A persistent control loop that automates the continuous improvement of AI skills through convergence detection and budget-aware orchestration.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install 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 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 specialized automation wrapper designed to manage the lifecycle of Openclaw Skills improvement. It repeatedly invokes the improvement-orchestrator in a persistent loop, moving a skill from emerging to solid quality tiers by capturing compounding gains across multiple iterations. Unlike single-shot tools, this controller handles the complexities of long-running optimization tasks, including state persistence, error handling, and resource management.
By wrapping the core generate-evaluate-gate pipeline, the Autoloop Controller ensures that each round of refinement builds upon the last. It is particularly effective for developers who need to optimize complex agents overnight or on a recurring schedule, providing the reliability required for production-grade Openclaw Skills development.
Autoloop Controller Use Cases
- Continuously improve a skill over multiple iterations until performance scores plateau.
- Run unattended overnight optimization research to maximize skill quality.
- Schedule periodic improvement cycles via system cron for production agents.
- Manage budget-constrained batch improvements with specific USD cost ceilings.
- Resume interrupted improvement runs seamlessly from a saved disk state.
How Autoloop Controller Works
- The controller initializes the session by loading the state of the target Openclaw Skills from the designated state root.
- It executes a full 5-stage improvement pipeline iteration: generate, discriminate, evaluate, execute, and gate.
- Upon completion of an iteration, the controller evaluates the results and updates the internal score history.
- It checks five independent termination conditions: maximum iterations, cost caps, score plateaus, oscillation patterns, and consecutive error limits.
- If all conditions are clear, the current state is persisted to disk and the loop continues (or waits for a cooldown in continuous mode).
- Once a termination condition is met, the controller halts and outputs a final summary and performance log.
Autoloop Controller Setup
To begin using the controller for your Openclaw Skills, use the following CLI command structure:
python3 scripts/autoloop.py \
--target ./skills/your-skill-path \
--state-root ./state-directory \
--max-iterations 5 \
--max-cost 20.0 \
--mode single-run
For scheduled improvements, you can add a script to your crontab that runs the controller in scheduled mode, allowing it to pick up the state from the previous run automatically.
Autoloop Controller Data Schema & Taxonomy
The controller organizes its data and metadata to ensure full traceability and recovery for Openclaw Skills:
| Artifact | File Format | Description |
|---|---|---|
| autoloop_state.json | JSON | Stores global session state, including iterations completed, total cost, and score history. |
| iteration_log.jsonl | JSONL | An append-only log with timing, cost, and decision metrics for every completed iteration. |
| Orchestrator Outputs | Directory | Contains iteration-specific artifacts like candidate diffs and gate evaluation results. |
| Console Summary | Stdout | A human-readable report generated upon loop termination. |
Autoloop Controller Advanced Features
- Intelligent Convergence Detection: Identifies both plateaus (no improvement) and oscillations (alternating keep/reject) to save resources.
- Automated Circuit Breaker: Halts execution after a configurable number of consecutive failures to prevent resource waste during infrastructure outages.
- Cross-Session Persistence: Saves the complete state after every iteration, allowing for zero-progress-loss recovery after system crashes.
- Budget Guardrails: Hard cost caps ensure that automated improvement of Openclaw Skills stays within defined financial limits.
- Multiple Execution Modes: Supports single-run, continuous long-running, and cron-friendly scheduled execution patterns.
SKILL.md
Loading
Related Openclaw Skills
Smart Summarizer
A sophisticated content processing tool that transforms long-form text, videos, and documents into actionable summaries with TL;DRs and key takeaways.
FitDietKernel
FitDietKernel is a precision nutrition knowledge base that transforms AI agents into expert diet coaches for carbon cycling and macro management.
Stock Summary
A professional-grade stock market analysis tool providing real-time quotes, technical indicators, and automated trend charting.
PubMed Literature Search & Analysis
An advanced clinical medicine literature retrieval system that integrates PubMed data with JCR impact factors for high-precision research analysis.
E-Invoice Local Compliance Partner Selection
A sophisticated AI skill designed to systematically identify, filter, and evaluate local e-invoicing partners and compliance channels in international markets.
Energy Market Intelligence
A specialized AI skill for generating sourced, decision-ready Alberta energy market briefings grounded in AESO data and real-world market structures.
Featured*
