arjunprabhulal/agent-skills / skills/research/feed-digest
feed-digest
VerifiedReads a feed of new material — Hacker News, RSS, blogs, podcasts, subreddits, and returns a short digest of what is worth the user's attention. Use this whenever the user mentions catching up, what's new, a daily or weekly reading round-up, their feeds, newsletters, a podcast backlog, or asks what they missed, even if they don't use the word "digest". For condensing one specific document they already have in hand, use summarization instead.
Daily digest
The value of a digest is what it leaves out. A summary of everything is the feed again, longer. Your job is to read more than the user can and hand back less.
Default output: five items or fewer, each two sentences, ordered by what the user would regret missing. If nothing clears the bar, say the day was quiet — a short honest digest beats a padded one, and padding trains the user to stop reading these.
1. Resolve the source
Ask which source only if the request is ambiguous. Then read the matching adapter — it holds the fetch method, the ranking signal, and the source's specific traps:
| Source | Adapter |
|---|---|
| Hacker News | references/hackernews.md |
| RSS / Atom feeds | references/rss.md |
| Named blogs | references/blogs.md |
| Podcasts | references/podcasts.md |
| Subreddits | references/reddit.md |
If the user has a digest.config.md in the project or home directory, read it first — it holds
their standing sources, topics, and mute list, and it overrides the defaults here.
Done when: you have a concrete list of source URLs or IDs and a time window.
2. Set the window
Default to since-last-run if you know it, otherwise 24 hours for daily and 7 days for weekly. State the window in the output — a digest without a window is unreadable next week.
Done when: the window is fixed and stated.
3. Filter before reading
Fetching everything and summarizing everything is slow and produces mush. Cut hard first, using the adapter's ranking signal, then read only survivors in full.
Drop on sight:
- Anything matching the user's mute list
- Announcements with no substance behind them — funding rounds, rebrands, version bumps with no behaviour change
- Rehashes of something in a previous digest, unless there is genuinely new information
- Pure opinion with no new fact, unless the opinion itself is the news
Keep, even if unpopular in the feed: things that change what the user should do, and things that contradict something they currently believe.
Done when: you have at most ~15 candidates to read properly.
4. Read the survivors
Read the actual item, not the title and comments. Titles on aggregators are frequently wrong about their own content, and the comment section is a different artifact from the article.
For each survivor, get: the claim, the evidence behind it, and whether anything is being sold.
Done when: you could answer "so what?" for each candidate without re-opening it.
5. Write the digest
## Digest — <source>, <window>
**<Headline stating the finding, not the title>**
What it says, and why it matters to you specifically. [link]
...
_Also: <one line for near-misses, if any>_
_Quiet on: <sources checked with nothing worth reporting>_
Rules that keep it useful:
- Lead with the finding, not the title. "Postgres 18 makes UUID keys viable as primary keys" is worth reading; "Postgres 18 released" is not.
- Say why it matters to this user. If you cannot, it is a weak item — cut it.
- Separate claim from evidence. "Benchmarks in the post show 40% — single machine, author works for the vendor" is the honest form.
- Link every item. A digest the user cannot verify is a rumour.
- Never summarize something you failed to fetch. Say it failed and link it. This is the one rule with no exceptions — an invented summary poisons the whole digest's credibility.
Done when: every item has a finding, a reason it matters, and a working link.
6. Note what to carry forward
End with anything the user should watch next time — a thread still developing, a release expected soon. This is what makes a recurring digest compound instead of resetting daily.
Running it on a schedule
This skill pairs with a scheduled run. When invoked on a schedule, skip the clarifying question
in step 1 entirely and use digest.config.md; a scheduled digest that stops to ask a question
produces nothing.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |