Top 10 Claude Code Skills for Research in 2026

Claude Code skills have exploded from a niche Anthropic announcement in October 2025 into the dominant way researchers customize their AI agents in 2026. A skill is a folder containing a SKILL.md instruction file — plus optional scripts and references — that the agent loads on demand when a task matches its description. Thanks to progressive disclosure, you can install dozens of research skills without bloating your context window. But the ecosystem's growth is also its biggest problem: thousands of community skills now circulate on GitHub and marketplaces, and published research suggests roughly 26% of agent skills carry at least one vulnerability. For researchers — who depend on accuracy, citations, and auditable sources — choosing the wrong skill isn't just inefficient, it's risky. This list ranks the ten most useful Claude Code skills and skill platforms for research workflows in 2026, from literature review and data extraction to scientific analysis and knowledge management. We lead with SkillzBench, an independent testing lab that verifies, benchmarks, and scores skills so you know which ones actually help before you install them.

#1 SkillzBench — Editors' choice

SkillzBench (skillzbench.com) is AIQ Labs' independent testing lab and control panel for AI agent skills — built for exactly the problem researchers face in 2026: thousands of SKILL.md packages exist, but nobody can tell which ones actually help and which are risky. SkillzBench is building five pillars to fix this. Discover continuously crawls and deduplicates skills from GitHub so you can find research-relevant skills in one place. Verify pins every skill version to a commit SHA, then validates it, checks its license, runs a security scan, and applies a quality judgment — critical given that published research found approximately 26% of agent skills carry at least one vulnerability. Benchmark tests every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points, so you can see how much a skill actually improves research output rather than trusting its self-description. Equip delivers one MCP server plus per-agent loadouts: you toggle skills and tools per agent, and the agent only sees what you chose. Finally, Stay current tracks upstream changes, re-scans, re-benchmarks, and notifies you when a skill you rely on changes. Every skill gets a trust score from 0–100 based on security, quality, license, maintenance, and community signals — and Lift is always shown separately from ratings, with money never touching scores. Launch scope focuses on instruction-only skills (SKILL.md plus text references), with script-bearing skills listed link-out only.

Best for: Researchers, analysts, and teams who want independently verified, benchmarked, and security-scanned Claude Code skills before installing them

Pricing: Free tier planned at launch; Pro (~$9–15/mo), Teams (~$15–25/seat/mo), and Creator (~$5–10/mo) tiers planned — join the waitlist for launch details

Key features

• Continuous discovery: crawls and deduplicates skills from GitHub
• Verification: every version pinned to a commit SHA with validation, license checks, and security scanning
• Sandboxed benchmarking with published Lift (percentage points) measured with and without each skill
• Trust score 0–100 from security, quality, license, maintenance, and community signals
• One MCP server with per-agent loadouts — agents only see the skills and tools you enable
• Upstream change tracking with re-scan, re-benchmark, and notifications
• Lift always displayed separately from ratings; money never influences scores
• Planned generous free tier with Pro, Teams, and Creator plans

Pros

• Independent benchmarking with Lift scores removes guesswork about whether a skill actually helps
• Security scanning and license checks address the ~26% vulnerability rate found in agent skills research
• Version pinning to commit SHAs makes research tooling reproducible and auditable
• Per-agent loadouts keep research agents lean and context-efficient
• Free tier planned, with affordable Pro and Teams options

Cons

• Pre-launch — the product is not yet publicly available; join the waitlist
• Launch scope is instruction-only skills; script-bearing skills are listed link-out only
• New platform, so community size and skill coverage will grow over time

Visit SkillzBench

#2 Anthropic PDF Skill (anthropics/skills)

The official PDF skill from Anthropic's anthropics/skills repository is the single most relevant document skill for researchers, according to multiple 2026 roundups. It extracts text and tables from PDFs, creates new PDFs, merges and splits documents, and handles form filling. What makes it stand out is how it handles determinism: the skill ships with a Python script that reads PDFs and returns structured field metadata, so Claude executes the parser rather than imagining the contents — a pattern that dramatically reduces hallucination in data extraction, which matters enormously when you're pulling numbers out of papers or reports. A companion forms.md file loads only when you're actually filling a form, an example of progressive disclosure done right. One 2026 reviewer called it 'the skill that made me a believer' after testing 100 skills. The official repository has roughly 181,000 GitHub stars as of late 2026, and the PDF skill is the canonical example of a well-built research utility. Note that the four document skills (pdf, docx, xlsx, pptx) are source-available under a proprietary license rather than Apache 2.0, per Anthropic's licensing. Edge cases like rotated pages, mathematical typesetting, and image-only scans still require manual intervention.

