[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M74MAX4RMD03EWWP4DDTWQ9T":3,"skill-same:01M74MAX4RMD03EWWP4DDTWQ9T":53},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":29,"ownerHtmlUrl":30,"ownerBlogUrl":6,"ownerDescription":31,"updatedAt":32,"install":33,"resources":46},"01M74MAX4RMD03EWWP4DDTWQ9T","figure-style",null,"Publication-grade figure correctness and legibility rules. Load before drawing any plot and call `apply_figure_style()` — sets a role-mapped font-size ladder, outward ticks, frameless legends, and 300-dpi output. The skill is a checklist, not a house look: data fidelity (claim-titles tested against every row, excluded data never enters summaries), label economy (floor and ceiling), colour threading, chart-choice-by-data-shape, layout, and a render-then-verify QA loop (bbox collision + per-panel perceptual check). Ships helpers: focal_palette, bar_with_points, strip_with_median, end_of_line_labels, panel_letter, set_frame, panel_crops. For multi-panel figures load `figure-composer`; for whole-paper figure arc load `paper-narrative`.","pengqianhan\u002FAI-Human-Research-OS","pengqianhan","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F4286531?v=4","01M74M3DT9HNZ6HPAK8TX69EX9","PengqianHan","01M74M3DTVSN2GXW0JH2W396FK",3,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fclaude-science-skills\u002Fskills\u002Ffigure-style\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: figure-style\ndescription: \"Publication-grade figure correctness and legibility rules. Load before drawing any plot and call `apply_figure_style()` — sets a role-mapped font-size ladder, outward ticks, frameless legends, and 300-dpi output. The skill is a checklist, not a house look: data fidelity (claim-titles tested against every row, excluded data never enters summaries), label economy (floor and ceiling), colour threading, chart-choice-by-data-shape, layout, and a render-then-verify QA loop (bbox collision + per-panel perceptual check). Ships helpers: focal_palette, bar_with_points, strip_with_median, end_of_line_labels, panel_letter, set_frame, panel_crops. For multi-panel figures load `figure-composer`; for whole-paper figure arc load `paper-narrative`.\"\nlicense: Apache-2.0\n---\n\n\n# Publication-Grade Figure Rules\n\n*A checklist for correct, legible, internally-consistent scientific figures. This\nskill does not impose a visual house style — frame, font, and palette are\nparameters. Load it and call `apply_figure_style()` before any plot.*\n\n## §0 Scope\n\n§1–§3, §8, and §9 are **correctness** — they apply to every plot, in every\ncontext, and have no aesthetic content. §4–§7 are **guidance** — defaults that\nproduce a clean result but that a deliberate alternative can override\n(individual rules inside §4–§7 that state a factual\u002Fperceptual invariant — e.g.\n§4.4 semantic-zero centring, §4.5 CVD, §6.9 leader anchoring — still bind). On\nits own, this skill is the inner tier (make one plot good); `figure-composer`\nand `paper-narrative` supply multi-panel and whole-paper context.\n\n---\n\n## §1 Data fidelity & self-consistency\n\n**1.1 Excluded rows.** A row marked excluded or flagged in the source data is\neither omitted entirely or drawn with a visually distinct open\u002Fhatched marker\nand named in the key. It **never** enters a summary statistic plotted alongside\nthe included rows.\n\n**1.2 Comparable conditions only.** Arms measured under non-comparable\nconditions (different N, epoch budget, initialisation, protocol) are not plotted\nas visual peers. Separate them with a facet break or a marker on the label, and\nstate the difference once in the caption.\n\n**1.3 Self-consistency.** Every key, threshold, and title inside the figure must\nbe satisfied by every plotted row. Before saving, walk each categorical outcome\nlabel back to the rule that defines it; if a row's value contradicts its label\nor the title, the figure is wrong, not the data.\n\n**1.4 Claim-titles must be true.** A sentence-title (§5.1) is tested against\nevery category on the axis before rendering. If any contradicts it, qualify the\ntitle (\"on 3 of 4 pairs\") or downgrade it to a description.\n\n**1.5 State n and what was held fixed.** Every panel that draws a summary mark\nstates `n` and the unit of replication, and every small-multiple that holds a\nvariable fixed states the fixed value — in the panel or, when §2 budget is\ntight, in the caption.\n\n**1.6 Reference structure is reference.** A tree, ordering, or topology drawn as\n*context* (a scale bar, a category strip) uses an established reference, not\none inferred from the plotted data. Infer the structure only when the structure\n*is* the result.\n\n**1.7 One number per claim.** A quantitative claim (runtime, accuracy, count)\nhas exactly one canonical value across every panel, caption, and the abstract.\nDefine what it measures and use that value everywhere.