A robust automation engine that coordinates the complete 5-stage pipeline for proposing, validating, and applying improvements to AI skills.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install auto-improvement-orchestrator
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-improvement-orchestrator 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 Improvement Orchestrator?
The Improvement Orchestrator is the central nervous system for skill evolution within the Openclaw Skills ecosystem. It manages a sophisticated pipeline that moves candidates through five distinct stages: Generator, Discriminator, Evaluator, Executor, and Gate. By automating these transitions, it ensures that every modification to an AI agent is rigorously tested and verified before being committed.
This skill is particularly valuable for developers looking to implement self-improving loops. It features a unique Ralph Wiggum feedback mechanism that captures execution traces from failed attempts and injects them back into the generator for smarter retries. This ensures that even complex Openclaw Skills can be refined iteratively without manual intervention, maintaining high standards for both code and documentation.
Improvement Orchestrator Use Cases
- Running a full, unattended improvement cycle on one or more Openclaw Skills.
- Coordinating a multi-stage pipeline that includes blind scoring, task-based evaluation, and quality gating.
- Automating bulk updates to documentation, reference files, or guardrails across multiple skill sets.
- Implementing self-correction logic where failed improvements are automatically retried with trace-aware feedback.
How Improvement Orchestrator Works
- The orchestrator performs a baseline evaluation on the current skill to identify existing failures and injects them as source feedback.
- It triggers the generator to create candidate proposals based on the identified issues and user-provided feedback.
- A discriminator panel scores the candidates to identify the best possible improvement proposal.
- If a task suite is provided, the evaluator validates the candidate through real or mock execution cycles.
- The executor applies the chosen changes to the target directory while ensuring rollback capabilities are preserved.
- A final 7-layer quality gate checks the results, deciding whether to keep the changes or revert based on regression tests.
Improvement Orchestrator Setup
To deploy the orchestrator for your Openclaw Skills, use the command-line interface as follows:
python3 scripts/orchestrate.py \
--target /path/to/skill \
--state-root /path/to/state \
--task-suite tasks.yaml \
--max-retries 3
Parameters such as --source can be used to provide additional JSON feedback or memory files to guide the optimization process.
Improvement Orchestrator Data Schema & Taxonomy
The orchestrator generates structured logs and state files to track the lifecycle of an improvement. Key data outputs include:
| Artifact | Format | Purpose |
|---|---|---|
| pipeline-summary.json | JSON | Final report containing target details, attempts, and the final decision (keep/revert). |
| trace-{run_id}.json | JSON | Failure traces captured from the evaluator or gate for use in retry loops. |
| state-root/receipts | Directory | Persistent storage for artifacts generated during the multi-stage pipeline. |
Decisions are tracked through statuses like keep, revert, reject, and pending_promote.
Improvement Orchestrator Advanced Features
- Ralph Wiggum Retry Loop: Automatically attempts to fix failed improvements by providing detailed error traces back to the generator for up to 5 retries.
- Adaptive Complexity Skip: Improves performance by automatically skipping the evaluator stage for low-risk documentation or reference changes.
- Baseline Auto-Discovery: Runs a standalone evaluation before proposing changes to ensure the generator understands current skill weaknesses.
- Seamless Integration: Designed to work natively with other Openclaw Skills like the improvement-generator and improvement-gate for a modular optimization workflow.
SKILL.md
Loading
Related Openclaw Skills
GEO Monitor
A sophisticated audit tool that tracks website optimization progress for AI engines by comparing current performance against historical baselines.
Improvement Discriminator
A multi-signal scoring engine that evaluates and ranks improvement candidates using heuristics, blind panels, and LLM-as-Judge.
Autoloop Controller
A persistent orchestration wrapper that automates the continuous improvement of AI skills through iterative loops, convergence detection, and strict cost management.
Yufeng A-Stock Quantitative Screener
A sophisticated AI-powered quantitative stock screening assistant designed for A-share short-term trading opportunities.
VCF Log Insight Health Check
A specialized AI agent skill for monitoring the health, connectivity, and node status of VMware Aria Operations for Logs clusters within VCF environments.
Companion Agent
A technical framework for deploying a persistent secondary AI personality that shares your workspace but maintains its own unique identity and communication channels.
Featured*
