A sophisticated multi-agent orchestrator that chains sequential and parallel AI tasks to build, debug, and research software projects.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-loops
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 agent-loops 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 Agent Loops?
Agent Loops is a powerful orchestration engine designed for the ecosystem of Openclaw Skills, enabling developers to execute complex, multi-step agent pipelines. By chaining specialized agents in sequential or parallel workflows, it facilitates the seamless transfer of real-world output between steps—ensuring that each agent builds directly upon the previous one's work. This skill acts as a project manager for AI agents, routing tasks to the most effective models and system prompts to handle everything from initial scoping to final deployment.
Agent Loops Use Cases
- Rapidly develop and ship new features, apps, or SaaS products using structured pipelines.
- Automated debugging and troubleshooting to diagnose root causes and generate regression tests.
- Comprehensive code auditing and security reviews performed by parallel specialized agents.
- Technical research and report generation to compare technologies or analyze market trends.
- Systematic refactoring and modernization of legacy codebases to improve maintainability.
- Seamlessly preparing and publishing new Openclaw Skills to the registry.
How Agent Loops Works
- The user selects a predefined workflow (like ship_feature or bug_fix) based on their specific intent.
- The orchestrator parses the user's natural language request and initiates the first step of the workflow.
- Each step runs via a specialized agent model routing with a role-specific system prompt designed for that task.
- Real output is captured and passed to the subsequent agents in the chain, maintaining context and project state.
- If the --apply flag is used, the agents perform the actions in the workspace; otherwise, a dry-run summary is provided.
Agent Loops Setup
To get started with this skill, install it via the hub:
clawhub install agent-loops
Or clone it manually into your skills directory:
git clone https://github.com/OpenClaw/agent-loops.git workspace/skills/agent-loops
This skill requires PyYAML for processing workflow configurations:
pip install pyyaml
Agent Loops Data Schema & Taxonomy
The orchestration logic is governed by YAML workflow definitions. The data flow is structured as follows:
| Component | Description |
|---|---|
| Workflow Registry | A collection of IDs (e.g., ship_feature, refactor) that map to specific logic gates. |
| Input Context | The user's natural language prompt which serves as the seed for the initial agent. |
| Agent Chaining | The mechanism where the output of Agent A becomes the context for Agent B. |
| Routing Metadata | Instructions that determine which model (e.g., Sonnet) is best suited for a specific step. |
| Execution Logs | Verbose or JSON-formatted output documenting every agent interaction and state change. |
Agent Loops Advanced Features
- Parallel execution support for running multiple agents (like a reviewer and a security auditor) at once.
- Fine-grained model control allowing users to override default agent-swarm routing with specific LLMs.
- JSON output mode for integrating agent results into external CI/CD pipelines or automated tools.
- Configurable per-step timeouts to manage long-running research or complex implementation tasks.
- Specialized hooks for the publication and distribution of other Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Spec-Brainstorm
A collaborative dialogue skill that transforms vague concepts into structured, system-agnostic project specification documents.
Vietnam Real-Time Air Quality Monitor (AQI)
A specialized skill for retrieving real-time Air Quality Index (AQI), temperature, and humidity data for major Vietnamese cities.
jmail
A powerful research and data analysis skill for querying the comprehensive Jeffrey Epstein archive via the jmail.world Data API and DuckDB.
Nail Art Pinterest Scraper
A specialized AI agent workflow for batch harvesting high-quality nail art images from Pinterest with built-in deduplication and anti-detection measures.
Subagent Dashboard
A centralized web dashboard for real-time monitoring, transcript viewing, and management of active OpenClaw subagents.
Subagent Tracker
A powerful monitoring tool for AI agent swarms that provides real-time visibility into subagent progress, status, and activity logs.