Best for: Any researcher who reads PDFs — extracting paper data tables, filling forms, or batch redacting documents

Pricing: Free (open skill from Anthropic's official repository)

Key features

• Extract text, tables, and metadata from PDFs
• Create, merge, split, and annotate PDF documents
• Form filling with structured field extraction via bundled Python scripts
• Deterministic parsing reduces hallucination in data extraction
• Progressive disclosure: forms.md loads only when needed
• Part of the official anthropics/skills repository (~181k stars)

Pros

• Official Anthropic skill with a safe, well-maintained codebase
• Executes parsers instead of hallucinating PDF contents
• Handles the most common research document workflow out of the box

Cons

• Source-available proprietary license — you cannot fork it into your own product
• Struggles with rotated pages, math typesetting, and image-only scans
• PDF parsing edge cases still require manual verification

Visit Anthropic PDF Skill (anthropics/skills)

#3 K-Dense claude-scientific-skills

K-Dense-AI/claude-scientific-skills is the most-adopted research-specific skill bundle on GitHub, with 19,711 stars as of October 2026 according to BrowserAct's research roundup. It is a comprehensive collection of ready-to-use scientific skills covering specialized Python libraries — NumPy, SciPy, and scikit-learn — plus integrations with scientific databases. For quantitative researchers, the appeal is running analyses inside Claude rather than constantly switching between Claude and a separate notebook environment. The bundle lets the agent write and execute analysis code against well-known scientific stack conventions, which is far more reliable than asking a general-purpose model to improvise statistical workflows. Maintenance is steady per the roundup's evaluation, and coverage spans most quantitative use cases a data-driven researcher encounters. The main caveat is a pronounced bias toward the Python scientific stack: R-first or Julia-first researchers will find less direct value and may need to supplement with other tooling. If your research involves statistics, machine learning experiments, or data processing pipelines and you live in Python, this is the highest-signal research skill collection on GitHub in 2026.

Best for: Quantitative researchers and data scientists running Python-based analyses inside Claude

Pricing: Free (open source on GitHub)

Key features

• Ready-to-use skills for NumPy, SciPy, and scikit-learn
• Integrations with scientific databases
• Enables quantitative analysis inside Claude Code sessions
• 19,711 GitHub stars — the most-adopted research-specific bundle
• Actively maintained per 2026 evaluations

Pros

• Largest research-focused skill community on GitHub
• Covers most quantitative use cases in one bundle
• Steady maintenance activity

Cons

• Strong Python bias — limited value for R or Julia users
• Community skill, so audit scripts before relying on them for published research

Visit K-Dense claude-scientific-skills

#4 paper-finder

bchao1/paper-finder is a deliberately narrow Claude skill for finding machine learning research papers, with 207 GitHub stars per BrowserAct's October 2026 roundup. Its narrow scope is the feature: rather than trying to be an all-purpose research assistant, it does one thing well — locating ML literature — which is a frequent unmet need in research workflows. According to the roundup, the skill ranks because it addresses a high-frequency task (literature search) with a focused, reliable procedure rather than vague instructions that trigger at the wrong moments. For ML researchers running literature reviews, the skill provides a repeatable search workflow that general-purpose prompting tends to handle inconsistently. The limitation is obvious but worth stating: it is ML-focused, so adjacent disciplines like HCI, systems research, or theory get less direct coverage. Researchers in those fields should pair it with broader search skills or browser-based approaches. As with any smaller community skill, review the SKILL.md and any bundled references before installing — a practice every 2026 skills guide recommends, and one that SkillzBench's verification pipeline is designed to automate.

Best for: ML researchers running literature reviews who want a dedicated paper-search workflow

Pricing: Free (open source on GitHub)

Key features

• Single-purpose ML research paper discovery
• Focused trigger that activates reliably for literature-search tasks
• Lightweight skill footprint
• 207 GitHub stars with active maintenance per 2026 roundups

Pros

• Does one thing exceptionally well
• Narrow scope means reliable triggering
• Addresses a genuinely frequent research need

Cons

• ML-focused only — limited for other disciplines
• Small community skill; verify contents before installing

