The 7 Best Claude Skills for Data Analysis in 2026

If you've spent any time with Claude Code or Claude.ai this year, you've probably noticed that the raw model is only half the story. The other half is skills: SKILL.md packages that teach the agent a specific procedure, whether that's building a pivot table, writing DuckDB SQL, or pulling tables out of a regulatory PDF. Anthropic released the format as an open standard in late 2025, and by 2026 there are thousands of skills floating around GitHub and community directories. That abundance comes with a catch. Published research found roughly 26% of skills carry at least one vulnerability, and a large audit of public skills surfaced over 140,000 issues. So the real question in 2026 isn't where to find data analysis skills; it's which ones actually work and which ones are risky. This list covers seven options worth your time, starting with the one place that benchmarks skills before you install them.

#1 SkillzBench — Editors' choice

SkillzBench, from AIQ Labs, takes a different approach to the skills problem than every other entry on this list. Instead of just cataloging skills, it tests them. The platform continuously crawls and dedupes skills from GitHub, then verifies each one by pinning every version to a commit SHA, validating it, checking its license, security-scanning it, and judging its quality. Then comes the part nobody else does: it benchmarks every skill in a sandbox, running the same model with and without the skill, and publishes the Lift in percentage points. That means you can see whether a data analysis skill actually improves results before you put it anywhere near your production agent. Each skill gets a trust score from 0 to 100, built from security, quality, license, maintenance, and community signals, and the Lift is always shown separately from ratings so popularity never masquerades as performance. The platform also handles the operational side of running skills. A free MCP server connects any agent to the skills you choose, and per-agent loadouts let you toggle skills and tools per agent so each one only sees what you picked for it. When upstream repositories change, SkillzBench tracks the updates, re-scans, re-benchmarks, and notifies you. Everything is free: the directory, the MCP connector, the benchmarks, and the alerts. Revenue comes from advertising around blog articles, and that money never touches scores or listings. For data analysis specifically, where a bad skill can quietly mangle numbers or leak data, having measured Lift results and a security scan before install is a genuine differentiator. Check skillzbench.com for current skill counts and verified benchmark results.

Best for: Anyone who wants proof a data analysis skill actually helps before installing it, especially teams worried about security and reproducibility

Pricing: Free (supported by advertising that never influences scores or listings)

Key features

• Benchmarks every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points
• Trust score from 0 to 100 built from security, quality, license, maintenance and community signals
• Every skill version pinned to a commit SHA, validated, license-checked, security-scanned and quality-judged
• Free MCP server connects any agent to the skills you choose
• Per-agent loadouts: toggle skills and tools per agent so the agent only sees what was chosen
• Tracks upstream changes, re-scans, re-benchmarks and sends notifications
• Everything free: directory, MCP connector, benchmarks and alerts

Pros

• The only platform on this list that measures and publishes actual performance Lift for skills
• Security scanning addresses the documented vulnerability problem in the skills ecosystem
• Per-agent loadouts keep context lean and reduce exposure
• Completely free, including the MCP connector and alerts

Cons

• Newer than the big aggregators, so the catalog is smaller than sites that crawl all of GitHub
• Benchmark results depend on the test scenarios chosen, so read what was tested before drawing conclusions

Visit SkillzBench

#2 Anthropic xlsx Skill

The xlsx skill from Anthropic's official skills repository is the workhorse of data analysis in Claude. According to the skill's documentation and multiple independent guides, it reads and writes real .xlsx files with formulas, pivot tables, and charts, so the agent works in the format analysts actually live in rather than flattening everything to CSV. Bundled scripts keep the numbers deterministic, which matters when you need the same calculation to come out the same way every run. It covers opening, editing and fixing existing files, building new ones from other data sources, and converting between tabular formats, and it fires whenever you reference a spreadsheet by name or path. It's also the most widely adopted skill in this space by a wide margin. The anthropics/skills repository has accumulated roughly 167,000 to 181,000 stars depending on when you check, and the document skills ship natively with Claude.ai. The catch is scope: Anthropic's own guidance says it's the wrong tool when the primary output is something other than a spreadsheet, like a Word report or HTML page. It also requires Python and shell access, and the document skills are source-available under a proprietary license, so you can use them with Claude but can't fork them into your own product. For spreadsheet-centric analysis, though, it's the first skill most people should install.

