A controlled autonomous iteration engine that executes agent tasks repeatedly until specific success criteria are met.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install loop
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 loop 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 Loop?
The Loop skill provides a structured framework for AI agents to perform iterative problem-solving within the Openclaw Skills ecosystem. Instead of relying on a single attempt, this skill allows agents to execute a task, verify the outcome against defined success criteria, and automatically retry if the goal is not met. It is designed to bridge the gap between simple task execution and complex, self-correcting autonomous workflows.
Built with safety as a priority, Loop incorporates strict guardrails, such as a hard limit on iterations and prohibitions against automatic production deployments or system configuration changes. This ensures that while the agent has the autonomy to keep trying, it remains within a controlled environment, making it a reliable tool for professional software development and automation tasks.
Loop Use Cases
- Debugging complex code issues where the initial fix may not satisfy all test cases.
- Refining documentation or data outputs through multiple passes until quality benchmarks are achieved.
- Automating repetitive troubleshooting steps that require verification after each action.
- Scenarios where a user provides a high-level goal and wants the agent to persist until the verification command passes.
How Loop Works
- Initialization: The agent defines a specific task, success criteria, and a maximum iteration count (defaulting to 5).
- Execution: The agent performs the primary task based on the current state and any learnings from previous attempts.
- Verification: A verification step is triggered, typically running a command like a test suite, to evaluate the result against the success criteria.
- Logging: The skill records the results of the iteration, including what worked and what failed, into a persistent history log.
- Evaluation: If criteria are met, the skill exits successfully. If not, it assesses whether to retry based on the iteration limit or identifies unrecoverable errors.
- Final Summary: Upon completion or reaching the limit, the skill provides a comprehensive summary of all attempts and identifies failure patterns.
Loop Setup
To initialize the Loop skill for use with Openclaw Skills, you must first create the necessary local storage directories to handle history and active state:
mkdir -p ~/loop/history
The skill operates locally and requires no external binary dependencies, relying on JSON and Markdown files for state management.
Loop Data Schema & Taxonomy
Loop maintains a structured data hierarchy within the ~/loop/ directory to ensure transparency and learning across sessions:
| File Path | Purpose |
|---|---|
~/loop/active.json |
Tracks the state and metadata of the currently active iteration loop. |
~/loop/history/{loop-id}.json |
Individual log files containing the step-by-step history of completed loops. |
~/loop/learnings.md |
A cumulative document used to track cross-loop patterns and improve future agent performance. |
Loop Advanced Features
- Pattern Recognition: Maintains a
learnings.mdfile to help agents identify and avoid repeating mistakes across different sessions. - Safety Hard Limits: Implements a non-bypassable hard limit of 10 iterations to prevent infinite loops and resource waste.
- Audit Trails: Generates detailed JSON logs for every attempt, providing a full forensic record of agent actions.
- Verification Integration: Supports standard CLI-based verification commands, allowing for objective, test-driven task completion.
SKILL.md
Loading
Related Openclaw Skills
London Intelligence & Navigation
A specialized knowledge layer providing contextual intelligence for visiting, living, and working in London.
Logo Generation Agent Skill
A comprehensive framework for generating professional-grade logos and app icons using advanced AI image models.
Lithuanian Natural Language
A specialized skill for generating natural, human-sounding Lithuanian text that avoids the typical stiff and formal AI tone.
Literature
An adaptive literary assistant that guides users from personal reading responses to rigorous, evidence-based scholarly analysis.
Los Angeles
A comprehensive intelligence layer for navigating Los Angeles as a visitor, resident, or professional.
macOS
A technical guide for mastering macOS system administration, command-line nuances, and automation workflows.
METADATA
Featured*
