Experiment Report Skill

A complete experiment-report workflow for AI coding agents. Describe an experiment, implement and run it, preserve the original results, build an interactive visualization, and approve the frontend before the final report is written.

The workflow requires at least four screenshots of the actual rendered frontend, both Markdown and DOCX deliverables, and at least three rounds of checks against the original data. Failed experiments and limitations are reported honestly; the workflow does not fabricate results or automatically reduce report quality.

The included Word template supports native OMML equations, automatic headings and table of contents, independent list numbering, consistent figure references, a cover page, and page numbering. It validates required report content and screenshots instead of silently generating an empty report.

Version 1.0.1 includes a cross-platform updater with backups, recovery after failed or interrupted updates, and manual rollback. The updater checks the public GitHub repository at most once daily at task start and downloads newer versions; this network and file-update behavior is intentional and documented. Git development checkouts are not automatically overwritten.

Requirements: a compatible AI coding agent with local code execution and browser/screenshot tools; Node.js 18+ for the updater and DOCX generation; docx 9.6.1 and image-size 2.0.4 (lockfile included). Experiment-specific dependencies vary. The instructions and default report structure are primarily in Chinese.

Free, open-source release under the MIT License. LICENSE is included in the package. The same source remains publicly available at https://github.com/lyzbcy/experiment-report-skill. This listing does not revoke the permissions granted by the MIT License.