Data-analyst work is two jobs taped together. Half the day is rigorous: write the SQL, validate the join, sanity-check the totals, build the chart, defend the methodology. The other half is logistics: pull a PDF apart, paste a CSV into a spreadsheet, format a board deck, rename forty columns, sweep up someone else's mess. The skills below compress the logistics half so you have more hours left for the rigorous half — and a few add hard guardrails so the rigorous half does not get undone by a destructive command at 6pm on a Friday. All from verified plugins with real commit history.
From the flagship skills plugin (122,478 stars). The single most useful skill for the median data analyst. Reads, writes, edits, and creates Excel files — formulas, formatting, charts, conditional formatting, named ranges, multi-sheet workbooks. Cleans messy CSVs (malformed rows, junk headers, mixed dtypes) into proper spreadsheets. Outputs real .xlsx with working formulas, not screenshots of grids.
When to use: any time the deliverable is a spreadsheet — a board reporting pack, a vendor data load, a model output you need to hand to a non-technical stakeholder. Especially valuable for the cleanup leg: when finance sends you a workbook with merged cells, footers in the data, and three header rows, this skill does the unmerging-and-flattening work that would otherwise eat thirty minutes.
Also from skills. Extracts text and tables from PDFs, runs OCR on scans, splits and merges files, fills forms. The data-analyst use case is narrow but high-value: pulling tables out of vendor reports, regulator filings, third-party research PDFs, and earnings releases. Anyone who has tried to copy a PDF table into a spreadsheet knows what column-and-row preservation across page breaks is worth.
When to use: when source data only exists as a PDF — quarterly filings, government reports, scanned invoices, supplier catalogs. Combine it with xlsx in the same session: extract the table from the PDF, write it to a real spreadsheet, then run the analysis. Skips the manual transcription step that introduces 80% of data-entry bugs.
From skills. Generates and edits PowerPoint files — slide layouts, embedded tables, charts, speaker notes, templates. For data analysts, the deck is often the actual deliverable: monthly business review, quarterly board pack, weekly stand-up summary. Building those manually from query output is hours of clicking; this skill takes structured numbers in and produces a finished deck.
When to use: any recurring reporting pack. Pair the skill with a templated layout once — chart placement, color palette, footer — and the next month's deck is one prompt instead of an afternoon. Especially useful for analysts who hate slide work and have learned to dread the last day of every reporting cycle.
From the flagship superpowers plugin (164,658 stars). Forces a disciplined hypothesis-and-test loop: state the hypothesis, design the minimal experiment, observe, update. For analysts the moment that matters is when a number changes and nobody knows why — revenue down 4% week-over-week, conversion up 12%, average order value silently drifting. This skill imposes the discipline that prevents "random fix and pray" debugging on data anomalies.
When to use: any time a metric moves and you can't immediately tell if it's a data pipeline issue, a definition change, a seasonal effect, or a real business signal. The first experiment should always be "run the same query against last week's snapshot" — reproducing under controlled conditions before testing causes. Random changes to SQL are an Ouija board.
A hard gate against claiming work is done before evidence supports it. Refuses to declare a task complete without running verification commands and showing output. For analyst work this is the defense against the most common silent failure: shipping a number that looks right but isn't — the row count looks plausible, the totals tie to the prior version, and a definition flip somewhere upstream means the metric now means something different than the slide says it does.
When to use: before every "the analysis is ready" claim. Make verification mean: row counts match the source, sums tie to a known control total, the date filter is the date filter you said it was, and the join did not silently drop rows. Evidence before assertions, every time.
Safety guardrails for destructive commands. Warns before DROP TABLE, TRUNCATE, DELETE without a WHERE, rm -rf, force pushes, and other operations with no "undo" button. From gstack (80,641 stars). For data analysts who have direct write access to a warehouse — even staging — this is cheap insurance against the career-defining bad command.
When to use: always on, but especially when working in a notebook against a live warehouse, cleaning up scratch tables, or running an ad-hoc data fix from a Slack request. The expensive lesson here is the time someone ran DELETE FROM dim_customer intending to scope it and forgot the WHERE. The recovery is a restore from backup if you're lucky and a postmortem either way.
Enforces the red-green-refactor cycle: write the assertion first, watch it fail, implement, watch it pass. The data-analyst translation is data tests — write the assertion about expected behavior before writing the transform. Examples: "customer_id is never null," "order_total is always within 0.01 of the sum of line_items," "the daily revenue series has no gaps," "refunds never exceed gross sales for any single day."
When to use: before touching any pipeline, dbt model, custom aggregation, or definition change. Data leakage bugs, off-by-one date-window bugs, and silent definition regressions all surface the moment you have a cheap assertion that fails when the assumption breaks. These are the bugs that ship to a board deck before anyone notices.
How to install
Each skill lives inside a plugin. Add the plugin marketplace once, then install with a single command. The skill detail page has the exact install string.
If you're new to Claude Code plugins and doing analyst work, the highest-ROI first install is the Anthropic skills plugin (for xlsx, pdf, pptx together) plus superpowers (for the debugging, verification, and TDD skills above). Between them you get the entire "messy file in → board-ready deliverable out" pipeline plus the rigor layer that keeps the deliverable defensible. Add careful from gstack the moment you have any kind of write access to production data.