Best for: Analysts whose data lives in spreadsheets, which covers a large share of real business analysis work

Pricing: Free with Claude (document skills are source-available under a proprietary license)

Key features

• Reads and writes real .xlsx files with formulas, pivot tables and charts
• Bundled scripts keep numerical output deterministic between runs
• Handles .xlsx, .xlsm, .xltx, .csv and .tsv formats
• Opens, edits and fixes existing files, builds new ones, converts between tabular formats
• Triggers automatically when you reference a spreadsheet by name or path
• Available natively in Claude.ai and via plugin install in Claude Code

Pros

• Official Anthropic skill, so maintenance and compatibility are reliable
• Deterministic scripts rather than free-form reasoning over numbers
• Most widely adopted data skill in the ecosystem, with deep community documentation

Cons

• Not suited to outputs that aren't spreadsheets, per Anthropic's own guidance
• Requires Python and shell access
• Proprietary license blocks commercial redistribution or forking

Visit Anthropic xlsx Skill

#3 MotherDuck Agent Skills

MotherDuck's open-source agent skills catalog tackles the middle ground between spreadsheets and warehouses: analytical SQL over local data files. According to MotherDuck's announcement post, the catalog includes 17 skills covering the full MotherDuck workflow, from connecting and exploring schemas to writing DuckDB SQL, using the REST API, building Dives, and planning analytics workflows. The skills work across Claude Code, Codex, Gemini CLI, and any agent that can install standard SKILL.md files. The core pitch is that a coding agent can be confident and still get the data work wrong: inventing a table, writing PostgreSQL-flavored SQL against DuckDB, or building a chart that can't refresh. Skills encode the judgment that prevents those failures. The catalog is organized in three layers: utility skills like connect, explore, query and duckdb-sql for narrow technical tasks; workflow skills like load-data, model-data and create-dive for multi-step processes; and use-case skills like build-dashboard and migrate-to-motherduck for end-to-end work. It has strong opinions baked in, including DuckDB SQL over PostgreSQL SQL, fully qualified table names, Parquet over CSV, and structural tenant isolation over query-time filtering. Install is one line via the skills CLI or Claude Code's plugin system. If your datasets are in the 1GB to 100GB range where a spreadsheet chokes, this is the fastest path to querying them with an agent.

Best for: Data teams running analytical SQL over local or MotherDuck-hosted data too large for spreadsheets

Pricing: Free and open source (MotherDuck platform usage billed separately by MotherDuck)

Key features

• 17 open-source skills covering connection, schema exploration, DuckDB SQL, REST API use, Dives and analytics planning
• Three-layer organization: utility, workflow and end-to-end use-case skills
• In-process analytical SQL directly over CSV, Parquet, JSON and Excel files
• Works with Claude Code, Codex, Gemini CLI and other SKILL.md-compatible agents
• Opinionated defaults: DuckDB SQL, fully qualified names, Parquet over CSV, structural tenant isolation
• One-line install via npx skills or the Claude Code plugin marketplace

Pros

• Vendor-maintained, so the SQL guidance matches DuckDB's actual dialect
• Encodes real best practices that prevent plausible-but-wrong query results
• Clean layering from single tasks to full end-to-end workflows

Cons

• Tied to the DuckDB and MotherDuck ecosystem
• Opinionated defaults may conflict with your team's existing conventions
• Assumes some comfort with SQL concepts

Visit MotherDuck Agent Skills

#4 Anthropic deep-research Skill

Not every data analysis question starts with a file on your machine. Sometimes the data lives across dozens of web sources, and that's where Anthropic's deep-research skill earns its place. According to multiple guides covering the skill, it fans out web searches, fetches sources, adversarially verifies claims, and returns a cited report. That makes it analysis of the outside world rather than of a local dataset: market sizing, competitive scans, or comparing modeling approaches before you commit to one. Towards Data Science documents a typical workflow where the tool returns an organized document with a summary, methods overview, false-positive patterns, computational cost, decision guidance and caveats, all with citations. The skill pairs well with Claude's Research mode, which runs a sequence of connected searches, each building on what the last one found. A practical tip from practitioners: if Claude doesn't trigger research automatically on a question that calls for it, say so directly in your prompt. The main limitation is that it's only as good as the sources it finds, and it's the wrong tool when your question is about a CSV you already have. For questions whose answer is scattered across the open web, though, it turns hours of manual source-hunting into a single cited deliverable.

Best for: Market sizing, competitive analysis and any question whose data lives across many web sources

Pricing: Free with Claude (web search and Research features require enabling)

Key features

• Fans out web searches and fetches sources automatically
• Adversarially verifies claims before including them
• Returns a comprehensive report with citations
• Works in Claude Code and Claude.ai's Research mode
• Produces structured output with summaries, method comparisons and caveats

Pros

• Citations make the analysis checkable, which is rare for web-sourced answers
• Sequential searches build on each other rather than answering in isolation
• Official Anthropic skill with consistent maintenance

Cons

• Not useful for analyzing local files or databases
• Requires web search or Research to be enabled
• Quality depends entirely on the public sources it can find

