- Home
- Skills
- Research & Analysis
- Publication-Quality Figures for Machine Learning Papers
More screenshots
Works with the AI Agents you already use
Publication-Quality Figures for Machine Learning Papers
Turn research results into clear, consistent figures: one claim per figure, the right chart form for each comparison, a shared style contract, and a reviewer checklist for every figure.
$5
Publication-Quality Figures for Machine Learning Papers
Example session with this skill installed
I have a paper on a fictional image classifier called Fernlight, tested on a fictional benchmark called Meadow-20. I need three figures: a training curve showing Fernlight reaching the baseline's final accuracy sooner, an ablation showing which component matters most, and an architecture diagram with four stages. Invent plausible numbers and give me the figure plan, the style contract, and the review outcome for each figure.
- Read your context and instructions
- Compiled the publication-quality figures
FIGURE PLAN — Fernlight on Meadow-20, invented numbers for this demo only.
Fig 1, line plot: Claim — Fernlight reaches the baseline's final accuracy in half the steps. Evidence — accuracy per step, 3 seeds each, Fernlight vs two baselines. Numbers: Fernlight 84% at step 5,000; stronger baseline 84% at step 10,000. Shaded bands show seed spread; axes "Training steps" and "Accuracy (%)".
Fig 2, grouped bars sorted by effect: full Fernlight 84%, without the gating module 77%, without the second stage 81%, without the third stage 83%. Y-axis starts at zero; full model leftmost as anchor.
Fig 3, block diagram: four boxes left to right — Intake, Gating, Refinement, Output — arrows labeled with what passes between stages, one accent color, no 3D effects.
STYLE CONTRACT: one sans family, 9 point minimum at column width; four data colors plus gray, colorblind-safe, Fernlight in the accent color and baselines in gray in every figure; legends outside the data region; files named fig1-fernlight-curve.pdf plus a 300 dpi review PNG, script kept beside each chart.
REVIEW: Fig 1 passed all ten checklist items. Fig 2 first failed the caption check — "Ablation results" stated nothing — and was rewritten to state setup, seed count, and takeaway. Fig 3 first failed the grayscale check — Gating and Refinement became identical in gray — and the palette was adjusted with different markers added. All three ship only after passing.
Connects securely to your tools. The creator never sees your data.
About this skill
A figure is an argument made visible, and this skill treats it that way from the start. Work begins with a figure brief for every figure: one claim, the exact evidence behind it, what the reader must conclude in ten seconds, and the form — concept diagram or data chart, never both at once. The chart form is picked from the comparison, not from habit: line plots for trajectories over steps, sorted grouped bars for ablations, scatter with a frontier for accuracy-versus-cost trade-offs, violin or box plots for distributions, and annotated heatmaps for confusion or correlation structure. Two pipelines stay separate: concept diagrams are generated from a structured four-block prompt and checked label by label, while data charts are scripted from the real numbers, with the generating script kept beside the figure so every number traces to its source. One style contract governs the whole paper — fonts, a colorblind-safe palette with the same color for the same method everywhere, axes with units, legends that never cover data, and a caption template that states what is shown, how it was computed, and the takeaway, in that order. Every figure finishes with a ten-item review checklist covering grayscale legibility, spread shown for any claim of better, and exact agreement with the tables. Figures that fail are reworked or cut, not shipped.
- It cannot create figures from numbers you do not supply, and it never invents or smooths data to fill a gap.
- Generated diagram text often needs manual correction; every label must be checked before submission.
- It assumes a plotting library is available for data charts; without one, it produces the plan and prompts, not finished charts.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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 3 days ago
Permissions
No special permissions declared or detected