qingyan_research_report

The Skill runs iterative search and page visits to gather authoritative sources, reflects on coverage and depth, then calls the generate_html script with a complete self-contained HTML document. The script reads the HTML from standard input and saves it as a file named after the report title.

Charts are built with Chart.js and embedded in the HTML using a canvas element wrapped in a figure tag with a figcaption. The Skill triggers chart generation when the text contains data comparisons, trends over time, distributions, or dense data tables.

Yes, a Table of Contents is inserted after the H1 heading, collecting all H2 headings and their following H3 subheadings. Each entry links to a slug-based anchor id, and numbering adapts to the document language, using Chinese or Arabic numerals.

Supported types include bar, line, and combined charts rendered via Chart.js, with variety encouraged across a report. Doughnut charts are explicitly forbidden, and the y-axis maximum should be 1.2 times the largest data value.

Reading from STDIN via a heredoc avoids shell argument length limits that would truncate large HTML documents. The script validates that content was received, sanitizes the title into a safe filename, and writes the file with UTF-8 encoding.