A persistent memory and learning system that allows AI agents to log technical experiments, analyze failures, and distill reusable lessons.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install experiment-notes
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 experiment-notes 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 Experiment Notes?
Experiment Notes is a robust memory layer designed to bridge the gap between agent action and long-term learning. While standard agents often repeat mistakes across sessions, this system provides a structured framework for recording trials, errors, and successful fixes. By utilizing Openclaw Skills integrated with this memory tool, agents can transition from simple execution to a continuous improvement cycle.
Built on a lightweight, local-first architecture, the skill ensures that every technical experiment—whether it is a Docker deployment or a complex script execution—is indexed and searchable. This creates a high-value knowledge base that prevents the repetition of known errors and accelerates the discovery of optimal solutions.
Experiment Notes Use Cases
- Preventing the repetition of failed CLI commands by checking historical experiment outcomes.
- Automatically documenting technical troubleshooting steps and successful fixes for future reference.
- Distilling high-level best practices from multiple trial-and-error attempts.
- Auditing agent performance and success rates across different technology stacks.
How Experiment Notes Works
- The agent queries the existing database using keywords or task descriptions to find similar historical experiments before attempting a new task.
- Upon completing a task, the agent logs the command used, the specific outcome (success, failure, or partial), and any error messages encountered.
- The system appends these records to a local JSONL file, ensuring data persistence without the need for a complex database server.
- Users or agents can periodically run the distillation process to aggregate multiple logs into generalized, high-level lessons.
- The statistical engine provides an overview of success trends and frequently used technical tags.
Experiment Notes Setup
To begin using this system within your Openclaw Skills environment, ensure you have Python installed. The data directory is initialized automatically upon first use.
# All interactions are handled via the CLI tool
python scripts/expnote.py log --task "example task" --outcome success
# Data is stored locally at:
# ~/.openclaw/memory/experiments/
Experiment Notes Data Schema & Taxonomy
The skill organizes its persistent memory using JSONL (JSON Lines) to allow for efficient appending and concurrent access. The schema is divided into two primary files:
| File | Description |
|---|---|
experiments.jsonl |
Contains raw logs of every trial, including timestamps, commands, errors, and specific fixes. |
lessons.jsonl |
Stores distilled best practices that have been extracted from multiple experiment logs. |
Typical Experiment Record:
id: Unique identifier for the experiment.task: Descriptive name of the action performed.outcome: Categorized as success, failure, or partial.tags: Array of technology labels (e.g., docker, git, python).lesson: A concise, actionable insight derived from the outcome.
Experiment Notes Advanced Features
- Similar Task Lookup: Uses keyword matching to provide agents with relevant historical context before execution.
- Knowledge Distillation: A dedicated command to transform raw, messy logs into clean, reusable technical documentation.
- Tag-Based Filtering: High-granularity search capabilities to isolate lessons learned within specific technical domains.
- Integration Snippets: Ready-to-use Markdown templates for AGENTS.md to standardize the experimental loop across different Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Wedding Planner Video
An AI-powered creative assistant that transforms wedding planning portfolios and service tiers into compelling short-form marketing videos.
Picasso
Picasso is a sophisticated design engineering skill that transforms AI agents into senior design engineers capable of building production-grade web interfaces.
aliyun-dns-cli
A professional CLI tool integration for managing Alibaba Cloud DNS records and Function Compute custom domain mappings.
Outlook Mail Reader
A sophisticated automation tool that controls local Chrome browsers via MCP to search, filter, and read Outlook emails.
Dayday: AI Learning & Practice Framework
Dayday is an AI learning skill that converts complex topics into actionable daily practice routines through a structured AI Teach, Learn, and Practice workflow.
Bound Memory
A hierarchical memory framework that enables AI agents to maintain a unified identity and context across fragmented sessions.
Featured*