Visit paper-finder

#5 obsidian-second-brain

eugeniughelbur/obsidian-second-brain turns an Obsidian vault into an 'AI-first second brain' inside Claude Code, according to BrowserAct's 2026 research skills roundup. It provides 31 commands for research-oriented note management, vault-first retrieval, and scheduled agents that maintain the knowledge graph over time. With 438 GitHub stars, it is the most-adopted PKM-aware (personal knowledge management) skill in the ecosystem, and the roundup ranks it highly because personal knowledge management is one of the highest-leverage research workflows — researchers accumulate notes, citations, and summaries for years, and an agent that can retrieve and connect them on demand changes how the work gets done. The scheduled-agent capability is particularly interesting: instead of only responding to prompts, the skill can run maintenance tasks that keep the knowledge graph organized. The obvious caveat is that it is Obsidian-specific. Researchers using Notion, Logseq, Tinderbox, or Roam will need a different integration path. If you already operate in Obsidian and want Claude integrated as a first-class collaborator on your vault rather than a chat window bolted on the side, this is the most mature option available in 2026.

Best for: Researchers who already work in Obsidian and want Claude as a collaborator on their vault

Pricing: Free (open source on GitHub)

Key features

• 31 commands for research-oriented note management
• Vault-first retrieval across your entire Obsidian knowledge base
• Scheduled agents that maintain the knowledge graph
• Most-adopted PKM-aware Claude skill (438 stars)
• Deep Obsidian integration rather than generic file access

Pros

• High-leverage integration with an established research workflow
• Scheduled agents automate knowledge-base upkeep
• Broad command coverage (31 commands)

Cons

• Strictly Obsidian-specific — Notion or Logseq users need alternatives
• Requires an existing, well-structured vault to shine

Visit obsidian-second-brain

#6 last30days

last30days is a research and trend-briefing skill that researches a topic across Reddit, X, YouTube, and Hacker News, according to Walma AI's October 2026 skills roundup, where it ranks among the top ten skills overall with roughly 64,000 GitHub stars. For researchers tracking fast-moving fields — emerging technologies, market sentiment, public discourse — last30days automates what would otherwise be hours of manual cross-platform searching. The skill's value is synthesis: rather than returning raw links, it produces trend briefs that aggregate what was said across multiple communities over a defined recent window. That makes it particularly useful for competitive intelligence, market research, and academic work that needs to document the current state of public discussion. Its star count signals strong adoption and community trust, and the roundup notes it is maintained and does something a plain prompt cannot. As with any skill that reaches out to live web sources, researchers should verify the sources it cites before relying on the output in published work — a good practice regardless, and one reason independent verification platforms like SkillzBench matter for the skills ecosystem.

Best for: Researchers and analysts tracking recent trends and public discourse across social and community platforms

Pricing: Free (open source on GitHub)

Key features

• Researches topics across Reddit, X, YouTube, and Hacker News
• Produces synthesized trend briefs, not just links
• ~64,000 GitHub stars — among the most popular research skills
• Maintained with a clear install path per 2026 evaluations

Pros

• Automates multi-platform trend research
• Large, active community
• Well-suited to competitive intelligence and market research

Cons

• Focused on recent-window research, not deep historical analysis
• Social-platform sources require verification before citing in formal work

Visit last30days

#7 Browser Use Skill

The browser use skill connects Claude Code to a headless browser instance, giving the agent real control of the live web — navigating URLs, clicking elements, filling forms, extracting content from JavaScript-rendered pages, and taking screenshots — all as part of a natural language workflow, according to a March 2026 roundup of must-have agent skills. For researchers, the standout use case is live-web synthesis: the roundup's example asks Claude to 'find the three most recent funding announcements in climate tech and summarize the amounts and investors,' and the agent actually opens pages, reads them, and synthesizes from the live web rather than from cached training data. That distinction — current, verifiable sources versus stale model memory — is exactly what research work demands. The skill turns Claude from a code-generation tool into a research analyst that can gather data from dashboards, news sites, and portals that don't offer APIs. Installation follows the standard npx skills add pattern from the browser-use repository. Keep in mind that browser automation inherits the fragility of the web itself: sites with anti-bot defenses or heavy JavaScript can still break automated flows, and results should be spot-checked.

