An automated engine that produces ranked improvement candidates for AI skills based on feedback loops, failure traces, and target analysis.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install auto-improvement-generator
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-generator 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 Generator?
The Improvement Generator is a core technical utility designed to optimize the lifecycle of Openclaw Skills by generating structured improvement candidates. It analyzes target skills, feedback signals, and historical memory to propose actionable changes across six distinct categories, including documentation, guardrails, and prompts.
By leveraging trace-aware generation, the tool ensures that AI agents do not repeat past mistakes. It intelligently deprioritizes categories that failed in previous runs and boosts alternative strategies, making it an essential component for developers building resilient and self-improving AI workflows.
Improvement Generator Use Cases
- Generating structured improvement candidates for a specific target skill.
- Injecting failure traces from previous runs to prioritize retries effectively.
- Avoiding repetitive failure patterns by analyzing memory sequences of three or more failed attempts.
- Creating targeted evaluator-driven fixes based on baseline-failures.json data.
How Improvement Generator Works
- The tool ingests a target skill directory and optional feedback sources like memory.json or failure traces.
- It scans for correction hotspots and evaluator failures to identify specific dimensions requiring optimization.
- Candidate strategies are generated across categories such as docs, guardrails, and prompts, each assigned a risk level.
- If a prior failure trace is provided, the generator adjusts candidate ranking to deprioritize the failed category.
- A structured JSON artifact is produced, containing an execution plan for the proposed improvements.
Improvement Generator Setup
To install and run the generator within your Openclaw Skills environment, use the following CLI pattern:
# Basic candidate generation
python3 scripts/propose.py --target /path/to/skill --state-root ./state
# Generation with failure trace injection
python3 scripts/propose.py \
--target /path/to/skill \
--trace failure_trace.json \
--max-candidates 4 \
--source feedback.jsonl
Improvement Generator Data Schema & Taxonomy
The skill produces a structured JSON artifact containing metadata and execution instructions.
| Field | Type | Description |
|---|---|---|
schema_version |
String | Version of the output format (e.g., 1.0). |
run_id |
String | Unique identifier for the generation session. |
candidates |
Array | List of proposed changes including category, risk_level, and execution_plan. |
failure_trace_used |
Boolean | Indicates if a prior failure trace influenced the output. |
truth_anchor |
String | Path to the generated candidate version file. |
Improvement Generator Advanced Features
- Trace-Aware Generation: Automatically shifts strategy when a specific category (e.g., docs) triggers a gate rejection.
- Evaluator-Driven Fixes: Direct integration with Claude to propose high-priority SKILL.md fixes for tasks with low evaluation scores.
- Correction Hotspot Analysis: Prioritizes improvements based on frequency of manual user corrections.
- Multi-Category Support: Handles diverse improvement types from low-risk documentation updates to medium-risk prompt restructures.
SKILL.md
Loading
Related Openclaw Skills
Improvement Learner
Improvement Learner provides a structural quality assessment and self-improvement loop for Openclaw Skills using a 6-dimensional evaluation framework.
Improvement Executor
A precision execution engine that applies approved AI candidates to target files with built-in safety guards and automatic backup systems.
ByteDance Organization Framework
A management methodology skill based on the ByteDance philosophy of Context not Control and algorithm-driven organizational systems.
Feelings Framework
A persistent emotional layer for AI agents that tracks mood and feeling intensities over time to shape dynamic response behaviors.
Lei Jun Strategic Mindset
A strategic AI skill that applies Lei Jun's five core mental models to evaluate products, business trends, and operational efficiency.
Emotion AI Companion
A sophisticated emotional intelligence layer that provides permanent memory and cross-agent experience sharing for your AI ecosystem.
Featured*
