一个面向 AI 信息跟踪的周报生成 CLI / Skill 原型。它把 RSS、Atom、sitemap 和手工事实源聚合到统一事件模型中,再通过规则筛选、去重、排序和模板渲染,生成可直接复制发布的 Markdown 周报。
- 支持
rss_atom、sitemap、manual_source三类信息源。 - 支持创作者源、官方源、趋势源的分层配置。
- 内置时间窗口过滤、事件聚类去重、主题/域名限额。
- 支持来源健康度降权,降低低可用源对主条选择的影响。
- 输出固定结构 Markdown:标题、编号主条、快讯、外显链接。
- 可通过 GitHub Actions 定时运行,也可本地手动生成。
./weekly-report run \
--end-date 2026-04-23 \
--window-days 7 \
--tz Asia/Shanghai \
--config config/weekly-report.json \
--output outputs/weekly-report-2026-04-23.mdMain configuration lives in config/weekly-report.json:
sources.creators[]: creator and analyst feedssources.official[]: official sources for fact anchorssources.trend[]: trend and quick-news candidatesmanual_items[]: manually curated factsrules.*: selection, deduplication, source-health and balance rulesrender.*: output language, style, title and display options
The workflow in .github/workflows/weekly-report.yml can run on a schedule or manually through workflow_dispatch. Generated reports are uploaded as workflow artifacts instead of being committed to the repository.
python3 -m unittest tests/test_weekly_report.pyThis project currently uses only Python standard library modules. Historical generated reports are intentionally excluded from the public repository through .gitignore.
MIT