A specialized feedback loop for AI agents to develop refined aesthetic judgment and calibration by learning directly from human preferences.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install taste
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 taste 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 Taste?
The Taste skill is a framework designed to bridge the gap between an AI's objective processing and human aesthetic judgment. It operates on the core principle that taste is a learned trait acquired through exposure and iterative feedback from a human collaborator. By implementing this skill, an agent moves from being uncalibrated to becoming a refined partner in creative tasks.
As a foundational part of the Openclaw Skills ecosystem, it provides a structured way for agents to ask genuine questions, record corrections, and extract transferable patterns. This ensures that the agent's output aligns with the specific aesthetic standards and nuances of the user, rather than relying on generic model defaults.
Taste Use Cases
- Calibrating an agent's visual design preferences for frontend development projects.
- Refining the prose and tone of voice in technical documentation or creative writing.
- Developing a shared aesthetic language between a developer and an AI coding partner.
- Training an agent to recognize and avoid specific design antipatterns in a professional context.
How Taste Works
- The agent states a tentative judgment, explaining why it believes a specific choice is aesthetically sound.
- The agent initiates the Genuine Curiosity Protocol to solicit specific, actionable feedback from the human.
- Upon receiving a correction, the agent probes for the underlying reasoning (the "Why") to ensure deep learning.
- The feedback is logged into the local workspace, where generalizable rules are extracted and stored.
- The agent updates its calibration confidence level for the specific domain, moving from Uncalibrated toward Calibrated.
Taste Setup
To initialize this functionality within Openclaw Skills, create the required directory structure in your home directory:
mkdir -p ~/taste/{corrections,preferences,patterns}
touch ~/taste/calibration.md
Ensure that you have the reference files like visual.md or writing.md available in your agent's knowledge base to serve as the initial baseline for judgment.
Taste Data Schema & Taxonomy
The skill organizes its learning data within the ~/taste/ directory using the following taxonomy:
| Directory/File | Description |
|---|---|
corrections/ |
Subfolders per domain containing date-stamped Markdown files of specific corrections. |
preferences/ |
Documents human-stated aesthetic preferences categorized by domain (e.g., design, writing). |
patterns/ |
Extracted generalizable rules derived from accumulated corrections. |
calibration.md |
A master ledger tracking the agent's current confidence level (Uncalibrated, Learning, Calibrating, Calibrated) per domain. |
Taste Advanced Features
- Genuine Curiosity Protocol: Enforces specific, high-signal questioning over vague queries to maximize knowledge transfer.
- Contextual Reference Loading: Dynamically loads domain-specific guidelines (e.g.,
antipatterns.md) when evaluating aesthetic work. - Pattern Extraction: Automatically transforms individual human corrections into reusable rules for future judgments.
- Calibration-Based Behavior: Adjusts agent autonomy based on confidence; uncalibrated domains require constant verification, while calibrated domains offer confident assertions as part of the Openclaw Skills workflow.
SKILL.md
Loading
Related Openclaw Skills
Task List
A conversational task manager for Openclaw Skills that brings professional GTD workflows to your AI coding agent through a local, file-based workspace.
Talk
Integrate real-time voice capabilities, phone agents, and live speech interactions into your AI agentic workflows.
Tai Chi (Practice Planner, Form Coach & Balance Tracker)
An AI-powered tai chi companion for building practice plans, improving form alignment, and tracking balance-focused consistency.
Systems Architect
A specialized architectural framework for designing high-availability infrastructure, secure networks, and scalable cloud systems.
Tattoo Skill
A comprehensive planning and research framework for tattoo discovery, cultural validation, and physical preparation.
Taxi
A comprehensive AI agent skill for managing every aspect of ground transportation from price comparison to expense tracking.
Featured*
