A comprehensive AI toolkit for benchmarking, comparing, and optimizing agent prompts and evaluation results through persistent, file-based logging.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-learner
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 agent-learner 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 Agent Learner?
Agent Learner is a specialized command-line toolkit designed for developers and researchers who need to rigorously fine-tune AI agent performance. By providing a structured, file-based system for logging configurations, benchmark results, and prompt variations, it acts as a central repository for iterative AI development. As a valuable part of the Openclaw Skills ecosystem, it allows for deep analysis of agent behavior without the overhead of external databases or complex API dependencies.
The tool facilitates a data-driven approach to prompt engineering, allowing users to compare different model outputs and track performance metrics over time. Its lightweight design ensures that all data remains local and accessible, making it an ideal choice for developers looking to maintain a clean and efficient workflow while scaling their agentic capabilities.
Agent Learner Use Cases
- Tracking performance benchmarks across different Large Language Models and system configurations.
- Versioning prompt engineering iterations to systematically identify the most effective instruction sets.
- Monitoring API consumption costs and token usage metrics to manage operational budgets.
- Comparing outputs from multiple agents to determine task-specific performance superiorities.
- Conducting full-text searches across historical logs to identify long-term trends or specific failure modes.
How Agent Learner Works
- Initialize the environment using the status command to verify the local data directory is ready for storage.
- Execute specific data commands like benchmark or prompt followed by your observations to create timestamped entries.
- Retrieve historical data instantly by running commands without arguments to view the most recent activity.
- Use the stats and search functions to analyze performance trends and locate specific historical records across all categories.
- Generate portable datasets using the export command to move your logs into JSON, CSV, or TXT formats for external reporting.
Agent Learner Setup
To get started with this tool from the Openclaw Skills collection, ensure your environment supports Bash and standard Unix utilities. No external API keys are required. You can initialize the system and check its health with the following command:
agent-learner status
All data is stored locally by default in the ~/.local/share/agent-learner/ directory, ensuring your benchmarking data remains private and under your control.
Agent Learner Data Schema & Taxonomy
Agent Learner organizes its data into a structured hierarchy of plain text files using a consistent timestamp|value taxonomy.
| Data Component | Format | Description |
|---|---|---|
| Category Logs | $DATA_DIR/<cmd>.log |
Individual files for each command category (e.g., benchmark.log, prompt.log). |
| History Log | $DATA_DIR/history.log |
A comprehensive audit trail of every command executed by the user. |
| Export Files | $DATA_DIR/export.<fmt> |
Processed data output in JSON, CSV, or TXT formats for external analysis. |
Agent Learner Advanced Features
- Case-insensitive full-text search across all log files for rapid data retrieval.
- Automated summary statistics including entry counts, total disk usage, and activity timelines.
- Multi-format data export capabilities to support integration with advanced data analytics tools.
- Dual-mode command logic that intelligently switches between data entry and data retrieval based on user input.
- High-performance logging using standard Unix pipes for minimal latency during benchmarking sessions.
SKILL.md
Loading
Related Openclaw Skills
Batch Content Sprint OS
A systematic framework for creators to plan and execute weekly batch content production while preventing burnout through structured time-blocking.
BBC Crawler MaxClaw
A high-performance universal web crawler designed for BBC News and dynamic websites using advanced extraction engines.
Weather Query
A streamlined AI skill for querying current global weather conditions and temperatures via simple city names.
Fixer Currency API Integration
A high-performance currency exchange and conversion skill for AI agents to access real-time financial data.
Agent QA Gates
A multi-tiered validation framework designed to prevent hallucinations, data leaks, and formatting errors in AI agent outputs.
Creator Attribution Lite
A lightweight attribution model designed to connect content outputs directly to business outcomes and ROI.
Featured*
