build-review-interface

Reviewing LLM traces at scale requires a browser-based annotation interface that renders raw JSON traces and tool-call logs. This Skill provides a single-file HTML UI to capture Pass/Fail labels and notes, streamlining human evaluation workflows.

To start labeling JSONL trace files, simply open the provided HTML UI in your browser and load your trace file. The interface supports keyboard shortcuts for rapid Pass/Fail labeling and optional notes attachment.

Yes, the annotation interface renders trace data in domain-native formats including Markdown, code, JSON, tables, and emails. This allows you to review intermediate steps and tool-call logs in their original context.

No dependencies or backend setup are required. The tool is a minimal, single-file HTML UI that runs entirely in the browser, allowing you to load trace JSONL files locally and export labeled data without external services.

The best way to export labeled LLM traces for eval pipelines is using the JSONL format. The UI exports your Pass/Fail labels and notes as a JSONL file, ensuring direct integration with QA workflows and evaluation systems.

The single-file HTML UI is limited to local browser processing without server-side storage or collaboration features. It is designed for individual rapid labeling of JSONL traces rather than multi-user annotation platforms.