\n\n---\n\n## §2 Label economy — floor and ceiling\n\nThe figure shows the pattern; the **caption** carries the context. Design for a\ngeneral scientific reader, not the author.\n\n**2.1 Floor (non-removable).** Every distinct mark, series, glyph, or comparator\nmust be identifiable from the figure alone. The caption explains *why it\nmatters*, not *what it is*. A label is non-removable if deleting it leaves a\nreader asking \"what is that?\"; it is removable only if the question becomes \"why\nis that there?\". Comparator labels name the thing (\"prior method\", \"no joint\ntraining\"), never a bare role word (\"baseline\", \"previous\"). Any term a general\nscientist can't parse gets a one-word gloss.\n\n**2.2 Ceiling.** Per panel: title + axis labels + tick labels + series identity\n(labelled once per row of small multiples) + at most 2–3 result annotations.\nCount the strings; >6 beyond axes\u002Fticks means you're over. The ceiling counts\n*narrative* annotations (callouts, value labels, brackets) — identity labels are\nfloor, not budget.\n\n**2.3 Move to the caption:** n=, what's-held-fixed, abbreviation expansions,\nnon-comparable footnotes, exclusion rationale, methodological caveats.\n\n**2.4 Titles are takeaways.** A reader seeing only the title knows what the\npanel shows. \"Robust to gene dropout\" passes; \"Fewer genes\" fails. Test: read it\naloud cold — if the listener asks \"fewer genes *what*?\", rewrite. For a row of\nsmall multiples that vary one thing, drop per-panel titles for one row-header.\n\n**2.5 Value-on-mark only for the headline number** — the one a reader would\nquote. Everything else is read off the axis.\n\n**2.6 When in doubt, delete the label and re-read.** If the message survives, it\nstays deleted.\n\n---\n\n## §3 Axes, scales, small multiples\n\n**3.1 Axis padding.** Axis limits clear the data by ≥ one marker radius on every\nside; markers and text never touch a spine. `ax.margins(0.04)` after plotting,\nor extend the limit past any annotation.\n\n**3.2 Axis breaks over wasted range.** When data occupy \u003C40 % of an axis, break\nthe axis or start it at the data floor with a clear non-zero tick. Never draw a\nreference line, threshold, or annotation inside a broken-axis gap — the gap has\nno coordinate.\n\n**3.3 Log axes get human-readable ticks** — `10²`, `10³`, or `1k \u002F 10k \u002F 100k`,\nnot raw exponents. **Never** draw filled bars on a log-scaled value axis (bar\nlength encodes ratio to an arbitrary floor); use points + median tick instead.\n\n**3.4 Shared axes across small multiples.** A row or column of small multiples\nshows tick labels once (leftmost \u002F bottommost panel); interior panels keep ticks\nbut drop labels. When the panels share a y-axis and differ only by x-variable,\nrender them as abutting subplots (`wspace≤0.06`) with one row-header title.\n\n**3.5 Fill the box.** A panel's data envelope occupies ≥75 % of its allotted\nrectangle. If a panel's natural aspect leaves dead bands, reshape the grid\n(rowspan, stacked complementary panels) — don't pad the panel.\n\n**3.6 Direction of goodness.** When higher- or lower-is-better is not obvious\nfrom the axis label, place a small upright cue (\"higher = better\") in the\nmargin — once per row of panels, never per panel, and never only in the caption.\nA directional glyph embedded in rotated text rotates with it; set the cue\nupright.\n\n**3.7 Physical width.** A single-row figure at 300 dpi fits the venue's\ndouble-column width. Adding a schematic or labels does not push data panels narrower than\nthey were before.\n\n---\n\n## §4 Colour\n\n**4.1 Threading.** Once a colour is bound to an entity (a method, a feature, a\ncondition), reuse that exact colour for every mark representing that entity\nacross the figure — line, fill, marker, text, heatmap row. Colour *is* the\ncross-reference; a reader should never have to consult a legend twice.\n\n**4.2 Limit hues.** Use as few distinct hues as the data require. When the\nfigure compares a focal series against others, make the focal series visually\ndominant (saturated, heavier weight) and render comparators with lower visual\nweight (desaturated, lighter, or thinner). The focal hue must not coincide with\nany hue in a categorical palette used in the same figure. The focal series must\nremain identifiable even when its mark is zero-width or coincident with others —\nvia outline, marker, or a light tinted band.\n\n**4.3 Hierarchical categories.** When categories nest (groups within groups),\nthe outer level picks the hue family and the inner level samples within it.\n\n**4.4 Continuous and diverging.** Use a perceptually uniform sequential map for\ngeneric continuous values; a single-hue ramp for ordinal rank or size; a\ndiverging map for signed quantities — **always** centred at the semantically\nmeaningful zero (0, 1.0, or median), never the data midpoint.