Hyperlogue has released R3, a new skill for AI agents that allows developers to review and refine generated artifacts locally. The tool, showcased on Hacker News, aims to solve the friction of iterating on AI-created UI prototypes, interactive explainers, and design docs by keeping the review process within the developer's local environment.

Visual Feedback Loop

R3 enables users to click directly on specific UI elements within an artifact to leave anchored notes. This feedback is batched and sent back to the agent with precise element and viewport context, ensuring the AI understands exactly which button, heading, or input needs adjustment. The system supports attaching images and commenting on specific video frames, providing unambiguous visual cues for the agent.

Version Control and Comparison

When an agent publishes a revision, R3 maintains a history of all versions, allowing users to revisit earlier iterations. The tool includes a side-by-side comparison feature, letting developers check the original against the fix to verify if the issue was resolved. This creates a clear audit trail of the design evolution, moving from the first draft to the finished product.

Installation and Access

Developers can install the R3 skill globally using the command npx skills add hyperlogue/r3 --skill publish-artifacts-to-r3 --global. Once installed, agents can create project forms and publish HTML artifacts to a local server (e.g., http://localhost:8891/artifact_example). A demo is available for those wanting to test the workflow before integrating it into their stack.

Key Takeaways

  • R3 facilitates local review of AI-generated artifacts without leaving the developer's environment.
  • Users can leave anchored comments on specific UI elements and send batched feedback to agents.
  • The tool supports version history and side-by-side comparison of revisions.
  • Installation is handled via the npx command line interface.

The Bottom Line

This is a smart move for the OpenClaw ecosystemβ€”by anchoring feedback to specific DOM elements, R3 removes the guesswork from AI iterations, making agent collaboration significantly more precise and efficient.