agentsclimarketplace

Loot

Skill RakhimovY/loot/skills/loot

Use when someone pastes a link (X/Twitter post or thread, article, blog, GitHub repo, Reddit or Hacker News thread, YouTube video) and wants the practical, actionable takeaways, what is worth taking and what to skip, with a before/after case for each. Triggers include /loot, "loot this", "what's worth taking", "break this down", "what should I take from this" followed by a URL. Optional lens preset (maker, engineer, marketer, researcher).From its SKILL.md

Install
npx -y skills add RakhimovY/loot --skill loot

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

8.0 KB, ~1.8k tokens by cl100k_base, as published. Nobody here has run it

loot

Overview

Paste a link, get a short, opinionated read: what is worth taking (with a before/after case for each) and what to skip. No summary, no retelling, just a decision.

One job: kill the noise and hand back something you can act on.

When to use

  • Someone pastes a link plus a short ask: "loot this", "what's worth taking", "break this down", /loot <url>.
  • Any link type in the fetch table below.

When NOT to use: they want a full, faithful summary of the source, or a deep archived write-up. This skill deliberately throws most of the source away.

How to get the content (escalate, do not give up)

Your job is to actually read the source. Use the best tool you have, and if you lack the capability, recommend the tool the user should add, do not just refuse. A flat "paste it yourself" is the last resort, not the first answer.

1. Static pages (article, blog, docs, GitHub repo, Reddit, Hacker News): fetch the URL with your agent's web-fetch tool. (Claude Code calls it WebFetch; other agents have an equivalent.) For a GitHub repo, also pull the raw README and note what it does, stars, and recent activity; use the gh CLI if available. This works out of the box.

2. JS-heavy or login-walled pages (X/Twitter, LinkedIn, Instagram, Threads): a plain fetch returns an empty shell, so this needs a real browser.

  • If you have a browser-automation tool or MCP connected, use it. One running with the user's logged-in session can read anything the user can see (needed for LinkedIn and most of X).
  • If you do not have one, do NOT stop at "paste it". Tell the user a browser tool is needed and recommend adding one: for example Playwright MCP (commonly run as npx @playwright/mcp@latest) for public pages, or a Chrome/browser MCP that reuses their logged-in session for login-walled sites. Then offer the paste fallback so they are unblocked right now.

3. Video (YouTube and others): there is usually no built-in transcript.

  • If you have a transcript tool or MCP, use it.
  • If not, recommend installing one: for example yt-dlp (it can pull a video's captions/subtitles) or a YouTube-transcript MCP. Then offer to continue the moment they paste a transcript.

Whatever path you take: if you could not actually read the source, say so plainly and never invent what it said. Recommending the right tool is part of the job; a refusal is not.

The lens (what separates signal from noise)

Everything is filtered through one lens. Default is maker. The user can switch it (--lens engineer, "as an engineer", etc.) or pass a custom one-line lens.

An item passes ONLY if both are true:

  1. it changes how you do the thing the lens cares about;
  2. you can actually act on it.

Personal stories, lifestyle, motivation, generic platitudes, hype, and tactics tied to a platform you do not use: all noise. There is no fixed cap on the number of items, the bar above is what keeps it short, not a counter.

LensAn item counts if it changes how you...Before/after cases are framed as...
maker (default)build OR sell your own producttime, money, users
engineerdesign, write, test, or operate softwarebugs, maintenance, performance, complexity, on-call pain
marketerreach, convert, or retain an audiencereach, conversion, trust, churn
researcherunderstand a topic well enough to actunderstanding, better decisions, dead-ends avoided

Custom lens: if the user passes a one-line lens, use it the same way (keep what changes how they do that thing; frame cases as consequences for it).

Output contract (exactly this, in this order)

Respond in the reader's language. Default to English; if the user wrote in another language, answer in that language.

1. Reality-check line (optional). One line at the top, ONLY if there is something to ground: small numbers, obvious flexing, unverified claims. If the source is solid, skip this line.

2. What's worth taking , a numbered list. Each item has exactly three parts:

N. **Short bold title**
One line: what they did.
- Without it: what happens to your work if you don't do this (the lens's consequence domain).
- With it: what happens if you do.

Every item must have both "Without it" and "With it". No case, no item.

3. What to skip , 2 to 4 lines. Only two things: (a) risky or against-the-rules tactics, as a warning; (b) things that look useful but do not apply to the reader (wrong platform, wrong stack). Drop pure jokes and lifestyle silently, do not list them.

4. In one line: a compressed plan, steps joined by arrows: step -> step -> step.

What usually counts as signal (priors, not a template)

These help recognize signal for the default maker lens. Take only what is actually in the source, never pad.

  • copy what already sells (proven demand), do not invent it;
  • aim at a strong human desire (money, health, status, fear of falling behind);
  • measure everything, pour effort only where money already lands;
  • kill dead projects faster, run more cheap attempts;
  • early signal means sales in the first days, not applause;
  • charge from day one, keep value behind payment;
  • strip friction from the first steps, push completion up;
  • notify only the people who actually engage;
  • test the price and the payment screen, not the button color.

For other lenses, the priors shift (engineer: simpler designs, fewer moving parts, tests that catch real bugs, observability; marketer: one channel done well, a sharp offer, proof over claims; researcher: primary sources, reproducible methods, falsifiable claims). Use them as a filter, not a checklist.

Example (copy the form, not the content)

This is what the final answer looks like, maker lens, source is a viral "how I hit $1k/mo" thread. No leading > markers, no "Example" heading.

Reality check: $1k/mo is small, and half the thread is flexing (gym, supplements, diet). Filter hard.

**What's worth taking**

1. **Copy what already sells**
He didn't invent an idea, he took a product that already makes money for others.
- Without it: you spend months building something nobody asked for, hoping they'll want it.
- With it: demand is already proven, so you only fight execution.

2. **Pour effort only where money already lands**
He exported all sales, found which countries and keywords actually pay, and spent more there.
- Without it: you spread time and budget evenly, including onto dead things.
- With it: you find the two or three channels that work and feed only those.

3. **Charge from day one**
The core sits behind a hard paywall, nothing valuable is free.
- Without it: people use it for free and never pay.
- With it: whoever wants the value pays, and you see real demand.

**What to skip**
- Bidding on competitors' names and asking for a review during onboarding (the store "patched" his trick): on the edge of the rules.
- Apple Ads, RevenueCat, SwiftUI: iOS-app stack only, not for you if you build on the web.

**In one line:** copy what already sells -> launch cheap -> watch where money shows up -> feed only that -> charge immediately.

Common mistakes

  • An item without "Without it / With it". Every item needs both cases.
  • Items labeled by source numbers (#1, #5) instead of a title plus a one-line "what they did".
  • Extra sections (meta-pattern, red flags, a separate "ignore" dump). Keep the 4 contract parts only; warnings go inside "What to skip".
  • Retelling the whole source instead of selecting. Take only what passes the lens.
  • Inventing what the source said when a fetch failed. If you couldn't read it, say so and ask for a paste.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,149. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.