Guide · Agent skill
How to use the Pareto Topic Sweep agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Pareto is an agent skill, a folder of instructions and scripts that an AI agent loads when it is asked for a specific kind of work. This guide explains what it does, how to install it, how to ask for a sweep in plain language, and what it needs to run. It is written for people who have not used agent skills before.
Quick start
- 1Download and unzip. Download the Pareto ZIP from the page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder.
- 2Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the ZIP.
- 3Add your keys. Put YOUTUBE_API_KEY and PERPLEXITY_API_KEY in your LifeOS .env file and make sure gh is authenticated.
- 4Check your setup. Confirm the GitHub CLI is logged in before running a full sweep, since it is the one source with no key in the .env file.
- 5Ask for a sweep. Use a phrase such as "pareto" or "what's the zeitgeist on" followed by a topic written as a category.
- 6Review the report. Open the saved report in ~/Downloads/, check that each frontier claim links to a source, and review any repo before opening it.
What the Pareto skill makes the agent do
When you ask about a topic, Pareto has the agent search YouTube, GitHub and the web, then filter the results down to the most relevant few by relevance, reach, recency and how unusual they are. It then writes a report covering how thinking on the topic used to be, what the mainstream thinks now, and where forward-thinking people seem to be heading, with a link for each claim.
Sources that fail or hit quota limits are noted in the result rather than stopping the run.
- Past meta: how thinking on the topic used to be, with links and years.
- Current mainstream meta: what most people are saying now.
- Where the forward thinkers are heading: people or projects ahead of the mainstream, each linked.
- A list of any source that failed.
Install the Pareto skill
You can install the skill in one of two ways. Download the ZIP from the Pareto page, then either unzip it into your personal skills folder or into a project's skills folder.
For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Per project: .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Set up the sources the skill needs
Pareto reads its keys from your LifeOS .env file. YouTube uses the YouTube Data API v3 key stored as YOUTUBE_API_KEY. Web search uses PERPLEXITY_API_KEY. GitHub uses the GitHub CLI (gh), which must be authenticated.
The skill's setup notes say each source is independent. If one key is missing or a quota is spent, that source is skipped and named in the result. You can run on GitHub alone with just the GitHub CLI, but a full sweep needs all three.
- Bun runtime, to run the TypeScript tools
- GitHub CLI (gh), authenticated
- YouTube Data API v3 key in YOUTUBE_API_KEY
- Perplexity API key in PERPLEXITY_API_KEY
- LifeOS .env file and LifeOS shared tooling (Env.ts, Paths.ts) at the expected relative paths
gh auth statusTrigger a sweep with a natural request
You do not need special syntax. Name the topic and use a phrase the skill recognises, such as "pareto", "what's the zeitgeist on", "what's the meta on", "who are the forward thinkers on" or "top content on a topic".
Phrase the topic as a category. For example, use "3D modeling software" rather than "best 3D modeling system", because words like "best" and "system" match almost everything.
- "Pareto agentic coding"
- "What's the zeitgeist on personal AI infrastructure?"
- "Who are the forward thinkers on local LLM inference?"
- "Run pareto on 3D modeling software"
Read the report and the console summary
The full report is saved to ~/Downloads/ with a name that includes the topic and the date. The console summary is five lines or fewer: the shift in one sentence, the two or three forward thinkers worth following, and any source that failed.
The frontier section is meant to name people or projects that are not simply the most-viewed items. If it only repeats the biggest videos, the skill's own checks treat that as a failed run.
- Report location: ~/Downloads/pareto-<topic-slug>-<YYYY-MM-DD>.md
- Each frontier claim links to its source
- Dates come from the source page, not from when it was crawled
Safety notes: scripts and network access
Pareto runs scripts and uses the network. It calls YouTube, GitHub and Perplexity, and runs TypeScript tools with Bun. Review the skill files before you install it if that matters for your setup.
GitHub searches for software categories can return cracked-software lure repositories with names such as fake activators. These look like malware bait. The skill's ranking drops repos with lure-style names or under 10 stars, but you should still review any linked repo before opening it.
- Review any GitHub repo before you open or download it.
- Do not run software from links in a report without checking it first.
- YouTube transcripts come from the Gemini watch pass on the public URL. A video that is refused is named as unread in the report.
What people use it for
- Scanning a fast-moving field, such as agentic coding, before a deeper research dive.
- Finding a short list of people to follow on a topic, with links to their work.
- Seeing how the mainstream view on a topic has changed over time, with dates from each source.
- Repeating the same topic sweep regularly to compare what is current.
Frequently asked questions
How do I install the Pareto agent skill?
Download the ZIP from the Pareto page and unzip it into ~/.claude/skills/ for a personal install, or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
What does Pareto require to run?
It needs the Bun runtime, an authenticated GitHub CLI (gh), a YouTube Data API v3 key in YOUTUBE_API_KEY, a Perplexity API key in PERPLEXITY_API_KEY, and a LifeOS .env file with the shared tooling at the expected paths. It also runs scripts and uses the network.
Is the Pareto skill free?
The source material for this page does not state pricing or licence terms beyond the licence being MIT. The YouTube API has a daily quota of 10,000 units by default, and each YouTube search costs 100 units, so check your own API usage.
Can I use Pareto without a YouTube or Perplexity key?
Yes. Each source is independent. If a key is missing or a quota is spent, that source is skipped and named in the result. You can run on GitHub alone with just the GitHub CLI, though a full sweep needs all three sources.
When should I use Research instead of Pareto?
Use the Research skill for a specific factual question. Pareto is for mapping a topic's current meta and who is ahead of the curve, so the skill's own description points factual questions to Research.
Why did my GitHub results include suspicious repos?
Software-category searches on GitHub can return cracked-software lure repos. The skill drops repos with lure-style names or under 10 stars from the ranked set, but you should review any linked repo before opening it.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Pareto suits people who want a quick, linked overview of what is current on a topic and who to follow, especially researchers scanning a fast-moving field who already have YouTube, GitHub and Perplexity access set up. It is not the right tool for a single factual question, for analysing one article or video in depth, or for AI industry news digests, where Research and other skills fit better.
Download the skill