A proactive self-reflection and memory management skill that enables AI agents to learn from corrections and improve performance permanently.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install claw-self-improving-pro
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 claw-self-improving-pro 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 Self-Improving Agent (Proactive Self-Reflection)?
The Self-Improving Agent is a sophisticated framework for AI agents built on Openclaw Skills that implements a continuous feedback loop of self-criticism and self-learning. By leveraging a tiered memory structure, the agent evaluates its own output, catches recurring mistakes, and organizes patterns to ensure knowledge compounds over time without manual maintenance.
This skill transforms a standard AI into a learning entity that proactively analyzes its work after every task. By monitoring user corrections and preference signals, the agent maintains a local database of learnings, ensuring that once a mistake is corrected or a preference is stated, it becomes a permanent part of the agent's operational logic.
Self-Improving Agent (Proactive Self-Reflection) Use Cases
- When a user corrects a mistake or points out an error in reasoning.
- After completing significant multi-step tasks to evaluate the outcome against the intent.
- Identifying and automating recurring workflow patterns or styling preferences.
- Managing project-specific or domain-specific knowledge that needs to persist across sessions.
How Self-Improving Agent (Proactive Self-Reflection) Works
- Pre-Task Loading: The agent scans the HOT memory file and relevant project namespaces before starting work to align with known user preferences.
- Execution and Monitoring: While working, the agent identifies detection triggers such as explicit user corrections or preference signals.
- Post-Task Reflection: Upon completion of a task, the agent pauses to evaluate if the work met expectations and identifies specific lessons for improvement.
- Memory Logging: Lessons and corrections are logged to a tiered file system (HOT, WARM, or COLD) located in the local user directory.
- Pattern Promotion: The system automatically promotes lessons to the HOT tier if they are applied successfully three times within a seven-day window.
- Maintenance: Unused patterns are automatically demoted or archived according to predefined aging rules (30 or 90 days) to keep context windows clean.
Self-Improving Agent (Proactive Self-Reflection) Setup
To begin using this capability within your Openclaw Skills environment, ensure the directory structure is initialized correctly.
# Install the skill via clawhub
clawhub install self-improving
# The skill stores memory in ~/self-improving/
# If the directory does not exist, run the setup guide
cat ~/self-improving/setup.md
Self-Improving Agent (Proactive Self-Reflection) Data Schema & Taxonomy
The skill organizes data in a structured hierarchy located at ~/self-improving/ to ensure efficient retrieval and context management.
| File/Folder | Tier | Description |
|---|---|---|
memory.md |
HOT | High-frequency rules (max 100 lines), always loaded into active context. |
projects/ |
WARM | Directory containing per-project learnings and specific patterns. |
domains/ |
WARM | Domain-specific files for coding, writing, or communication styles. |
corrections.md |
LOG | A history of the last 50 corrections for evaluation and promotion. |
archive/ |
COLD | Historically relevant but currently unused patterns moved to long-term storage. |
reflections.md |
LOG | A dedicated log for the agent's post-task self-evaluations and lessons learned. |
Self-Improving Agent (Proactive Self-Reflection) Advanced Features
- Automatic Promotion and Demotion: Smart logic that manages context window usage by moving unused data to cold storage while keeping frequent lessons in HOT memory.
- Namespace Isolation: Prevents project-specific rules from leaking into global workflows unless explicitly promoted to the global memory tier.
- Conflict Resolution: Intelligent hierarchy (Project > Domain > Global) that ensures the most relevant context always wins when instructions overlap.
- Self-Compaction: Automated summarization of verbose entries and merging of similar corrections to keep memory files efficient and readable.
- Transparency Citations: The agent explicitly cites which memory file it is using during its operations to build user trust.
SKILL.md
Loading
Related Openclaw Skills
SDD Requirements Normalizer
A professional tool that converts fragmented product ideas into structured, high-fidelity requirement documents optimized for Spec Driven Development (SDD).
Hotels Near Attraction (flyai-hotel-near-attraction)
A location-focused travel skill that finds the closest hotels to specific landmarks with real-time distance sorting and integrated attraction data.
Beijing Pollen Monitor
A comprehensive utility for monitoring and forecasting pollen concentrations across all 16 districts of Beijing.
Saudi Noon Product Selection Skill
An AI-driven market research tool for identifying profitable product opportunities on the Saudi Arabian Noon e-commerce platform.
DaoReview: AI Document Auditor & Scorer
An intelligent document auditor that analyzes, scores, and provides actionable feedback on text-based files across multiple quality dimensions.
Jingyi Module Skill
An AI agent skill for searching official Jingyi Module commands and generating runnable E-Language code snippets within Openclaw Skills.
Featured*
