自托管加密定位分析，基于 tradingriot.com 的加密货币 完全基于免费数据源构建的 stack。

• 数据：小时资金，未平合约（币安 + Bybit + 超流动）， 汇总清算（Coinalyze）、BTC/ETH期权（Deribit）、1小时OHLCV。 历史记录自2021年1月起通过Binance Vision大批量档案回填。
• 存储：parquet 文件，位于data/中，通过DuckDB查询。GitHub操作 cron 按小时收取并提交。
• 分析：资金/开放投资/动量的滚动Z分数，横断面分析 筛选器、综合制度指数、期权期限结构和25-Delta偏置。
• Dashboard：Streamlit（streamlit run dashboard/app.py）。
• LLM 层：CLAUDE.md 用 Claude 文档用于临时分析的模式 代码;python -m briefing.generate 通过 无头 Claude 代码（[claude -p）——使用 Claude 订阅，无 API 密钥 （每周通过Actions发布）

布置

Python 3.13。requirements.txt 钉住通过测试的精确版本;升级 一个是故意的，运行python -m unittest discover -s tests，然后提交。

秘密（GitHub Actions 仓库秘密;本地复制 .env.example 到 .env）：

Crypto Dashboard

Self-hosted crypto positioning analytics, modeled on tradingriot.com’s crypto stack, built entirely on free data sources.

• Data: hourly funding, open interest (Binance + Bybit + Hyperliquid), aggregated liquidations (Coinalyze), BTC/ETH options (Deribit), 1h OHLCV. History backfilled from Jan 2021 via Binance Vision bulk archives.
• Storage: parquet files in data/, queried with DuckDB. A GitHub Actions cron collects hourly and commits.
• Analytics: rolling z-scores for funding/OI/momentum, cross-sectional screener, composite regime index, options term structure and 25-delta skew.
• Dashboard: Streamlit (streamlit run dashboard/app.py).
• LLM layer: CLAUDE.md documents schemas for ad hoc analysis with Claude Code; python -m briefing.generate writes a Claude market briefing via headless Claude Code (claude -p) - uses the Claude subscription, no API key (weekly via Actions).

Setup

Python 3.13. requirements.txt pins exact versions that pass the tests; upgrade one deliberately, run python -m unittest discover -s tests, then commit.

pip install -r requirements.txt
python -m collectors.universe
python -m collectors.hourly --daily
streamlit run dashboard/app.py

Secrets (GitHub Actions repo secrets; locally copy .env.example to .env):

• COINALYZE_API_KEY - free at https://coinalyze.net/account/api (liquidations, aggregated OI)
• CLAUDE_CODE_OAUTH_TOKEN - for the weekly briefing workflow only; generate once from your Claude subscription with claude setup-token. Local briefings just use your Claude Code login.

See CLAUDE.md for schemas and commands, docs/ for system docs.

Reliability and development

See AGENTS.md and .agents/skills/crypto-dashboard/SKILL.md for portable project guidance. Current rankings and briefings share freshness/coverage gates; Research and Coin Detail are available in the sidebar. New observations carry ingestion provenance and options retain actual capture times. Existing historical data is not rewritten and legacy limitations remain visible.

Run python -m unittest discover -s tests -v for synthetic regression checks. Tests and analytics imports do not load local credentials. Operational collector entry points load their existing local environment configuration explicitly.