A persistent orchestration wrapper that automates the continuous improvement of AI skills through iterative loops, convergence detection, and strict cost management.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install auto-autoloop-controller
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 auto-autoloop-controller 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 Autoloop Controller?
The Autoloop Controller is a sophisticated orchestration tool designed for the Openclaw Skills ecosystem. It acts as a persistent wrapper around the improvement-orchestrator, enabling developers to run long-term, automated optimization cycles. By managing state across sessions, it ensures that your AI agents undergo rigorous refinement without requiring manual intervention for every iteration.
This skill is particularly valuable for complex optimization tasks that require multiple passes to achieve peak performance. It intelligently monitors progress using convergence detection algorithms to identify score plateaus or oscillations, ensuring that computational resources are not wasted on marginal or non-existent gains. Whether you are performing overnight runs or scheduled maintenance, it provides the stability and oversight needed for high-quality agent development.
Autoloop Controller Use Cases
- Running multi-iteration optimization cycles for Openclaw Skills during overnight batch processing.
- Scheduling periodic skill refinement via cron jobs to maintain performance against evolving benchmarks.
- Executing fixed-length improvement sprints with automatic termination upon reaching a performance plateau.
- Maintaining persistent improvement state across different developer sessions to ensure continuity.
How Autoloop Controller Works
- The controller initializes by loading or creating an autoloop_state.json file in the designated state root.
- It triggers a subprocess call to the improvement-orchestrator for the target skill directory.
- After each iteration, it collects dimension scores from the improvement-learner to calculate a weighted performance metric.
- It evaluates four termination conditions: maximum iterations, cost cap, score plateau, or oscillation detection.
- If conditions allow, it enters a cooldown phase (in continuous mode) or saves state for the next scheduled trigger.
- All artifacts, including handoff notes and iteration logs, are persisted to ensure cross-session context survival.
Autoloop Controller Setup
To begin using the controller for your Openclaw Skills, ensure you have the core repository cloned and dependencies installed. Use the following CLI structure to launch an improvement loop:
# Standard run with a 5-iteration limit and $10 cost cap
python3 scripts/autoloop.py \
--target /path/to/your/skill \
--state-root ./state_data \
--max-iterations 5 \
--max-cost 10.0 \
--mode single-run
# Continuous mode for long-term refinement
python3 scripts/autoloop.py \
--target /path/to/your/skill \
--state-root ./state_data \
--mode continuous \
--cooldown-minutes 15
Autoloop Controller Data Schema & Taxonomy
The Autoloop Controller maintains a structured data environment to ensure reliability and transparency throughout the improvement lifecycle:
| File | Purpose | Format |
|---|---|---|
autoloop_state.json |
The master state file containing iteration counts, cumulative costs, and score history. | JSON |
iteration_log.jsonl |
A line-delimited log record of every completed iteration for analytics. | JSONL |
handoffs/iteration-N.md |
Contextual handoff files containing scores and remaining tasks for the next iteration. | Markdown |
last_failure_trace |
Captured stderr and tracebacks in the event of a subprocess crash. | Text |
Autoloop Controller Advanced Features
- Intelligent convergence detection that stops the loop if scores haven't improved over a user-defined window.
- Oscillation protection to prevent the agent from repeatedly alternating between conflicting improvements.
- Cross-session resume capability allowing the controller to pick up exactly where it left off after a manual stop or system reboot.
- Flexible execution modes including single-run, continuous sleep-loop, and scheduled execution for integration with system schedulers.
- Integrated cost tracking with a hard stop USD limit to prevent unexpected API expenditure during automated runs.
SKILL.md
Loading
Related Openclaw Skills
Yufeng A-Stock Quantitative Screener
A sophisticated AI-powered quantitative stock screening assistant designed for A-share short-term trading opportunities.
HTTP Requests
A Python-powered skill for executing complex HTTP requests without the escaping and quoting issues common in traditional curl commands.
Improvement Evaluator
A technical benchmarking tool that measures whether Skill modifications actually improve AI performance on real-world tasks using automated pass/fail criteria.
Zhang Lei Thinking Operating System
An advanced AI cognitive framework modeled after Zhang Lei's value investment philosophy to guide long-term business and investment decision-making.
Improvement Discriminator
A multi-signal scoring engine that evaluates and ranks improvement candidates using heuristics, blind panels, and LLM-as-Judge.
GEO Monitor
A sophisticated audit tool that tracks website optimization progress for AI engines by comparing current performance against historical baselines.
Featured*
