Taste

What This Skill Does

The Taste skill transforms your OpenClaw agent from a basic processor into a discerning aesthetic partner. Unlike standard AI models that rely on generic averages, the Taste skill enables the agent to learn your specific aesthetic preferences, stylistic boundaries, and creative nuances over time. By utilizing a structured 'Learning Loop,' the agent moves from an 'Uncalibrated' state to a 'Calibrated' state, where it can make judgments, curate content, and offer design feedback that feels authentically aligned with your personal or brand identity. The skill maintains a local repository of your feedback, turning every correction into a repeatable design pattern. It operates on the principle that taste is not inherent to AI, but a learned artifact of human interaction.

Installation

To integrate the Taste skill into your OpenClaw environment, execute the following command in your terminal:

clawhub install openclaw/skills/skills/ivangdavila/taste

Once installed, ensure your local filesystem has write access to the ~/taste/ directory, as the agent requires this space to maintain its correction history and calibration logs.

Use Cases

• Design Curation: Use the agent to filter through mood boards or UI mockups. It will learn whether you prefer minimalist whitespace or dense, information-rich interfaces.
• Writing Tone Alignment: Train the agent to recognize your specific voice. Over time, it will identify when a draft sounds too academic, too casual, or inconsistent with your preferred stylistic flair.
• Content Quality Control: Have the agent act as a first-pass critic for creative projects, checking work against established 'taste patterns' to save you time during the review process.

Example Prompts

1. "I'm working on a new landing page design; can you evaluate the color palette and typography against the preferences we've stored in the visual/domain?"
2. "I've just written a blog post. Based on our past corrections, does this prose match my preferred tone, or does it feel too stiff?"
3. "Here is a critique I received from a client. Let's record this as a correction in the design domain and extract a new pattern for future projects."

Tips & Limitations

• Be Specific: The agent relies on your feedback quality. Use the 'Genuine Curiosity Protocol' when correcting the agent; the more context you provide, the faster it reaches a calibrated state.
• Start Small: Begin by applying the skill to one specific domain (like writing or UI) before expanding. This ensures the agent builds a high-confidence calibration log rather than a diluted set of rules.
• Limitations: The agent cannot replicate human emotion, only recognize patterns associated with it. Do not use this skill to replace human judgment in highly sensitive or high-stakes artistic decisions; use it to augment your existing workflow.