// the find
simonlin1212/vibe-astock
A 股短线复盘看板：涨停池·连板梯队·龙虎榜·板块资金一屏看完，赚钱效应/晋级率/梯队断层/情绪周期等派生指标纯计算直出（不经过 AI），AI 只把数据串成能读的盘面研判。全本地运行，可用 Claude/Codex 订阅免 API key。| A-share short-term daily-review dashboard: derived sentiment metrics computed locally, AI only writes the narrative. No API key needed.
A local-only dashboard for short-term A-share trading review — limit-up pools, promotion ladders, dragon-tiger list, sector flow — where all the actual metrics are computed by plain Python and only the narrative layer is handed to an LLM. Built for Chinese day-traders who already pay for a Claude or Codex subscription and want to skip a separate API key.
Hard separation between deterministic computation and AI output: market breadth, ladder continuation, and sentiment scores come from a Python/FastAPI pipeline, and the LLM's job is explicitly to narrate and cross-check evidence, not produce numbers — this avoids the usual failure mode of an LLM inventing stats. The backtest module is a real deterministic engine (gate.py, validation.py) that reports zero-fill conditions honestly instead of faking a fill. It reuses your existing Claude Code / Codex CLI login instead of requiring a separate paid API key, which is a genuinely useful trick if you already have one of those subscriptions. Binds to 127.0.0.1 by default and requires an explicit VIBE_ALLOW_HOSTS allowlist before it'll answer on any other interface.
Single maintainer, and the README itself is mostly a wall of self-reported audit language ('1287 项通过', '34 项运行契约') rather than a CI badge or public dashboard — there's no way to verify those numbers independently. Data pipeline leans on scraping public sources (akshare, 问财/iwencai) that can change or rate-limit without notice, and the project's own docs acknowledge partial coverage and stale snapshots as a known failure mode. It's a Chinese A-share tool only — the English README exists but the product and its data sources don't generalize past that market. Seven UI modules, five+ AI backend integrations, and a bespoke backtest engine is a lot of surface for one person to keep consistent; expect version drift between modules (the README calls this out directly for older reports not being re-validated after upgrades).