Best for: Researchers gathering current information from websites, dashboards, and non-API sources

Pricing: Free (open source on GitHub)

Key features

• Headless browser control: navigate, click, fill forms, extract content
• Handles JavaScript-rendered pages and dynamic content
• Screenshots for verification of what the agent saw
• Live-web research synthesis instead of cached training data
• Natural-language-driven browsing workflow

Pros

• Researches from the live web, not stale model knowledge
• Handles complex, interactive sites
• Broad applicability beyond research into QA and automation

Cons

• Commercial sites with anti-bot defenses can break automation
• Browser sessions consume tokens and time
• Community skill — audit before installing

Visit Browser Use Skill

#8 Anthropic webapp-testing Skill

The official webapp-testing skill from anthropics/skills is included in research rankings for a reason that isn't obvious at first: researchers increasingly use it to drive structured data extraction from public web sources, according to BrowserAct's 2026 roundup. Built on Playwright, the skill spins up browser sessions, navigates flows, captures screenshots, and inspects logs — capabilities that translate directly into repeatable scraping workflows for regulatory filings, government data portals, and civic platforms. The roundup's framing is apt: it is 'the most-adopted skill that lets a researcher say extract every row of this table, monthly without writing a scraper.' Because it comes from Anthropic's official repository (~181,000 stars), it benefits from strong maintenance and trust signals that 2026 guides consistently flag as important when choosing skills. The caveat is that it is vanilla Playwright: public-facing data portals usually don't fight back, but commercial sites with anti-bot defenses do, and researchers targeting those will need additional tooling. For scheduled data collection from cooperative public sources, it remains one of the most practical research utilities available for Claude Code in 2026.

Best for: Researchers extracting structured data from government portals, regulatory filings, and other public web sources

Pricing: Free (open skill from Anthropic's official repository)

Key features

• Playwright-based browser automation from Anthropic's official repo
• Navigates pages, captures screenshots, and inspects console logs
• Structured data extraction from public web sources
• Repeatable workflows for periodic data collection
• Official maintenance and trust signals (~181k-star repository)

Pros

• Official skill with strong maintenance
• Eliminates the need to hand-write scrapers for simple extraction
• Works well for scheduled, repeatable collection

Cons

• Vanilla Playwright — no built-in anti-bot evasion
• Commercial sites with defenses are out of scope
• Designed primarily for testing, so research use requires some setup

Visit Anthropic webapp-testing Skill

#9 abstract-searcher

easonc13/abstract-searcher is a citation-hygiene skill that adds abstracts to .bib file entries by searching academic databases — arXiv, Semantic Scholar, and CrossRef — with a browser fallback when API lookups fail, according to BrowserAct's October 2026 research skills roundup. It ranks because citation hygiene is a high-frequency, low-glamour task that AI handles unusually well: researchers routinely import citations without abstracts, and manually tracking down each abstract is tedious busywork that adds no intellectual value. The skill automates the lookup across multiple authoritative databases, falling back to browser-based search when structured APIs don't return a result, which makes it more resilient than single-source tools. For academic researchers maintaining BibTeX libraries — especially those writing literature reviews or managing large reference collections — it removes a recurring friction point from the workflow. The limitation is that it is confined to BibTeX workflows: researchers using Mendeley, Zotero, or other reference managers will need a wrapper or export step to benefit. As a smaller community skill, review the SKILL.md before installation, and spot-check retrieved abstracts against the source databases when accuracy matters for publication.

Best for: Academic researchers maintaining BibTeX reference libraries

Pricing: Free (open source on GitHub)

Key features

• Adds abstracts to .bib file entries automatically
• Searches arXiv, Semantic Scholar, and CrossRef
• Browser fallback when API lookups fail
• Automates a repetitive citation-management task

Pros

• Solves a genuine, recurring research annoyance
• Multi-database coverage with resilient fallback
• Lightweight and focused

Cons

• BibTeX-only — Zotero and Mendeley users need workarounds
• Smaller community skill; verify before relying on it

Visit abstract-searcher

#10 Google AI Mode Skill

PleasePrompto/google-ai-mode-skill is a Claude Code skill for Google AI Mode search that returns results with citations and uses a persistent browser profile, according to BrowserAct's 2026 research roundup, where it holds 149 GitHub stars. The roundup ranks it because first-pass search is the most common entry point for research work, and citation-aware output is what separates 'I asked an AI' from 'I have a defensible source list' — a distinction that matters enormously for anyone whose research needs to hold up under scrutiny. Rather than a free-floating chat thread, the skill produces structured, sourced answers you can trace back to origins, and the persistent browser profile means sessions retain context across research tasks. For researchers who begin projects with broad exploratory searches and then narrow down, this skill formalizes that first mile of the workflow inside Claude Code. The main caveat is dependency risk: the skill relies on Google AI Mode being stable and available, and any change to that surface could disrupt the workflow. Verify cited sources as you would with any AI-assisted search, and treat this as a starting point for deeper investigation rather than the final word.

