How to use the Multi-Source Research Skill agent skill

What is an agent skill, and what does this one make the agent do?

An agent skill is a folder of instructions and supporting files that an AI agent loads when a request matches the skill's description. You do not run the skill yourself; you ask for something in ordinary language and the agent decides whether the skill applies.

Without this skill, an agent researching a topic tends to hallucinate URLs and answer from the first few search results. With it installed, the agent runs several vendors in parallel, cross-checks their findings, verifies each URL before delivering it, and routes opinion questions to community sources first. The skill offers four depths: Quick (one Perplexity call, about 10-15 seconds), Standard (four vendors cross-checked, about 30-60 seconds), Extensive (seven explorers and two independent verifiers, about 60-90 seconds) and Deep Investigation (progressive iteration with a persistent vault, roughly 3-60 minutes).

• Every delivered URL is fetched and checked against the claim it supports; unverifiable links are dropped.
• Findings are tagged [HIGH], [MED], [LOW] or [CONFLICT].
• Disagreements between sources are shown with both sides rather than averaged away.
• Sentiment questions are answered from community sources first, with verbatim quotes and thread URLs.

Install the skill from the download ZIP

Download the ZIP for the Research skill from this page and unzip it. The skill folder must be named Research and contain the SKILL.md file at its top level alongside the Intents folder and the other files listed on this page.

For a personal install that works in every project, place the unzipped Research folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder instead.

If you use the Claude apps rather than a local setup, you can upload the skill instead. Open Settings, go to Capabilities, then Skills, and upload it there.

• Personal (all projects): ~/.claude/skills/Research
• Single project: .claude/skills/Research inside the project folder
• Claude apps: Settings → Capabilities → Skills

Trigger the skill with natural requests

You do not need a special command. The skill's description lists the phrases that route to it, so ordinary requests work. Saying "research" alone runs the Standard tier by default. Saying "quick research" runs the single-call Quick tier, and "extensive research" or "deep research" runs the Extensive tier. "Deep investigation" or "investigate [topic]" starts the Deep Investigation tier.

The skill also responds to find-and-identify requests such as "find this product", "what is this", "track down" or "source this", and to everyday look-ups such as "best X in a place", "what's good in <city>", "is X open" or "hours". For those, the skill is meant to be used instead of a raw web search.

• "Do research on the best noise-cancelling headphones under current reviews." (Standard)
• "Quick research: what are the opening hours of the central library?" (Quick)
• "Extensive research on how small bakeries handle wholesale pricing." (Extensive)
• "Investigate the electric scooter rental market." (Deep Investigation)

Choose the right depth tier

Quick uses a single vendor and a single query. Standard cross-checks four vendors as parallel tool calls in the session. Extensive and Deep Investigation use all five vendors. The skill's own reasoning is that the more a question is worth, the more independent model training distributions check it.

Deep Investigation builds a persistent vault on disk and deepens it over several iterations, so it suits ongoing topics rather than a single question.

• Quick: one vendor, one query, a few seconds to about fifteen.
• Standard: four vendors, the default for "research".
• Extensive: all five vendors, seven explorers and two verifiers.
• Deep Investigation: progressive iteration into a vault; slowest.

Requirements and what each feature needs

The skill itself needs no paid key for everything. ClaudeResearch uses Claude's built-in WebSearch only and needs no API keys. The higher tiers call vendor APIs for Perplexity, Gemini, OpenAI/Codex and Grok, which means you need access to those vendors. The source excerpt does not specify how keys or accounts are supplied, so check the setup guide included with the skill (Setup.md) before running those tiers.

Some features need extra tools. The URL verifier (Tools/VerifyUrls.ts) and its tests need the Bun runtime. YouTube extraction and Fabric patterns need the Fabric CLI. The DeepVerifiedResearch script needs a Node.js-compatible environment.

• Claude WebSearch only: no API keys needed.
• Standard, Extensive, Deep tiers: vendor API access (Perplexity, Gemini, OpenAI/Codex, Grok at higher tiers).
• URL verifier and its tests: Bun runtime.
• YouTube extraction and Fabric patterns: Fabric CLI.
• DeepVerifiedResearch: Node.js-compatible environment.

Safety notes: scripts and network access

This skill runs scripts and uses the network. Its verifier fetches the pages it cites, and its research tiers call paid vendor APIs. Review the Intents folder and the scripts before installing, especially Intents/DeepVerifiedResearch.mjs and the verifier in Tools/VerifyUrls.ts.

The skill is not meant for every task. It should not be used for background on a private person or entity, which the skill says belongs to a different OSINT skill. Papers go to ArXiv, wisdom extraction goes to ExtractWisdom, and raw page fetching or crawling is routed to BrightData.

• Runs scripts: yes.
• Uses network: yes, including paid vendor API calls at higher tiers.
• Not for entity background, papers, JSON entity extraction, wisdom extraction, or raw crawling.

Tips for getting reliable answers

Ask for the depth you actually need. A simple factual question usually does not justify a multi-vendor run, and the deeper tiers take minutes. When you need opinion or lived experience, say so, because the skill then starts with community sources and keeps the criticism next to the praise.

Read the confidence tags before acting on a finding. A [CONFLICT] tag means the sources disagree and both sides are shown, which is often the most useful part of the answer.

• Match the tier to the question's importance.
• For venue or hours questions, expect checks against the official page of the venue or its host property.
• Treat [LOW] and [CONFLICT] findings as leads to confirm, not as settled facts.