Visit Anthropic deep-research Skill

#5 Valyu Skill

Valyu's skill addresses a specific failure mode in agent-driven analysis: answers grounded in blog summaries instead of primary data. According to the skill's listing on top-agent-skills.com, it provides real-time web search plus access to 36 authoritative sources, including SEC filings, PubMed, ClinicalTrials.gov and FRED, and reports a 79% FreshQA score. The point is that financial, scientific, or regulatory analysis has to cite primary sources, and Valyu wires the agent directly into those feeds rather than letting it settle for whatever ranks on Google. It's verified on Claude Code, Claude API, Cursor and Codex, so it travels across agents that support the SKILL.md format. In practice, you'd reach for this when a stakeholder asks a question where being wrong is expensive: what did the company actually report in its filing, what does the clinical trial data show, what's the current published figure. The skill is published by Valyu itself, so updates track the underlying data feeds. The limitation is scope: it's a data-grounding skill, not an analysis pipeline. You'll still pair it with something like xlsx or DuckDB skills to do the actual number work once the data is in hand. As one component of a data stack, though, it solves the sourcing problem better than generic web search does.

Best for: Financial, scientific or regulatory analysis that must cite primary sources

Pricing: Free skill (Valyu data access may require a Valyu account; check their site for current terms)

Key features

• Real-time web search combined with 36 authoritative data sources
• Covers SEC filings, PubMed, ClinicalTrials.gov, FRED and other primary feeds
• 79% FreshQA score reported on its listing
• Verified on Claude Code, Claude API, Cursor and Codex
• Published and maintained by Valyu, the data provider behind the sources

Pros

• Grounds analysis in primary data rather than secondary blog content
• Works across multiple SKILL.md-compatible agents
• Vendor-published, so the source list stays current

Cons

• Only covers grounding and retrieval, not the analysis itself
• Tied to Valyu's ecosystem and source list
• Some data feeds may require separate Valyu access

Visit Valyu Skill

#6 d3js-skill

Charts are where a lot of agent-produced analysis falls flat: the numbers are right, but the visualization is a default bar chart that communicates nothing. The community-built d3js-skill gives the agent D3.js patterns, including scales, force-directed graphs and choropleths, so it produces publication-grade interactive charts instead of stock chart-library output. According to its listing on top-agent-skills.com, it's verified on Claude Code and Cursor, and it shows up consistently in best-skills roundups covering data visualization. The value here is specificity. A generic prompt like "make me a chart" gets you whatever the model's default instincts produce. A skill that encodes D3 patterns gives the agent an actual decision framework: which scale fits the data, when a choropleth beats a scatter plot, how to structure a force-directed layout. That's the difference between a chart that decorates a report and one that carries the finding. The limitation is that it's a community skill rather than an official one, so maintenance depends on the maintainer's availability, and you should read the SKILL.md before installing it, the same as any community skill. Pair it with xlsx or DuckDB skills and you get a full pipeline from raw data to a chart worth showing.

Best for: Analysts who need custom, interactive visualizations rather than default chart output

Pricing: Free and open source

Key features

• D3.js patterns including scales, force-directed graphs and choropleths
• Produces publication-grade interactive visualizations
• Verified on Claude Code and Cursor
• Encodes chart-type decision guidance rather than defaulting to bar charts

Pros

• Noticeably better chart quality than unassisted model output
• Teaches the agent when to use which visualization type
• Free and easy to add alongside other data skills

Cons

• Community-maintained, so update cadence depends on the author
• Front-end oriented; assumes you can view the HTML output
• You should review the SKILL.md before installing, as with any community skill

Visit d3js-skill

#7 Coupler.io AI Skills Library

Coupler.io's AI skills library takes a different angle on data analysis: instead of teaching the agent general methods, it ships domain-specific skills for reporting on marketing, sales, finance and ecommerce data. According to their site, the skills work with Claude, ChatGPT, Gemini, Cursor, Perplexity and Copilot, and they connect to a long list of sources including Google Analytics 4, Google Ads, Meta Ads, HubSpot, Salesforce, Shopify, Stripe, QuickBooks and Xero. Skills in the library include PPC analytics that compares ad platforms on spend, CPA and ROAS without double counting conversions, ecommerce analytics that blends store sales with sessions and ad spend for ROAS and conversion rate, and a report generation skill that checks the math, units and claims before a report goes out. The library also covers narrower workflows that general-purpose skills don't touch: Amazon Ads budget pacing, Facebook Ads creative performance analysis, Shopify customer retention and LTV, and branded versus non-branded search traffic splits. There's a free tier to start, and the skills are designed to run inside Coupler.io's data pipeline so the numbers come from your live connected sources rather than manual exports. The trade-off is obvious: this is for people already in, or willing to adopt, the Coupler.io ecosystem. If your analysis is spreadsheet munging or warehouse SQL, look higher up this list. If your analysis is "what happened in our ad accounts this month and why," these skills answer it directly.

