메뉴
Visual (VLM) and array-completeness check for plots and figure files. Use proactively after savefig or export; before citing a figure in a paper or report. Axis styling and export settings → scientific-plotting; domain physics → physics-check.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | plot-check |
| description | Visual (VLM) and array-completeness check for plots and figure files. Use proactively after savefig or export; before citing a figure in a paper or report. Axis styling and export settings → scientific-plotting; domain physics → physics-check. |
Inspect the figure file and, when available, the arrays the plotting script used. Do not infer quality from file size, timestamps, or log lines alone.
VLM alone is not enough for curves and scatter plots. A figure can look smooth
while regions, bins, or series are missing. Run Step 2 when numeric artefacts exist;
never mark OK on visuals only in that case.
savefig, PDF/PNG/SVG export).
PNG: always dpi=300 (see scientific-plotting).Open the figure with the Read tool (PNG/PDF at roughly final print size).
Verify:
Data completeness (VLM): for every line, curve, scatter cloud, or error-bar series:
If incomplete or ambiguous → run Step 2. If arrays are unavailable, verdict is at
most WARN with "completeness not verified numerically".
Project-specific visual bands (reference curves, expected ranges) → repo CLAUDE.md.
Load the arrays the plotting script used. Count before concluding the VLM pass.
import numpy as np
x, y = np.asarray(x).ravel(), np.asarray(y).ravel()
assert x.size == y.size, "x and y length mismatch"
n_total = x.size
plottable = np.isfinite(x) & np.isfinite(y)
n_plotted = int(plottable.sum())
n_dropped = n_total - n_plotted
print(f"total={n_total} plottable={n_plotted} dropped(non-finite)={n_dropped}")
assert n_dropped == 0, f"{n_dropped} points dropped — check mask/NaN handling"
| Check | Fail when |
|---|---|
| Expected count | Caption/manifest says N but n_plotted < N |
| Shared abscissa | Compared curves differ in x length or finite mask |
| NaNs | Non-finite values not documented in caption |
| Histogram | Occupied bins ≪ expected without explanation |
| 2D / image | Most pixels non-finite without stated mask |
| Replot spot-check | Re-plot (x, y) and diff against saved PNG if unsure |
Step 1 OK but Step 2 shows drops → BLOCKING. Scientific plot without arrays → WARN
unless user confirms completeness.
Path from CLAUDE.md. Confirm git_hash, data_paths, seed, script/command
match what produced the figure.
Exactly one of:
OK: <one-line summary>WARN: <summary>; <what to investigate>BLOCKING: <reason>; <fix>Record in research-figure-manifest or source comment. Then run physics-check on saved arrays when they exist.
| Skill | Role |
|---|---|
| scientific-plotting | Export before Step 1 |
| physics-check | Numeric/domain asserts after plot looks complete |
| vlm-figure-audit | Print-size QA; defer final OK on curves until plot-check passes |
| research-figure-manifest | Store plot_check verdict |
| paper-writing-workflow | Manuscript build loop references this category |
Deep-read an academic paper and produce a durable, critical research note — not a loose summary. Use when given a paper title, DOI, arXiv ID, publisher/PDF/GitHub URL, or local PDF, or when asked to read closely, summarize, extract contributions, analyze experiments, critique, compare to related work, reproduce results, or maintain a paper-reading repository index. Works across fields (astro/physics, ML, bio, security, …); examples lean astro-ph but the workflow is general.
Remote, slice-level access to FLAMINGO HDF5 files on COSMA via the `hdfstream` Python module — no need to download whole snapshots. Covers install, server connection, navigation, dataset slicing, multi-slice batched reads, and integration with swiftsimio. Use when reading any FLAMINGO data product remotely.
FLAMINGO past-lightcone products: virtual observers, particle lightcones, HEALPix shell maps, integrated_maps, and halo lightcones. Use when building mock surveys, stacking Compton-y or lensing maps, or cross-correlating with observations.
Review code changes against requirements and quality standards before merge. Use when dispatching a review subagent, after major features, before merging, or when the user asks for a code review. Requires reading the actual diff.
Style and structure for research manuscript text: IMRaD, hourglass narrative, formal tone, topic sentences, reproducible Methods, captions that tell the story. Use when drafting or editing titles, abstracts, sections, captions, or reports.
Iterative VLM review of compiled LaTeX PDFs for layout clashes (text vs figures, tables, long equations). Use after the first compile or any float/equation edit; loop fix-recompile-recheck until no clashes remain.