agentleFS
megandmartinmegandmartin/agent-skills-repo/skills/productivity/reading-digest/SKILL.md

reading-digest

Skill4 starsChanged 4 months agoMITView on GitHub

Turns a backlog of saved links into a prioritized reading digest — relevance-ranked, three-sentence summaries, and a read/skim/skip verdict for each. Use when the user says "digest my reading list", "summarize these links", "what's worth reading here", "clear my saved tabs", or pastes a pile of URLs. Don't use for deep multi-source research on one question — use a research workflow — or for the daily schedule brief — use morning-brief.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Files
SKILL.md

Reading Digest

Takes the guilt pile — 15 saved links nobody will ever "get to" — and converts it into a ranked digest with honest verdicts: read this one properly, skim these three, skip the rest and here's why. The standard: every summary comes from actually fetched content, never from what the headline implies, and "skip" is a first-class recommendation, not a failure.

When to Use

  • User has a list of saved links (text file, bookmarks export, pasted URLs) and wants to know what deserves attention.
  • Weekly reading-list cleanout; prepping reading for a flight.
  • Not for: answering one research question across many sources (that's a research task, not a digest). Not for daily news/calendar — use morning-brief.

Quick Reference

ActionCommand / Call
Load the listread_file the links file, or take pasted URLs
Count + dedupepython3 -c "urls=[l.strip() for l in open('links.txt') if l.strip().startswith('http')]; print(len(urls), len(set(urls)))"
Fetch an articleweb_extract on the URL (the web toolset)
Fallback fetchcurl -sL --max-time 15 "URL" | head -c 4000
Verdict scaleRead (whole thing, soon) · Skim (intro + headers) · Skip (with reason)
Save the digestreading-digest-<YYYY-MM-DD>.md
Reading-time estimatepython3 -c "print(round({word_count}/230))" — minutes at ~230 wpm
Detect a paywall wrapperfetched text < ~400 chars or contains "subscribe to continue"

Procedure

  1. Precheck — get the link list and dedupe it (python3 one-liner). Ask one calibration question if you don't already know it: what is the user optimizing for right now (current projects, interests, deadlines)? Relevance ranking is meaningless without a "relevant to what".
  2. Fetch — pull each URL with web_extract (curl fallback). Record per link: title, source, length estimate, publish date if visible. Paywalled or dead links aren't skipped silently — they go in the digest marked "couldn't fetch" with whatever the metadata reveals.
  3. Summarize — for each fetched piece, exactly three sentences: (1) what it actually says/claims, (2) the strongest specific point, insight, or number in it, (3) who it's for / what you'd do with it. Summaries come from fetched text only — if you only got the intro, say so.
  4. Rank + verdict — order by relevance to the user's stated priorities, factoring freshness and substance-per-minute. Assign Read / Skim / Skip. Be ruthless and honest: a well-written but irrelevant piece is a Skip; a clumsy but load-bearing one is a Read. Every Skip gets a one-clause reason so the user can overrule.
  5. Deliver — top 3 "Read" items first with time estimates, then Skims, then Skips as a compact list. Close with the total: "reading the Reads costs ~{N} min; you just saved ~{M} min of backlog guilt." Save the file. Offer (don't do unprompted) to clear the processed links from the source list — never delete without an explicit yes.

Output Template

# Reading Digest — {date} · {n} links → {r} read / {s} skim / {k} skip
Ranked for: {user's stated priorities}

## 📖 Read ({r}) — ~{N} min total
### 1. {Title} — {source} ({est. min})
{Three-sentence summary.}
Why it ranks: {one line tying it to the user's priorities}

## 👀 Skim ({s})
- **{Title}** — {source}: {three-sentence summary}. Skim: {which part}.

## ⏭️ Skip ({k})
- {Title} — {reason: outdated / duplicate of #1 / listicle padding / not your problem}

## ⚠️ Couldn't fetch
- {URL} — {paywall / 404} — {what the title suggests, labeled as guess}

Pitfalls

  • Summarizing from the headline — the cardinal sin; headlines routinely misrepresent the piece. Recovery: no fetched text, no summary — the link goes to "Couldn't fetch" with its guess clearly labeled, never into the ranked list.
  • Everything ranked "Read" — a digest that recommends everything recommends nothing. Recovery: cap Reads at 3 (5 for lists over 25 links); force-rank; remember Skip-with-reason is the digest's main gift.
  • Paywall page summarized as the article — you fetched a subscription pitch and summarized cookie-consent text. Recovery: sniff for paywall markers (very short text, "subscribe to continue"); mark as couldn't-fetch rather than summarizing the wrapper.
  • Ranking without calibration — relevance to nothing in particular produces a generic list. Recovery: if step 1's priorities question went unanswered, rank by substance + freshness and say explicitly that ranking is uncalibrated.
  • Deleting the user's saved links after processing — never unprompted. Recovery: offer removal, act only on explicit confirmation, and only ever remove links that appear in the digest.

Verification

  • Every summarized link was actually fetched; unfetchable links are in their own labeled section
  • Each summary is exactly three sentences and contains at least one specific detail from the text
  • Reads capped and force-ranked; every Skip has a reason
  • Dedupe count reported (n links, m unique)
  • Source list untouched unless the user explicitly approved cleanup
  • Time estimates present on every Read item and the digest total

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.