P
podanypepa/skim
Personal RSS agent: Claude Code skills score your feeds against your interests and write a short daily digest.
更新 2026-11-18
在 GitHub 查看 ↗直链 · 无中转页
装完之后怎么算装好了
- 装完该看到什么 — 命令正常结束时退出码为 0,终端里没有 "command not found" 或 "permission denied"。
- 一行自检 — 按该仓库 README 里给出的最小用法跑一次;能按 README 的描述输出就是通了。
- 答不出来怎么办 — 这个项目没有写排错说明,去仓库的 Issues 搜 install。
安装命令取自该仓库;以上判据只描述可观察到的结果,不替代项目自己的文档。
- ★ 数取值
- 0 ★ · podanypepa/skim · 取自 GitHub 搜索 API · (2026-11-19 20:00:48 核对)
- 来源明细
- 从该仓库的 GitHub 页面直接读取 · 条目类型:开源项目
- 收录与活跃
- 收录于 2026-11-19 · 最近见到 2026-11-20 · 共见到 2 次 · 被 1 个来源列出
- 评分构成
- 综合分 103 · 1 个来源 · repo
- 安装
- 安装命令见上方「安装与配置」。
常见问题(答案都来自我们的记录)
它适合谁 / 用来做什么?
我们把它归在这些标签下: AI Agent。标签来自条目自身,不是我们强加的分类。
谁维护它、还在更新吗?
仓库
podanypepa/skim;我们收录于 2026-11-19;最近见到 2026-11-20。这里没有"官方认证"或"质量评级"——我们只记录来源与时间。README 原文(英文,未翻译) · 在 GitHub 读完整 README →
# skim
以下摘录自该仓库的 README(片段)
A personal RSS agent that reads the feeds for you and hands you a short daily digest.
skim is a replacement for scrolling through hundreds of headlines in a
feed reader. You tell it what you care about, it follows a small set of
hand-picked sources, scores every new article against your interests,
groups duplicate coverage into stories and writes a concise daily digest
with correct links back to the originals.
It is built as a set of Claude Code
skills (the judgement: what is relevant, what is one story, how to
summarise it) on top of small, deterministic Go tools (the mechanics:
fetching, parsing, validating, storing, rendering).
> skim
Fetched 151 · triaged 151 · 36 stories · 0 feeds failed
## ai-novinky
1. Claude Fable 5.2 is the top model again — 20 % cheaper per token, but hungrier …
2. Claude Code leaks secrets to GitHub — "don't ask again" stores commands verbatim …
## golang
21. Go turns 17: the year in review and a look at 1.28 …
## python
28. Tim Peters has died …
Feedback? e.g. "22 good, 26 not interesting".
## Why
Feed readers show you everything and leave the filtering to you. skim
flips that around:
You describe interests once (topics.md) — what you want, what you
don't, how deep, how much per day.
The agent does the reading — every article gets a 1–5 score and a
one-sentence reason, so mistakes are visible and easy to correct.
You read one digest a day — stories, not headlines; the same news
from five outlets becomes one entry with five sources.
The split between skills and Go tools is deliberate: language models are
good at judgement and bad at bookkeeping, so every step where a mistake
would be silent (duplicates, broken links, half-written files) is handled
by code with tests, and the model only ever produces content that the
code validates before anything is written.
## How it works
feeds.yaml topics.md topics.md
│ │ │
▼ ▼ ▼
┌────────────┐ ┌──────────────┐ ┌──────────────┐
│ skimfetch │────▶│ skim-triage │───────▶│ skim-digest │◀── your feedback
│ (Go) │ │ + skimtriage │ │ + skimdigest │
└─────┬──────┘ └──────┬───────┘ └──────┬───────┘
▼ ▼ ▼
data/articles/ data/triage/ data/digest/D.md
D.jsonl D.jsonl (+ D.json)
1. Fetch — skimfetch downloads every feed in feeds.yaml
concurrently, uses conditional requests (ETag / Last-Modified), parses
RSS 2.0, RSS 1.0 and Atom, deduplicates by GUID / link, converts HTML
to plain text and appends new articles to a daily JSONL file. A broken
feed never stops the run.
2. Triage — the skim-triage skill reads topics.md and scores each
new article 1–5 with a short reason. skimtriage pending hands out
batches of untriaged articles; skimtriage save validates the batch
(known ids, integer scores, reason length…) and writes it all or
nothing.
3. Full text — many feeds only carry a teaser (or nothing). Before
the digest, skimdigest fulltext fetches the article page for
candidates with short content, extracts the readable text and stores
it, so summaries have real substance. It honours robots.txt (RFC 9309,
including every redirect hop), only connects to public addresses, and
tries each page once — transient failures (network, 429, 5xx) are
retried on the next run. Sites that block bots (e.g. Reddit) keep their
skim is a replacement for scrolling through hundreds of headlines in a
feed reader. You tell it what you care about, it follows a small set of
hand-picked sources, scores every new article against your interests,
groups duplicate coverage into stories and writes a concise daily digest
with correct links back to the originals.
It is built as a set of Claude Code
skills (the judgement: what is relevant, what is one story, how to
summarise it) on top of small, deterministic Go tools (the mechanics:
fetching, parsing, validating, storing, rendering).
> skim
Fetched 151 · triaged 151 · 36 stories · 0 feeds failed
## ai-novinky
1. Claude Fable 5.2 is the top model again — 20 % cheaper per token, but hungrier …
2. Claude Code leaks secrets to GitHub — "don't ask again" stores commands verbatim …
## golang
21. Go turns 17: the year in review and a look at 1.28 …
## python
28. Tim Peters has died …
Feedback? e.g. "22 good, 26 not interesting".
## Why
Feed readers show you everything and leave the filtering to you. skim
flips that around:
You describe interests once (topics.md) — what you want, what you
don't, how deep, how much per day.
The agent does the reading — every article gets a 1–5 score and a
one-sentence reason, so mistakes are visible and easy to correct.
You read one digest a day — stories, not headlines; the same news
from five outlets becomes one entry with five sources.
The split between skills and Go tools is deliberate: language models are
good at judgement and bad at bookkeeping, so every step where a mistake
would be silent (duplicates, broken links, half-written files) is handled
by code with tests, and the model only ever produces content that the
code validates before anything is written.
## How it works
feeds.yaml topics.md topics.md
│ │ │
▼ ▼ ▼
┌────────────┐ ┌──────────────┐ ┌──────────────┐
│ skimfetch │────▶│ skim-triage │───────▶│ skim-digest │◀── your feedback
│ (Go) │ │ + skimtriage │ │ + skimdigest │
└─────┬──────┘ └──────┬───────┘ └──────┬───────┘
▼ ▼ ▼
data/articles/ data/triage/ data/digest/D.md
D.jsonl D.jsonl (+ D.json)
1. Fetch — skimfetch downloads every feed in feeds.yaml
concurrently, uses conditional requests (ETag / Last-Modified), parses
RSS 2.0, RSS 1.0 and Atom, deduplicates by GUID / link, converts HTML
to plain text and appends new articles to a daily JSONL file. A broken
feed never stops the run.
2. Triage — the skim-triage skill reads topics.md and scores each
new article 1–5 with a short reason. skimtriage pending hands out
batches of untriaged articles; skimtriage save validates the batch
(known ids, integer scores, reason length…) and writes it all or
nothing.
3. Full text — many feeds only carry a teaser (or nothing). Before
the digest, skimdigest fulltext fetches the article page for
candidates with short content, extracts the readable text and stores
it, so summaries have real substance. It honours robots.txt (RFC 9309,
including every redirect hop), only connects to public addresses, and
tries each page once — transient failures (network, 429, 5xx) are
retried on the next run. Sites that block bots (e.g. Reddit) keep their
收录徽章(可选)
想把收录标记放进 README 就复制这行,可选,不影响收录。
[](https://anclum.com/zh/s/podanypepa-skim/)
- 来源
- github · https://github.com/podanypepa/skim
- 按领域浏览
- AI Agent
- 收录于
- 今天
更多条目
所属领域与标签
每个标签都能点进去看同类的全部条目。数字是当前库里的真实条数。