More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Figure Audit: Catch Figure Problems Before Reviewers Do

    1

    This skill turns your agent into a careful figure reviewer.

    $9

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Figure Audit: Catch Figure Problems Before Reviewers Do

    Figure Audit: Catch Figure Problems Before Reviewers Do

    Example session with this skill installed

    Review Figure 3 before I submit. It's in research/paper/draft.pdf, and research/figures/fig3.py is the script that made it (the data files are next to it). It will print at double-column width, 183 mm. Tell me what a reviewer would object to, check it for colour-blind readers, and go ahead and redraw it without waiting for me.

    • Read your context and instructions
    • Compiled the figure audit
    • Generated the document

    Figure 3 isn't ready to submit. A reviewer would say both claims in the text that cites it depend on how it was drawn more than on the data.

    What a reviewer would object to, worst first
    1.

    Panel a, cut-off bars. The y-axis starts at 66 %, so catalyst C's bars are about 1.6× taller than B's when the yield is only about 5 % higher.
    2.

    Panel b, two y-axes. The temperature axis is set to 59.5–66.5 °C, and that range is what makes the two curves sit on top of each other.
    3.

    "Outperformed all others" isn't supported with n = 3. C is not clearly separated from D in water (Welch t = 1.6) or from B in ethanol (t = 1.9).
    4. The heat-transfer conclusion has no test of which way the cause runs. The outlet is up to 6 °C above the 60 °C jacket, so the reaction could be heating the reactor.
    5.

    Error bars aren't defined, n isn't stated, and there are no a/b panel letters although the text cites Fig. 3a and 3b.
    6. The figure in the draft is a 100 ppi image; include the vector fig3.pdf instead.

    Colour-blind readers: panel a fails. Red and green are the only cue separating water from ethanol, and they merge under simulated deuteranopia. Panel b gets by because one line is dashed.

    Redraws (both at 183 mm, no audit errors, pass greyscale and all three simulations):

    • good.png: your layout, with every replicate plus mean ± SD, blue/orange plus marker shapes, and the twin axis split into two stacked panels.
    • better.png: catalyst C in one accent colour, every replicate with its 95 % interval, solvents labelled on the plot.

    Full report, scripts and check sheets are in research/reviews/figure-audit-fig3/. Your script, data and draft are untouched.

    fig3.pdf

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    What it does

    A figure is read before its caption, so a truncated axis or a convenient second y-axis can oversell a result that the text then has to defend. This skill turns your agent into a careful figure reviewer. Point it at a figure (PNG, a PDF page, or the matplotlib script that drew it) and it:

    • Pins down the claim the figure supports, from the caption and the sentence that cites it, and judges everything against that claim.
    • Audits the plotting script automatically without overwriting your files: twin y-axes, bars that don't start at zero, legends far from the data, grids, rotated or crowded ticks, missing units and panel letters, error bars that need defining, and text below the minimum at the width the figure will actually print.
    • Checks colour-blind and greyscale readers: one sheet with greyscale plus simulated protanopia, deuteranopia and tritanopia views, and a pairwise check of the script's data colours with the numbers.
    • Checks the numbers behind the claim when the data are there: replicates, intervals, how far a truncated axis exaggerates a difference.
    • Ranks every finding by harm (misleads, slows reading, clutter, polish) with the evidence and a concrete fix, and says which findings came from the automatic audit and which are judgement.
    • Redraws the figure twice when it has the script and data: Good, a minimal fix in your own style, and Better, a message-first redesign with one accent colour, direct labels and a caption that says what to see. Both are re-audited and inspected before you see them.

    Reviews figures; doesn't make them from scratch

    Figure Audit reviews figures that already exist and the scripts behind them. It pairs well with Aurora Lab's Paper Figures (which designs figures from data) and Figure Editor (which assembles panels into one figure), but needs neither.

    What's included

    • SKILL.md: the review workflow, a checklist in five groups (integrity, chart form, construction, colour, message), severity bands, report format and how to phrase a critique for a co-author or a referee report
    • scripts/figtools.py: renders PDF pages, crops panels, makes greyscale and colour-blind views and a comparison sheet, and audits matplotlib scripts headless
    • scripts/plotkit.py: matplotlib techniques for the redesign (direct labels, accent colour, replicates with mean and 95 % CI, 1-2-5 ticks)
    • examples/: a deliberately flawed synthetic figure script with its data, and the report, minimal fix and redesign a real run produced

    Part of the Aurora Lab research pipeline

    Works entirely on its own. With a research/ folder it reads figures from research/figures/ and the manuscript from research/paper/, and writes each review to research/reviews/figure-audit-<figure>/. No skill calls another.

    It reviews existing figures; it does not design figures from scratch or assemble panels. The automatic script audit covers matplotlib only; for R, Excel, Prism or other tools it reviews the exported image or PDF, and redraws are made in matplotlib. Auditing a script means running it, so audit only scripts you trust; files the script writes other than its figures are not intercepted. Colour-blind views are simulations of full dichromacy (milder colour-vision deficiencies are common and less severe); they are a strong screen, not a test with real readers. With only an image, it describes the redesign and asks for the data rather than reading values off pixels. Judgements about the claim, the caption and the statistics are made by the agent and reported as such. Minimum text sizes differ between venues; check your journal's current author guidelines.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 28 days ago

    Requires

    Python 3.10+
    NumPy
    Pillow
    Matplotlib
    Poppler or PyMuPDF
    Python
    macOS

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    File Scopes

    research/figures/**
    research/paper/**
    research/reviews/**

    Reads the figures, manuscript PDF, plotting scripts and data files you point it at (by default in research/figures/ and research/paper/). Runs your matplotlib script locally to audit it, with its savefig calls intercepted so your figure files are not overwritten. Writes only into one review folder per figure (research/reviews/figure-audit-<figure>/, or figure-audit-<figure>/ next to the figure): the report, check images and two redraw scripts with their output. Never modifies your figure, script, data or manuscript. No network access; no installs without asking.

    Listed29 days ago
    Updated28 days ago

    What's inside

    Frequently Asked Questions

    Paper Figures designs and draws figures from your data. Figure Editor assembles finished images into one multi-panel figure. Figure Audit reviews figures that already exist (and the scripts behind them), ranks what is wrong and redraws them. Each works on its own.

    Python 3.10+ with numpy, Pillow and matplotlib. To review pages of a PDF, also poppler's pdftoppm (brew install poppler, apt install poppler-utils) or the pymupdf package. PNG, JPEG and TIFF figures need neither.

    Automatic: the script audit (twin axes, bars not starting at zero, legends, grids, crowded or rotated ticks, missing units and panel letters, error bars to define, colour pairs that merge for colour-blind readers or in greyscale, text size at the printed width) and the greyscale and colour-blind images. Judgement: what the figure claims, whether the axis range or the statistics support it, the caption, and the redesign. The report says which is which.

    Yes. It renders the page, crops panels, runs the colour checks and works through the checklist by eye. Without the data it describes the redesign precisely and asks for the data; it never reads values off pixels and presents them as data.

    Yes. It phrases findings as observations with a suggested remedy, ready for a report. The skill itself makes no network calls, but your agent's model provider sees what the agent reads, as with any agent task, so follow your journal's confidentiality rules for manuscripts under review.

    No. It writes only into a separate review folder: the report, the check images, and new good.py and better.py scripts with their output. Your files stay untouched.

    New in Research & Analysis