\n\n**4.5 CVD safety.** Never rely on a red\u002Fgreen contrast for a binary or opposing\ndistinction. Any binary pair should remain distinguishable in deuteranopia\nsimulation. Reserve one alarm hue for error\u002Fanomaly\u002Fperturbation marks and do\nnot reuse it as a data-series colour.\n\n**4.6 Two palettes, two legends.** When a figure uses two categorical colour\nsystems, each legend sits adjacent to the first panel where its palette applies.\n\n---\n\n## §5 Typography\n\n**5.1 Sentence titles.** A panel title states the comparison in plain language,\nregular weight, left-aligned. Metric names go on the axis, not in the title.\n\n**5.2 Role-mapped size ladder.** A figure uses **at most three** font sizes,\nmapped to *role* not space: titles\u002Faxis-labels\u002Fseries-identity at the base size;\nlegend\u002Fannotation text one step down; tick labels one step further. Panel\nletters are the only exception (bold, larger). If a label doesn't fit at its\nrole's size, fix the layout or shorten the text — don't reach for an\nintermediate size. `apply_figure_style(sizes=(8,7,6))` sets the ladder.\n\n**5.3 Nomenclature.** Species, gene, and variable names that scientific\nconvention italicises are italicised. Abbreviated codes inherit the rule; expand\nonce on first appearance.\n\n**5.4 Magnitude suffixes.** Large counts use `k \u002F M \u002F B` (`4.2B`, `120 kb`),\nnot comma-grouped full numerals.\n\n**5.5 Numeric annotations.** On-mark numbers use at most 2 significant figures —\nunless 2-sf rounding would make two distinct rows print the same value, in which\ncase show the digit that separates them. Text on a filled mark reaches ≥4.5:1\ncontrast; if it doesn't, place the text outside the mark.\n\n**5.6 No internal codes.** Axis labels use plain-language names; codebase\nabbreviations appear only in parentheses after the readable name or in the\ncaption. Comparator series are labelled with what they *are*, not a role word.\n\n**5.7 Panel letters.** Bold, top-left, outside the axes box. Case follows the\ntarget venue's convention; `panel_letter(ax, 'a', case=...)` handles either.\n\n---\n\n## §6 Chart-family guidance (by data shape)\n\n**6.1 Categorical × numeric.** Show the distribution, not just the summary.\nChart choice follows n: jittered strip with a median tick for small n; box or\nviolin for large n; bar + overlaid raw points or bar + interval when the mean is\nthe message. The `errorbar='ci95'` interval is the t-distribution 95% CI of the\nmean (half-width `t_{0.975,n−1}·s\u002F√n`), so it is valid at small n. Error bars\nand raw-point overlays are alternatives — showing both is usually redundant. A\ncategory absent from a group is marked (`n.d.`, `—`, or a hatched ghost) at its\nslot; an empty slot reads as zero. A zero-valued bar\ngets a visible stub or dot at the baseline.\n\n**6.2 Single-observation categories.** A filled dot with a thin neutral stem to\nthe semantic zero (lollipop). Value labels sit beside the dot.\n\n**6.3 Continuous series.** Mean-per-x as a line with markers; individual runs as\nthin translucent lines or points behind it. Label each series with direct text\nat the right end of its line in preference to a legend box. Summary glyphs\n(per-bin mean\u002Fmedian) use a shape that cannot be mistaken for a raw observation,\nidentical across series, drawn below the raw points in z-order.\n\n**6.4 Distributions on shared support.** When two distributions overlap heavily,\nstack them as small panels with a shared x-axis or use a ridgeline. Overlay only\nwhen the separation is visually clear.\n\n**6.5 Matrices.** When a heatmap is small enough to read (\u003C ~200 cells), print\nthe value in every cell. State the threshold once in the colourbar label.\n\n**6.6 Embeddings.** Dimensionality-reduction scatters (UMAP, t-SNE, PCA) drop\nticks and tick labels; a small corner arrow pair names the axes. Clusters are\nlabelled by thin leader lines to text in surrounding whitespace.\n\n**6.7 Paired prediction vs. observation.** Stack the two as adjacent tracks with\nidentical x and colour; let the alignment carry the comparison. Target regions\nare translucent spans registered in the legend.\n\n**6.8 Insets.** Connect a detail inset to its source region visibly — a bounding\nbox with connector lines, or a translucent wedge.\n\n**6.9 Label the extremes.** On a scatter of named observations, direct-label at\nleast the maximum, minimum, and any flagged point with a thin leader line. After\nrendering, verify every leader endpoint terminates within one marker radius of\nthe row it names.\n\n---\n\n## §7 Layout & narrative\n\n**7.1 Show what is measured before the result.** A reader should grasp what's\nbeing compared before seeing the comparison — via a plain-language title, a\nlabelled schematic, or panel ordering. Any schematic uses the same words and\nglyphs as the data panels' labels.