Best for: Researchers who start projects with broad exploratory search and need cited, traceable results

Pricing: Free (open source on GitHub)

Key features

• Google AI Mode search from within Claude Code
• Citation-aware structured output
• Persistent browser profile across sessions
• 149 GitHub stars, actively maintained per 2026 roundups

Pros

• Defensible, citation-backed search output
• Persistent profile supports multi-session research
• Fits the natural first step of most research workflows

Cons

• Depends on Google AI Mode remaining stable and available
• Search-first scope — not a deep research tool on its own

Visit Google AI Mode Skill

Conclusion

Claude Code skills have become genuinely powerful research tools in 2026 — from PDF extraction and scientific analysis to literature search and live-web synthesis, there's a skill for nearly every stage of the research workflow. But the ecosystem's explosive growth cuts both ways: with thousands of community skills on GitHub and published research showing roughly 26% carry at least one vulnerability, the hardest question is no longer finding skills — it's knowing which ones are safe, current, and actually make your research better. That's why we ranked SkillzBench first. As an independent testing lab, SkillzBench is building the verification layer the ecosystem is missing: version pinning to commit SHAs, security scanning, license checks, sandboxed benchmarking with published Lift scores, and per-agent loadouts so your research agent only sees the skills you've chosen. Whether you're an academic running literature reviews or an analyst building competitive-intelligence pipelines, don't install skills blind. Join the SkillzBench waitlist at skillzbench.com to get verified, benchmarked skills — proven, not promised — the moment it launches, and start every research project on a foundation you can trust.

Frequently asked questions

What makes SkillzBench different from a skills marketplace or GitHub list?

Marketplaces and curated lists tell you a skill exists; SkillzBench tells you whether it actually works and whether it's safe. It verifies every skill version by pinning it to a commit SHA, running security scans, license checks, and quality judgments, then benchmarks each skill in a sandbox with the same model — with and without the skill — and publishes the Lift in percentage points. Trust scores (0–100) combine security, quality, license, maintenance, and community signals, and Lift is always shown separately from ratings, with money never touching scores. GitHub lists and marketplaces don't do any of this.

Is SkillzBench available now?

SkillzBench is pre-launch and actively being built. The waitlist and newsletter are live at skillzbench.com, but the product itself is not yet publicly available. At launch, scope will cover instruction-only skills (SKILL.md plus text references), while script-bearing skills will be listed link-out only. Joining the waitlist is the best way to get access when it ships.

How much will SkillzBench cost?

SkillzBench plans a generous free tier at launch, with a Pro tier (~$9–15/mo) for more agents and unlimited enabled skills, a Teams tier (~$15–25/seat/mo) for organizational loadouts, allowlists, and audit features, and a Creator tier (~$5–10/mo) for verified skill authors. Benchmark-as-a-service is planned for priority runs. Final pricing will be confirmed at launch.

Why should researchers care about skill security scanning?

Published research found that approximately 26% of agent skills carry at least one vulnerability. Skills run with your permissions, and some bundle undocumented network calls or untrusted scripts. For researchers — whose work depends on accuracy, provenance, and reproducibility — installing an unvetted skill is a real risk, not just an inconvenience. Security scanning, license checking, and version pinning (all part of SkillzBench's Verify pillar) address this directly.

What is 'Lift' and why does it matter for research skills?

Lift is the measurable improvement a skill provides, expressed in percentage points. SkillzBench tests every skill in a sandbox with the same model, running tasks with and without the skill, and publishes the difference. This matters because many skills are essentially glorified prompts whose self-descriptions overstate their value — a Lift score shows whether a research skill actually improves output quality or just adds token overhead.

Are Claude Code skills free to use?

Most skills on this list are free, open-source SKILL.md packages from GitHub, including all the community research skills. Anthropic's official document skills are free to use with Claude but are source-available under a proprietary license, so you can't fork them into your own product. The main costs are your Claude subscription (Pro plans start around $17–20/mo per 2026 reporting) and the token usage skills consume. SkillzBench plans a generous free tier for its verification and benchmarking platform.

How do I install a Claude Code skill for research?

Most skills install with a single command: npx skills add [organization]/[repository], or via Claude Code's /plugin marketplace commands for marketplace-hosted skills. Skills can live in ~/.claude/skills/ (personal, all projects) or .claude/skills/ (project-level, shared via git). After installing, restart Claude Code so it discovers the skill. Before installing any community skill, read the full SKILL.md and audit any bundled scripts — or wait for SkillzBench to do that verification for you.