trading-agents

Authenticate against the QuantMind API, pull the stock's features, risk score, inference history, K-line, and news endpoints, then run at least two opposing bull and bear sub-agents citing concrete data. Assemble the standardized Markdown template and convert it to PDF with md_to_pdf_report.py.

It uses QuantMind local APIs: 371-dimension research features, a 6-dimension risk scorecard, model inference score history across multiple models, 120-day K-lines, daily market selection rankings, and RSS news articles filtered by ticker, industry, sentiment, and event tags.

Yes, the autonomous mode works with any LLM such as deepseek, qwen, glm, openai, or minimax because the agent itself fetches data and assembles the report. The optional container TradingAgents pipeline requires configured LLM API keys inside the container.

Chinese rendering fails when no embeddable font is found. Place WQY MicroHei and ZenHei TTF files in docker/training/fonts/ so they bind-mount into the container; otherwise the script falls back to STSong-Light CID fonts that some viewers cannot render.

The directory name must exactly match the backend market names such as A股市场 with no space, under /data/reports/trading_agents/{market}/{stock name}/. Verify with the files/list API and check that the filename follows the {name}{code}_{date}_投研分析报告 convention.

The report must mark the section as data missing rather than fabricating content. For missing news, the agent reminds the user to add RSS sources in the admin panel; for missing inference records, it notes that no recent inference data exists.