agentleFS
Tha-vivid-oneTha-vivid-one/claude-skills/progressive-summary/SKILL.md

progressive-summary

Skill0 starsChanged 7 months agoMITView on GitHub

Lay a non-destructive "scan-layer" onto ONE long note the user explicitly points you at: bold the keeper sentences and ==highlight== the few keepers-of-keepers, leaving the note otherwise completely untouched (AI-assisted progressive summarization — it only ADDS emphasis and never rewrites, summarizes, condenses, deletes, or restructures). This is a deliberate, opt-in move, NOT an ambient one. Trigger only when the user clearly asks to emphasize/mark up a specific note in place for faster rereading or skimming — e.g. "/progressive-summary", "progressive summarize this", "lay a scan-layer on this", "bold the keepers and highlight the essence", "mark up this transcript so I can reread it fast". Operate ONLY on the single note or section they name; never auto-run, never pick the target yourself, never touch other notes. Do NOT trigger merely because a long transcript or article is mentioned, and explicitly do NOT trigger for: summarize / condense / shorten / TL;DR (those rewrite), extracting content into a separate note, splitting a note into atomic notes, answering "what are the key takeaways" in chat, cleaning up or reformatting prose, or anything aimed at the user's own evolving journal / working / trade notes. When unsure, do NOT trigger — this is invoked deliberately, so under-triggering is the safe failure.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Don't over-mark: If everything is bold, nothing is.

    line 42
  2. Distinctive:

    On multi-section notes, target the body's wording: When the same idea appears in both a summary block and the raw prose, match the prose …

    line 64
  3. Distinctive:

    After the pass, ALWAYS surface the essence back in chat so the user gets value immediately:

    line 70
  4. Distinctive:

    Don't rewrite or "clean up" the prose: If you feel the urge to improve a sentence, stop — that's a different task and it breaks the one …

    line 83

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Progressive Summary

What this is, and why it works

Progressive summarization lays a scannable layer over a long note without changing a word of it. You bold the load-bearing sentences (Layer 2), then highlight the few that are the absolute essence (Layer 3). Future-you then reads the note at three depths:

  • highlights only → the ~10-second essence
  • highlights + bold → the full argument in ~30 seconds
  • drop into the raw prose → only where a section earns it

The reason this beats rewriting a note into a summary: the note stays whole. Nothing is lost, the author's voice and the surrounding context remain, and the full thing is always there to reread. You're adding a fast path, not paving over the road. (It's the deliberate opposite of breaking a note into atomic fragments — keep it intact.)

This is best on long source material the user wants to revisit — podcast/NotebookLM transcripts, saved articles, research dumps — especially before a second read-through.

Core rules — these are the skill, don't break them

  • Add only. Wrap text in **bold** and ==highlight==. Never reword, summarize, delete, reorder, or "fix" typos or links. The entire value is that it is still the original note, just easier to scan — it must be fully reversible, so a git checkout or undo restores the original exactly.
  • Only touch what you're pointed at. Mark up the specific file or section the user named — nothing else. Don't go looking for other notes, and never run on your own initiative.
  • Don't over-mark. If everything is bold, nothing is. Over-marking is the most common failure. Aim for roughly 10–20% of the body bolded, and of that, roughly 10–20% highlighted. When in doubt, mark less.
  • Leave existing distillations alone. Source notes often already carry a hand-written summary or an already-highlighted block (frequently at the top). Don't re-mark those — aim the layer at the raw body. If it's genuinely ambiguous which part to mark, ask before editing.

Procedure

  1. Read the target the user pointed you at. If they named a section ("the article part", "just section III"), scope to exactly that.
  2. Check for an existing distillation. Multi-section notes often stack a hand-written highlights block + a worksheet/template + the raw source. Mark up the raw source; skip what's already distilled, and say which part you worked on.
  3. Bold the keepers (Layer 2). Pick the sentences that carry the argument — the ones that, read alone in order, deliver the whole point. Bold the meaningful clause, not the throat-clearing around it. Spread them across the whole piece so scanning the bold traces the full arc, not just the opening.
  4. Highlight the keepers-of-keepers (Layer 3). From the bolded set, choose the 4–8 lines that are the irreducible essence — the thesis, the turn, the line you'd quote on a card. Wrap those in ==...==.
  5. Report the spine (see Report format below) so the user gets the payoff without reopening the file.

Choosing what counts as a "keeper" — the judgment that matters most

A keeper isn't merely an interesting sentence — it's one that does work in the argument: the claim or thesis, a definition, the turn ("but here's the real reason…"), an actionable instruction, or a memorable formulation. Skip setup, illustrative examples of a point you already bolded, transitions, and hedging.

The test: read only the bolded sentences, top to bottom — do you get the argument? If yes, you bolded the right ones. If you get a disconnected highlight reel, rebalance (usually you over-bolded the first section and under-bolded the rest).

A keeper-of-keepers is the line you'd put on a flashcard — the one that survives if everything else is forgotten. There should only be a handful per piece. If you're highlighting a dozen, you're really just bolding again — pull back.

Editing safely (you're making many small wraps in one file)

  • Match a unique substring for each wrap. On long notes short phrases repeat, so include enough surrounding words to be unambiguous.
  • On multi-section notes, target the body's wording. When the same idea appears in both a summary block and the raw prose, match the prose version (usually longer / phrased differently) so you mark the right occurrence and don't fail on a non-unique match.
  • Preserve exact characters. Copy curly quotes (' " "), em dashes (—), and the user's own [single-bracket] link convention verbatim — don't normalize them, or the match will miss and the text will drift.
  • Wrap at clean boundaries. Bold/highlight render across a soft line break inside one paragraph, but don't span a blank line (that's two paragraphs) — wrap each part separately instead.

Report format

After the pass, ALWAYS surface the essence back in chat so the user gets value immediately:

Scan-layer's on [note] — [N] bold keepers tracing the argument, [M] ==highlighted== as the essence. (Worked the [raw article body / section X]; left your existing [Highlights block] untouched.)

The keepers-of-keepers — your reread spine:

  1. ==…==
  2. ==…== … (4–8 total)

Reread path: scan the highlights → scan the bold → drop into full prose only where a section grabs you. Nothing deleted, fully reversible.

What NOT to do

  • Don't rewrite or "clean up" the prose. If you feel the urge to improve a sentence, stop — that's a different task and it breaks the one promise this skill makes.
  • Don't add a TL;DR, summary paragraph, or new headers unless asked. The whole bet is that emphasis-in-place beats a separate summary.
  • Don't touch frontmatter.
  • Don't mark up the user's own evolving working notes (journals, trade logs, drafts) unless they explicitly point you there — those are theirs to develop, not yours to annotate.

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.