\n\n**7.2 One figure, one message.** A multi-panel figure has a single sentence it\nis trying to make true. Every panel either states it, supports it, or bounds it;\npanels that do none of these belong in supplement.\n\n**7.3 Legends live in whitespace.** Frameless, placed inside the figure's\nnatural whitespace, or replaced by direct labelling. Legend entries are\nswatch-first, left-aligned, and resolve every visually distinct glyph on the\npanel.\n\n**7.4 Row-band headers for nested faceting.** When small multiples are grouped,\neach group gets one spanning header, not repeated per-panel titles.\n\n**7.5 The figure arc.** For a paper: Figure 1 renders the paper's one-sentence\npitch as data — scope, not architecture. Subsequent figures cover mechanism,\nevidence, robustness, application. A panel is judged against the paper's pitch,\nnot just its own figure's claim; content moves between figures if that's where\nthe story needs it. (`paper-narrative` runs this review.)\n\n**7.6 Don't re-decorate a passing panel.** Between revision rounds, a panel that\nalready passes is not made more visually complex to fix nothing. Adding marks or\nlabels to a clean panel is a regression.\n\n---\n\n## §8 Anti-patterns\n\nThese are correctness failures, not style preferences:\n\n- Red and green as opposing categories.\n- Filled bars on a log-scaled value axis.\n- Colourbar ticks that are evenly spaced but miss the semantic centre.\n- A diverging colormap whose centre is the data midpoint, not the semantic zero.\n- An axis title that restates the tick labels.\n- Explaining the direction of goodness only in the caption.\n- A \"reference\" line drawn at a value that is itself one of the plotted points.\n- An excluded row that enters a plotted summary statistic.\n- A leader line whose nearest mark is not the row it labels.\n\n---\n\n## §9 Render-then-verify\n\nAfter `fig.savefig(...)`, before `save_artifacts`:\n\n**9.1 Geometric (bbox) check.**\n```python\nr = fig.canvas.get_renderer()\ntexts = [(t, t.get_window_extent(r)) for t in fig.findobj(mpl.text.Text)\n         if t.get_text().strip() and t.get_visible()]\nspines = [(s, s.get_window_extent(r)) for ax in fig.axes\n          for s in ax.spines.values() if s.get_visible()]\nticklabels = {ax: set(ax.get_xticklabels(which='both') + ax.get_yticklabels(which='both'))\n              for ax in fig.axes}\noverlaps  = [(a, b) for i, (a, ba) in enumerate(texts) for b, bb in texts[i+1:] if ba.overlaps(bb)]\noverlaps += [(t, s) for t, bt in texts for s, bs in spines\n             if bt.overlaps(bs) and t not in ticklabels[s.axes]]\n# assert: overlaps == [] and every text box lies within fig.bbox\n```\nOverlap is defined between *visible* boxes, and a tick label sitting on its own\nspine is not a finding. Fix (move, shorten, stagger) and re-save until clean.\n\n**9.2 Perceptual check.** The bbox check is geometric, not perceptual — it will\nnot catch a low-contrast label, a leader that crosses three others, or a series\ncolour mistakable for another. Crop the saved PNG to each panel and look:\n```python\nfig.savefig(\"figure.png\")\nfor letter, box in panel_crops(fig).items():\n    host.view_image(\"figure.png\", crop=box)\n```\nFor each crop: Is every glyph and mark legible against its background? Does the\nsmallest plotted element have a stroke or stub? Do any leaders cross? Could any\nseries colour be mistaken for another? Does the legend sit beside what it keys?\nA perceptual defect that passes §9.1 is still a defect.\n\n---\n*When in doubt: fewer hues, more direct labels, raw data over summary stats, and\nstate what is being measured before showing the result.*\n","# Publication-Grade Figure Rules\n\n*A checklist for correct, legible, internally-consistent scientific figures. This\nskill does not impose a visual house style — frame, font, and palette are\nparameters. Load it and call `apply_figure_style()` before any plot.*\n\n## §0 Scope\n\n§1–§3, §8, and §9 are **correctness** — they apply to every plot, in every\ncontext, and have no aesthetic content. §4–§7 are **guidance** — defaults that\nproduce a clean result but that a deliberate alternative can override\n(individual rules inside §4–§7 that state a factual\u002Fperceptual invariant — e.g.\n§4.4 semantic-zero centring, §4.5 CVD, §6.9 leader anchoring — still bind). On\nits own, this skill is the inner tier (make one plot good); `figure-composer`\nand `paper-narrative` supply multi-panel and whole-paper context.\n\n---\n\n## §1 Data fidelity & self-consistency\n\n**1.1 Excluded rows.** A row marked excluded or flagged in the source data is\neither omitted entirely or drawn with a visually distinct open\u002Fhatched marker\nand named in the key. It **never** enters a summary statistic plotted alongside\nthe included rows.\n\n**1.2 Comparable conditions only.