Swarma enables AI agent teams to run growth experiments and build self-improving playbooks using the GROWS loop.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install swarma
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install swarma using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Swarma: Growth Experiment Loops for Agent Teams?
Swarma is a specialized framework designed for growth-oriented developers and marketers who want to move beyond one-shot agent execution. By implementing the GROWS loop—Generate, Run, Observe, Weigh, Stack—it transforms standard AI agents into a learning squad that discovers what actually works for your specific audience. As part of the Openclaw Skills ecosystem, Swarma allows teams to test hooks, landing pages, and outreach at scale, replacing guesswork with data-backed strategies.
The core value of Swarma lies in its ability to evolve. Unlike static automation tools, it tracks experiment signals and automatically updates a strategy.md file with validated patterns. This ensures that your agents aren't just repeating tasks, but are actively optimizing for specific performance metrics over time.
Swarma: Growth Experiment Loops for Agent Teams Use Cases
- Running high-velocity growth experiments for hooks, landing pages, and pricing.
- Building agent teams that learn and improve through iterative A/B testing.
- Scaling experiment volume from a few per week to over 50 automated cycles.
- Developing a searchable, validated playbook of marketing strategies for specific audiences.
- Optimizing the full AARRR funnel with pre-built specialized agent squads.
How Swarma: Growth Experiment Loops for Agent Teams Works
- Generate: Agents analyze the current strategy and propose a new experiment hypothesis.
- Run: The agent squad executes the experiment, producing actionable outputs like copy or research.
- Observe: A specialized judge agent scores the output based on defined metrics.
- Weigh: After a set number of cycles, the system compares the average performance against the baseline to issue a verdict.
- Stack: Validated patterns are automatically written to the team's playbook and strategy file to inform future cycles.
Swarma: Growth Experiment Loops for Agent Teams Setup
To get started with Swarma, install the package via pip and initialize your instance:
pip install swarma
swarma init
Next, set your OpenRouter API key to enable LLM calls:
echo "OPENROUTER_API_KEY=sk-or-..." >> ~/.swarma/instances/default/.env
For users integrating with Claude Code or other tools in the Openclaw Skills suite, you can start the MCP server using:
swarma serve --mcp
Swarma: Growth Experiment Loops for Agent Teams Data Schema & Taxonomy
Swarma organizes its intelligence in a structured directory format, making it easy to manage via Openclaw Skills:
| File | Purpose |
|---|---|
| team.yaml | Defines the goal, agent flow, schedule, and budget. |
| strategy.md | A living document containing validated patterns and anti-patterns. |
| agent.yaml | Configuration for individual agent roles, models, and targets. |
| program.md | Global context and constraints for the entire agent squad. |
All experiment results and metrics are logged locally, allowing for complete data ownership and transparency.
Swarma: Growth Experiment Loops for Agent Teams Advanced Features
- Automated Strategy Evolution: The strategy.md file updates itself based on experiment verdicts.
- 18 Pre-built Funnel Squads: Includes templates for SEO, pricing, cold outreach, and viral loops.
- MCP Integration: Full support for Model Context Protocol to work with modern AI IDEs.
- Real-world Metric Ingestion: Import external data via CSV or CLI to ground agent scoring in actual performance data.
- Cross-Team Knowledge Sharing: Use the QMD engine to share learnings across multiple different agent teams.
SKILL.md
Loading
Related Openclaw Skills
CET-4 Word Memory System
An AI-driven English vocabulary learning system based on the Ebbinghaus forgetting curve designed specifically for CET-4 exam preparation.
ClawLite CEO Review
A strategic AI planning agent that reviews, challenges, and upgrades your technical implementation plans from an executive perspective.
ClawLite Design Consult
An opinionated AI design consultant that builds cohesive design systems and visual documentation for your coding projects.
ClawLite Doc Sync
An automated workflow skill that synchronizes project documentation with actual code changes to ensure delivery consistency.
Textile & Clothing Repair
A comprehensive automation and instruction suite for essential clothing repair, maintenance, and stain removal.
Survival Basics: Home & Food
A comprehensive survival guide for mastering essential home maintenance, plumbing repairs, and high-efficiency budget cooking.
Featured*
