tribunal17

Imported from Yiqing-dev/tribunal17 (AGENTS.md). Install upstream with npx skills add Yiqing-dev/tribunal17. Copyright stays with the author.

Codex Project Notes

When the user asks to start the local workbench, run:

python start_workbench.py

Primary launch phrase:

论衡十七司，升堂！【600519，000858，002594】

When the user sends this phrase, extract the bracketed ticker list and start the daily stock-research workflow for those tickers. The scripts can consume the same phrase directly:

python start_workbench.py "论衡十七司，升堂！【600519，000858，002594】"
python -m subagent_pipeline.batch_process "论衡十七司，升堂！【600519，000858，002594】"
python -m subagent_pipeline.lunheng_court "论衡十七司，升堂！【600519，000858，002594】"

When this phrase is used, the current Claude Code / Codex language model should call subagents directly. Run lunheng_court first to write data/reports/daily_watchlist.json and data/reports/lunheng_runbook.md; then use subagents for bounded roles per ticker: bull case, bear case, risk/review, and final synthesis. Python scripts do not call subagents themselves.

If the user provides a daily stock list, scope the workbench to that list:

python start_workbench.py --tickers 600519,000858,002594
python start_workbench.py --tickers-file watchlist.txt

The command regenerates data/reports/workbench.html, starts the local server, and prints the URL:

http://127.0.0.1:8765/workbench.html

Use another port if 8765 is already occupied:

python start_workbench.py --port 8770

Use --no-generate when the user only wants to serve existing reports.

To build HTML reports from existing agent outputs for the same daily list:

python -m subagent_pipeline.batch_process --tickers 600519,000858,002594
python -m subagent_pipeline.batch_process --tickers-file watchlist.txt

For 论衡十七司，升堂！【...】, multi-agent/subagent discussion is the default LLM-runtime behavior. Keep internal review plumbing separate from the stockholder-facing HTML report unless the user asks to include it.