** Arms measured under non-comparable\nconditions (different N, epoch budget, initialisation, protocol) are not plotted\nas visual peers. Separate them with a facet break or a marker on the label, and\nstate the difference once in the caption.\n\n**1.3 Self-consistency.** Every key, threshold, and title inside the figure must\nbe satisfied by every plotted row. Before saving, walk each categorical outcome\nlabel back to the rule that defines it; if a row's value contradicts its label\nor the title, the figure is wrong, not the data.\n\n**1.4 Claim-titles must be true.** A sentence-title (§5.1) is tested against\nevery category on the axis before rendering. If any contradicts it, qualify the\ntitle (\"on 3 of 4 pairs\") or downgrade it to a description.\n\n**1.5 State n and what was held fixed.** Every panel that draws a summary mark\nstates `n` and the unit of replication, and every small-multiple that holds a\nvariable fixed states the fixed value — in the panel or, when §2 budget is\ntight, in the caption.\n\n**1.6 Reference structure is reference.** A tree, ordering, or topology drawn as\n*context* (a scale bar, a category strip) uses an established reference, not\none inferred from the plotted data. Infer the structure only when the structure\n*is* the result.\n\n**1.7 One number per claim.** A quantitative claim (runtime, accuracy, count)\nhas exactly one canonical value across every panel, caption, and the abstract.\nDefine what it measures and use that value everywhere.\n\n---\n\n## §2 Label economy — floor and ceiling\n\nThe figure shows the pattern; the **caption** carries the context. Design for a\ngeneral scientific reader, not the author.\n\n**2.1 Floor (non-removable).** Every distinct mark, series, glyph, or comparator\nmust be identifiable from the figure alone. The caption explains *why it\nmatters*, not *what it is*. A label is non-removable if deleting it leaves a\nreader asking \"what is that?\"; it is removable only if the question becomes \"why\nis that there?\". Comparator labels name the thing (\"prior method\", \"no joint\ntraining\"), never a bare role word (\"baseline\", \"previous\"). Any term a general\nscientist can't parse gets a one-word gloss.\n\n**2.2 Ceiling.** Per panel: title + axis labels + tick labels + series identity\n(labelled once per row of small multiples) + at most 2–3 result annotations.\nCount the strings; >6 beyond axes\u002Fticks means you're over. The ceiling counts\n*narrative* annotations (callouts, value labels, brackets) — identity labels are\nfloor, not budget.\n\n**2.3 Move to the caption:** n=, what's-held-fixed, abbreviation expansions,\nnon-comparable footnotes, exclusion rationale, methodological caveats.\n\n**2.4 Titles are takeaways.** A reader seeing only the title knows what the\npanel shows. \"Robust to gene dropout\" passes; \"Fewer genes\" fails. Test: read it\naloud cold — if the listener asks \"fewer genes *what*?\", rewrite. For a row of\nsmall multiples that vary one thing, drop per-panel titles for one row-header.\n\n**2.5 Value-on-mark only for the headline number** — the one a reader would\nquote. Everything else is read off the axis.\n\n**2.6 When in doubt, delete the label and re-read.** If the message survives, it\nstays deleted.\n\n---\n\n## §3 Axes, scales, small multiples\n\n**3.1 Axis padding.** Axis limits clear the data by ≥ one marker radius on every\nside; markers and text never touch a spine. `ax.margins(0.04)` after plotting,\nor extend the limit past any annotation.\n\n**3.2 Axis breaks over wasted range.** When data occupy \u003C40 % of an axis, break\nthe axis or start it at the data floor with a clear non-zero tick. Never draw a\nreference line, threshold, or annotation inside a broken-axis gap — the gap has\nno coordinate.\n\n**3.3 Log axes get human-readable ticks** — `10²`, `10³`, or `1k \u002F 10k \u002F 100k`,\nnot raw exponents. **Never** draw filled bars on a log-scaled value axis (bar\nlength encodes ratio to an arbitrary floor); use points + median tick instead.\n\n**3.4 Shared axes across small multiples.** A row or column of small multiples\nshows tick labels once (leftmost \u002F bottommost panel); interior panels keep ticks\nbut drop labels. When the panels share a y-axis and differ only by x-variable,\nrender them as abutting subplots (`wspace≤0.06`) with one row-header title.\n\n**3.5 Fill the box.** A panel's data envelope occupies ≥75 % of its allotted\nrectangle. If a panel's natural aspect leaves dead bands, reshape the grid\n(rowspan, stacked complementary panels) — don't pad the panel.\n\n**3.6 Direction of goodness.** When higher- or lower-is-better is not obvious\nfrom the axis label, place a small upright cue (\"higher = better\") in the\nmargin — once per row of panels, never per panel, and never only in the caption.\nA directional glyph embedded in rotated text rotates with it; set the cue\nupright.