figure-review

Colorblind reference

D="${CLAUDE_PLUGIN_ROOT:+$CLAUDE_PLUGIN_ROOT/skills/figure-review}"; D="${D:-$HOME/.claude/skills/figure-review}"; cat "$D/COLORBLIND.md" 2>/dev/null || echo "(colorblind guide not found)"

House style

D="${CLAUDE_PLUGIN_ROOT:+$CLAUDE_PLUGIN_ROOT/skills/figure-review}"; D="${D:-$HOME/.claude/skills/figure-review}"; cat "$D/CC-STYLE.md" 2>/dev/null || echo "(no house style configured — criterion 6 skipped)"

Inputs

Accept any subset of: figure image, plotting code, caption, surrounding text claim. Mark cant-assess for any criterion that requires input not provided.

Scope — establish this before applying any criterion

Determine each of the following from the inputs.

Standalone figure, or one panel of a composite? Look for siblings before deciding — list the directory, and if plotting code is an input, check whether one script saves several panels. Signals: a panel-style filename (fig2C.pdf, Fig_3B.pdf); a sibling differing by one letter. A panel delegates its legend and symbol definitions to the composite; do not flag those. It does not reliably delegate n or data source — a composite caption routinely covers several panels without mapping which is which — so keep checking them. When the composite is not itself an input, mark delegated items cant-assess; never pass them silently.

Draft, or published? Published means post-peer-review: apparent defects are far more likely to be considered choices, journal requirements, or context carried by the caption. For a published figure, raise concerns as by-design? rather than flag.

Deliberate-choice discipline

An author's considered tradeoff is not a defect. Before flagging, ask whether a competent author would plausibly have chosen this on purpose — journal or house constraints, panel economy, or fidelity to a figure being reproduced. If so, use by-design? and name why it might be intentional, so the author can confirm or overrule in one line.

When the author says a choice was deliberate, withdraw it immediately and do not re-raise it in any later pass.

Reproductions are out of scope for criteria 1–4. When a figure deliberately reproduces a published one, its choices are inherited rather than chosen. Review it for fidelity to the source instead, and say so.

Routing — this is what makes the verdicts above reachable. Criteria 1–6 emit flag on their own terms; they do not consider scope or intent. Before reporting, apply ## Scope and the rules above to every flag: it may be dropped (panel delegation), or reclassified by-design?. A criterion's bare "flag" is never the final verdict.

Criteria

1. Colormap Flag colormaps listed as unsafe in the colorblind reference above (jet, rainbow, red–green). Verify sequential vs diverging choice matches data type — a diverging map on an ordered non-diverging sequence (warming levels, time, dose) misreads as opposite signs, and a light midpoint can render a series member invisible. Assess luminance contrast. Warn if color is the sole encoding channel.

2. Uncertainty Distinguish two different objects before judging:

• Variability is the subject. A PDF, histogram, violin, box, or ensemble-spread plot already represents spread. pass — do not ask for a confidence band around a density.
• A point estimate or derived scalar is claimed (mean, trend, ratio, probability, count, fraction). Estimation uncertainty is then required: CI, error bars, ensemble range, or shading. Flag if absent. Significance masking does not satisfy this — stippling is a binary test result, and no interval is recoverable from it.

Before prescribing a band, check whether the inputs to compute one are present. If only aggregated output is available, report that instead of prescribing a fix.

3. Axes Labels and units on all axes. Ticks legible. Scale: for a positive-definite quantity, flag if plotted values span ≥2 orders of magnitude on a linear axis — and compute the span rather than asserting it (e.g. "daily rainfall 0.3–420 mm = 3.1 decades"). This test does not apply to signed anomalies (log is undefined) or to probability densities (a tail always →0, and log destroys the area comparison a density exists to convey). Flag a truncated axis or colorbar floor that renders small nonzero values identically to zero. Flag inconsistent limits across panels that invite cross-panel comparison.

4. Caption Judge self-containment at the level of the artifact under review (see ## Scope): defines all symbols, abbreviations, and line styles; states n, time period, data source. Never infer a caption's contents from the image.

5. Claim support The figure shows what the surrounding text asserts. Flag over-reach. Example over-reach: figure shows 2°C warming at one station; text claims "warming is accelerating across the region."

6. House style (only when --style is passed) Check colors against the CC palette above, graphic fonts (Effra/Bebas/Work Sans), and visual direction (minimal clutter, white space).

Report

Open with one line stating the scope determined above: standalone vs panel, draft vs published, and any assumption made.

[Criterion]: pass / flag / by-design? / cant-assess — [detail; if flagged, specific fix]

Example: Colormap: flag — Uses jet. Replace with viridis or cmo.thermal. Example: Colormap: by-design? — Grayscale only, no redundant line styles. Ignore if the journal requires grayscale.

Archive

Unless --no-archive was passed: after emitting the report, write it verbatim to <repo-root>/.ai/reviews/<YYYY-MM-DD>-figure-review[-<figure-slug>].md (mkdir -p "$(git rev-parse --show-toplevel)/.ai/reviews"; suffix -2, -3… on filename collision). Best-effort — if the cwd isn't a git repo or the write fails, add a one-line note and move on; never alter the review itself. Probe the ignore with the file path at repo root, not the bare directory: if git check-ignore -q "$(git rev-parse --show-toplevel)/.ai/reviews/probe.md" exits non-zero, warn and suggest adding .ai/ to .gitignore. (git check-ignore -q .ai false-negatives for the directory-form .ai/ pattern whenever the directory does not yet exist.)

Provenance header. Prepend a YAML block to the archived file only (never to the report shown to the user), so a report re-read months later is still interpretable — in particular, so you can tell whether it predates a revision of this skill's criteria:

---
skill: figure-review
skill-version: <toolkit version>
reviewed-repo: <repo basename> @ <short SHA>[ (dirty)]
cli: <claude --version>
date: <YYYY-MM-DD>
---

Collect the values with:

SD="${CLAUDE_PLUGIN_ROOT:+$CLAUDE_PLUGIN_ROOT/skills/figure-review}"; SD="${SD:-$HOME/.claude/skills/figure-review}"
cat "$SD/.version" 2>/dev/null || echo unknown    # skill-version ("unknown" under plugin install)
git rev-parse --short HEAD 2>/dev/null || echo unknown
git status --porcelain 2>/dev/null | head -1      # non-empty → append " (dirty)"
claude --version 2>/dev/null || echo unknown

.version is written into the deployed skill dir by sync.sh push; it is absent under a plugin install, which is expected. Resolve any value that fails to unknown — never drop the key, and never let a failed lookup block or alter the archive.

Does not

• Make aesthetic or branding calls beyond the six criteria.
• Rewrite plotting code unless explicitly asked after the report.