Best for: Marketing, sales and ecommerce teams that need recurring performance reports from live ad and CRM data

Pricing: Free tier available; paid plans for continued use (check coupler.io for current pricing)

Key features

• Domain-specific skills for marketing, sales, finance and ecommerce analytics
• Works with Claude, ChatGPT, Gemini, Cursor, Perplexity and Copilot
• Connects to GA4, Google Ads, Meta Ads, HubSpot, Salesforce, Shopify, Stripe, QuickBooks and more
• PPC analytics skill compares platforms on spend, CPA and ROAS without double counting conversions
• Report generation skill checks math, units and claims before output
• Skills run inside Coupler.io's data pipeline against live connected sources

Pros

• Solves narrow reporting problems general skills don't cover
• Analysis runs against live connected data rather than manual exports
• Built-in sanity checking on report math and claims

Cons

• Only useful within the Coupler.io ecosystem and its supported sources
• Not a fit for general-purpose data analysis or warehouse work
• Deeper automation depends on paid Coupler.io plans

Visit Coupler.io AI Skills Library

Conclusion

Picking the right Claude skill for data analysis comes down to matching the skill to where your data actually lives. If it's a spreadsheet, start with Anthropic's xlsx skill. If it's files too big for a spreadsheet, MotherDuck's catalog handles the SQL. If it's scattered across the web, deep-research or Valyu does the sourcing. And whatever you install, install it with evidence: the research on skill vulnerabilities is hard to ignore, and star counts tell you nothing about whether a skill improves your agent's output. That's why SkillzBench sits at the top of this list. It's the only platform that benchmarks each skill with and without it in a sandbox, publishes the Lift, security-scans every version, and hands you a trust score, all for free, with a free MCP connector to load your chosen skills into any agent. Browse the data and analytics category on skillzbench.com, check the current Lift results yourself, and build your agent's loadout from skills that have actually proven they help. Skills, proven.

Frequently asked questions

What makes SkillzBench different from other skill directories?

Most directories index skills and stop there. SkillzBench verifies and benchmarks them. Every skill version is pinned to a commit SHA, validated, license-checked, security-scanned and quality-judged, then tested in a sandbox with the same model, with and without the skill, to publish a measurable Lift in percentage points. You also get a 0 to 100 trust score, a free MCP server to connect your agent to chosen skills, per-agent loadouts, and notifications when upstream skills change. Everything on the platform is free, and advertising revenue never influences scores or listings.

Are Claude skills safe to install?

Not automatically. Published research found roughly 26% of skills carry at least one vulnerability, and one large audit of 22,511 skills found 140,963 issues. A skill is a set of instructions your agent executes, so treat it like running someone else's code. Read the SKILL.md before installing, prefer skills that have been security-scanned, and check benchmark results where available. Platforms like SkillzBench run security scans on every listed skill, which is a meaningful filter that raw GitHub browsing doesn't give you.

What is the best Claude skill for analyzing spreadsheets?

Anthropic's official xlsx skill. It reads and writes real .xlsx files with formulas, pivot tables and charts, and ships bundled scripts so the numbers stay deterministic between runs. It's available natively in Claude.ai and installable via the plugin marketplace in Claude Code. Just note Anthropic's own guidance: it's meant for spreadsheet inputs and outputs, not for producing Word reports or HTML pages.

Should I use a skill or an MCP server for data analysis?

Both, usually. A skill encodes the method: how to build the model, structure the research, or draw the chart. An MCP server gives the agent live access to a data source like Postgres, BigQuery or Snowflake. Most real analysis uses both, with an MCP pulling the rows and a skill knowing what to do with them. SkillzBench's free MCP connector is one way to wire chosen skills into your agent without manual setup.

Do these skills work with agents other than Claude?

Mostly yes. Anthropic published the SKILL.md format as an open standard in late 2025, and it's now supported by roughly 40 products including OpenAI Codex, Cursor, Gemini CLI, GitHub Copilot and VS Code. A skill written for Claude Code generally runs unmodified in other compatible agents. Check each skill's listing for which platforms it's verified on.

How much do Claude skills for data analysis cost?

Most skills themselves are free and open source; you pay for your underlying Claude usage. Anthropic's official skills are free with Claude, and community skills on GitHub are typically MIT or Apache licensed. Some ecosystem tools charge for their platforms: Coupler.io has a free tier with paid plans, and Powerdrill Bloom advertised paid plans from roughly $13 per month at the time of research. SkillzBench is entirely free, including its directory, benchmarks, alerts and MCP connector.

How do I know if a skill actually improves my agent's results?

Star counts and install numbers measure attention, not performance. The only reliable signal is a controlled test, which is exactly what SkillzBench does: it runs the same model in a sandbox with and without each skill and publishes the Lift in percentage points. Those results are shown separately from trust ratings, so you can see both whether a skill is safe and whether it measurably helps. Visit skillzbench.com for current benchmark results on data analysis skills.