\n\n**3.7 Physical width.** A single-row figure at 300 dpi fits the venue's\ndouble-column width. Adding a schematic or labels does not push data panels narrower than\nthey were before.\n\n---\n\n## §4 Colour\n\n**4.1 Threading.** Once a colour is bound to an entity (a method, a feature, a\ncondition), reuse that exact colour for every mark representing that entity\nacross the figure — line, fill, marker, text, heatmap row. Colour *is* the\ncross-reference; a reader should never have to consult a legend twice.\n\n**4.2 Limit hues.** Use as few distinct hues as the data require. When the\nfigure compares a focal series against others, make the focal series visually\ndominant (saturated, heavier weight) and render comparators with lower visual\nweight (desaturated, lighter, or thinner). The focal hue must not coincide with\nany hue in a categorical palette used in the same figure. The focal series must\nremain identifiable even when its mark is zero-width or coincident with others —\nvia outline, marker, or a light tinted band.\n\n**4.3 Hierarchical categories.** When categories nest (groups within groups),\nthe outer level picks the hue family and the inner level samples within it.\n\n**4.4 Continuous and diverging.** Use a perceptually uniform sequential map for\ngeneric continuous values; a single-hue ramp for ordinal rank or size; a\ndiverging map for signed quantities — **always** centred at the semantically\nmeaningful zero (0, 1.0, or median), never the data midpoint.\n\n**4.5 CVD safety.** Never rely on a red\u002Fgreen contrast for a binary or opposing\ndistinction. Any binary pair should remain distinguishable in deuteranopia\nsimulation. Reserve one alarm hue for error\u002Fanomaly\u002Fperturbation marks and do\nnot reuse it as a data-series colour.\n\n**4.6 Two palettes, two legends.** When a figure uses two categorical colour\nsystems, each legend sits adjacent to the first panel where its palette applies.\n\n---\n\n## §5 Typography\n\n**5.1 Sentence titles.** A panel title states the comparison in plain language,\nregular weight, left-aligned. Metric names go on the axis, not in the title.\n\n**5.2 Role-mapped size ladder.** A figure uses **at most three** font sizes,\nmapped to *role* not space: titles\u002Faxis-labels\u002Fseries-identity at the base size;\nlegend\u002Fannotation text one step down; tick labels one step further. Panel\nletters are the only exception (bold, larger). If a label doesn't fit at its\nrole's size, fix the layout or shorten the text — don't reach for an\nintermediate size. `apply_figure_style(sizes=(8,7,6))` sets the ladder.\n\n**5.3 Nomenclature.** Species, gene, and variable names that scientific\nconvention italicises are italicised. Abbreviated codes inherit the rule; expand\nonce on first appearance.\n\n**5.4 Magnitude suffixes.** Large counts use `k \u002F M \u002F B` (`4.2B`, `120 kb`),\nnot comma-grouped full numerals.\n\n**5.5 Numeric annotations.** On-mark numbers use at most 2 significant figures —\nunless 2-sf rounding would make two distinct rows print the same value, in which\ncase show the digit that separates them. Text on a filled mark reaches ≥4.5:1\ncontrast; if it doesn't, place the text outside the mark.\n\n**5.6 No internal codes.** Axis labels use plain-language names; codebase\nabbreviations appear only in parentheses after the readable name or in the\ncaption. Comparator series are labelled with what they *are*, not a role word.\n\n**5.7 Panel letters.** Bold, top-left, outside the axes box. Case follows the\ntarget venue's convention; `panel_letter(ax, 'a', case=...)` handles either.\n\n---\n\n## §6 Chart-family guidance (by data shape)\n\n**6.1 Categorical × numeric.** Show the distribution, not just the summary.\nChart choice follows n: jittered strip with a median tick for small n; box or\nviolin for large n; bar + overlaid raw points or bar + interval when the mean is\nthe message. The `errorbar='ci95'` interval is the t-distribution 95% CI of the\nmean (half-width `t_{0.975,n−1}·s\u002F√n`), so it is valid at small n. Error bars\nand raw-point overlays are alternatives — showing both is usually redundant. A\ncategory absent from a group is marked (`n.d.`, `—`, or a hatched ghost) at its\nslot; an empty slot reads as zero. A zero-valued bar\ngets a visible stub or dot at the baseline.\n\n**6.2 Single-observation categories.** A filled dot with a thin neutral stem to\nthe semantic zero (lollipop). Value labels sit beside the dot.\n\n**6.3 Continuous series.** Mean-per-x as a line with markers; individual runs as\nthin translucent lines or points behind it. Label each series with direct text\nat the right end of its line in preference to a legend box. Summary glyphs\n(per-bin mean\u002Fmedian) use a shape that cannot be mistaken for a raw observation,\nidentical across series, drawn below the raw points in z-order.\n\n**6.4 Distributions on shared support.** When two distributions overlap heavily,\nstack them as small panels with a shared x-axis or use a ridgeline. Overlay only\nwhen the separation is visually clear.\n\n**6.5 Matrices.** When a heatmap is small enough to read (\u003C ~200 cells), print\nthe value in every cell. State the threshold once in the colourbar label.\n\n**6.6 Embeddings.** Dimensionality-reduction scatters (UMAP, t-SNE, PCA) drop\nticks and tick labels; a small corner arrow pair names the axes. Clusters are\nlabelled by thin leader lines to text in surrounding whitespace.\n\n**6.7 Paired prediction vs. observation.** Stack the two as adjacent tracks with\nidentical x and colour; let the alignment carry the comparison. Target regions\nare translucent spans registered in the legend.\n\n**6.8 Insets.** Connect a detail inset to its source region visibly — a bounding\nbox with connector lines, or a translucent wedge.\n\n**6.9 Label the extremes.** On a scatter of named observations, direct-label at\nleast the maximum, minimum, and any flagged point with a thin leader line. After\nrendering, verify every leader endpoint terminates within one marker radius of\nthe row it names.\n\n---\n\n## §7 Layout & narrative\n\n**7.1 Show what is measured before the result.** A reader should grasp what's\nbeing compared before seeing the comparison — via a plain-language title, a\nlabelled schematic, or panel ordering. Any schematic uses the same words and\nglyphs as the data panels' labels.\n\n**7.2 One figure, one message.** A multi-panel figure has a single sentence it\nis trying to make true. Every panel either states it, supports it, or bounds it;\npanels that do none of these belong in supplement.\n\n**7.3 Legends live in whitespace.** Frameless, placed inside the figure's\nnatural whitespace, or replaced by direct labelling. Legend entries are\nswatch-first, left-aligned, and resolve every visually distinct glyph on the\npanel.\n\n**7.4 Row-band headers for nested faceting.** When small multiples are grouped,\neach group gets one spanning header, not repeated per-panel titles.\n\n**7.5 The figure arc.** For a paper: Figure 1 renders the paper's one-sentence\npitch as data — scope, not architecture. Subsequent figures cover mechanism,\nevidence, robustness, application. A panel is judged against the paper's pitch,\nnot just its own figure's claim; content moves between figures if that's where\nthe story needs it. (`paper-narrative` runs this review.)\n\n**7.6 Don't re-decorate a passing panel.** Between revision rounds, a panel that\nalready passes is not made more visually complex to fix nothing. Adding marks or\nlabels to a clean panel is a regression.\n\n---\n\n## §8 Anti-patterns\n\nThese are correctness failures, not style preferences:\n\n- Red and green as opposing categories.\n- Filled bars on a log-scaled value axis.\n- Colourbar ticks that are evenly spaced but miss the semantic centre.\n- A diverging colormap whose centre is the data midpoint, not the semantic zero.\n- An axis title that restates the tick labels.\n- Explaining the direction of goodness only in the caption.\n- A \"reference\" line drawn at a value that is itself one of the plotted points.\n- An excluded row that enters a plotted summary statistic.\n- A leader line whose nearest mark is not the row it labels.\n\n---\n\n## §9 Render-then-verify\n\nAfter `fig.savefig(...)`, before `save_artifacts`:\n\n**9.1 Geometric (bbox) check.**\n```python\nr = fig.canvas.get_renderer()\ntexts = [(t, t.get_window_extent(r)) for t in fig.findobj(mpl.text.Text)\n         if t.get_text().strip() and t.get_visible()]\nspines = [(s, s.get_window_extent(r)) for ax in fig.axes\n          for s in ax.spines.values() if s.get_visible()]\nticklabels = {ax: set(ax.get_xticklabels(which='both') + ax.get_yticklabels(which='both'))\n              for ax in fig.axes}\noverlaps  = [(a, b) for i, (a, ba) in enumerate(texts) for b, bb in texts[i+1:] if ba.overlaps(bb)]\noverlaps += [(t, s) for t, bt in texts for s, bs in spines\n             if bt.overlaps(bs) and t not in ticklabels[s.axes]]\n# assert: overlaps == [] and every text box lies within fig.bbox\n```\nOverlap is defined between *visible* boxes, and a tick label sitting on its own\nspine is not a finding. Fix (move, shorten, stagger) and re-save until clean.\n\n**9.2 Perceptual check.** The bbox check is geometric, not perceptual — it will\nnot catch a low-contrast label, a leader that crosses three others, or a series\ncolour mistakable for another. Crop the saved PNG to each panel and look:\n```python\nfig.savefig(\"figure.png\")\nfor letter, box in panel_crops(fig).items():\n    host.view_image(\"figure.png\", crop=box)\n```\nFor each crop: Is every glyph and mark legible against its background? Does the\nsmallest plotted element have a stroke or stub? Do any leaders cross? Could any\nseries colour be mistaken for another? Does the legend sit beside what it keys?\nA perceptual defect that passes §9.1 is still a defect.\n\n---\n*When in doubt: fewer hues, more direct labels, raw data over summary stats, and\nstate what is being measured before showing the result.*\n",{"name":5,"license":15,"description":7},"main","e71245365df78e06cd166602744bf2491f79910e","research-skills-hub\u002Fclaude-science-skills\u002Fskills\u002Ffigure-style",0,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS","https:\u002F\u002Fpengqianhan.github.io\u002FAI-Human-Research-OS\u002F","https:\u002F\u002Fgithub.com\u002Fpengqianhan","Ph.D. candidate of The University of Auckland, New Zealand","2026-11-14 18:08:16",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add pengqianhan\u002FAI-Human-Research-OS --skill figure-style",{"tool":39,"command":40},"bunx","bunx skills add pengqianhan\u002FAI-Human-Research-OS --skill figure-style",{"tool":42,"command":43},"pnpm","pnpm dlx skills add pengqianhan\u002FAI-Human-Research-OS --skill figure-style","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fclaude-science-skills\u002Fskills\u002Ffigure-style\u002FSKILL.md",true,[47],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"kernel.py","SCRIPT","脚本",13366,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fclaude-science-skills\u002Fskills\u002Ffigure-style\u002Fkernel.py",[54,62,69,76,83,90],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":58,"licenseFileUrl":6,"sourceHtmlUrl":59,"categories":60,"keywords":61,"publishedAt":19},"01M74MWYTHT498SVMN823896T7","workflow-skill-creator","Distills a completed user workflow or interaction into a reusable agent skill. Use when the user asks to turn their workflow, interaction, or multi-step process into a skill, or when they say \"make this a skill\", \"create a skill from what we just did\", \"package this workflow\" or similar. Do not use for creating skills from scratch without an existing workflow (use a generic skill-creator for that).","MIT","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fworkflow_skill_creator\u002FSKILL.md",[],[],{"skillNo":63,"name":64,"nameZh":6,"description":65,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":58,"licenseFileUrl":6,"sourceHtmlUrl":66,"categories":67,"keywords":68,"publishedAt":19},"01M74MWWVCFZ8HQS8840F1V1FT","uv","Checks whether the uv Python package manager is installed and installs it if missing. Ensures uv is on PATH. Use when another skill requires uv as a prerequisite.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fuv\u002FSKILL.md",[],[],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":58,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":19},"01M74MWRH5BTSWZ2C5PCTD5FV9","uniprot-database","Access protein metadata, function, taxonomy, and sequences across UniProtKB, UniParc, and UniRef. Use when searching for proteins, mapping identifiers, or retrieving functional annotations and publications. Don't use for sequence alignment, protein folding, or sequence similarity search (use specialized skills for those tasks).","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Funiprot_database\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":58,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M74MWN1D3ER9F1FZPWH8HQC1","unibind-database","Queries the UniBind database for experimentally validated transcription factor (TF) binding sites. Use when retrieving direct TF-DNA interaction datasets, downloading binding site coordinates (BED\u002FFASTA) for local analysis, or listing available datasets by species, cell line, or TF name. Don't use to query specific intervals, locations, genes, motif models or expression data.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Funibind_database\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":58,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M74MWHHB5QZY3NR2FYA1EK4N","ucsc-conservation-and-tfbs","Fetch Evolutionary Conservation scores (phyloP, phastCons) and Transcription Factor Binding Sites (TFBS) from the UCSC Genome Browser. Use when analyzing whether genomic variants or regions are evolutionarily conserved, functionally important, or bounded by TF regulators across major projects (ENCODE, JASPAR, ReMap).","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fucsc_conservation_and_tfbs\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":58,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":19},"01M74MWERYHSY87P21FDMVE6BA","string-database","Query the STRING database for protein-protein interactions (PPIs), functional enrichment, and homology. Use when the user asks about interactions between specific proteins, interaction evidence, confidence scores, protein interaction partners, or pathway enrichments.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fstring_database\u002FSKILL.md",[],[]]