agent development more transparent, auditable, and ethical. The AGPL ensures that these values propagate through all derivatives. Anyone can use and improve this work, but no one can make it proprietary. If someone builds something better on top of this, that improvement stays available to everyone.

Acknowledgments

Built by humans who believe AI development should be transparent, auditable, and in service of people — not the other way around.

Roadmap

  • Pre-built teams: Marketing, DevOps/SRE, Customer Success, Sales, Product Management
  • 230 pre-built teams across engineering, marketing, sales, operations, healthcare, education, and more
  • Multi-model support: Claude, Ollama, OpenAI-compatible APIs, and llm tool
  • Automated eval runner for feedback cycle benchmarking
  • Inter-team communication protocols (agents from different teams collaborating)
  • Web UI for feedback submission and audit review
  • Agent promotion system (advisory → semi-autonomous → autonomous based on trust score)

About

An open framework for building self-improving AI agent teams. Includes a meta-agent architecture with independent auditing, git-backed evolution, and structured feedback loops. Ships with a complete marketing team, a reusable team-builder skill, and a portable prompt that works with any LLM.

Topics

Resources

License

Code of conduct

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages