Figure Advisor

For each figure in figure_todo.md, examine the rasterized image and check:

• Axis labels present and readable
• Legend present if multiple series
• Color-blind safe palette (avoid red+green only)
• Line width sufficient (≥1.5pt)
• Caption explains the takeaway, not just "Results"
• No JPEG artifacts on line plots (use PDF/vector format)

Step 3 · For each figure, choose action

Data-driven plots (training curves, bar charts, tables, heatmaps): Generate new matplotlib code and save to figures/fig_N_revised.py:

# Example structure
import matplotlib.pyplot as plt
import numpy as np

# Use ONLY numbers already in the paper
# Never invent data points

plt.rcParams['font.size'] = 11
# ... plot code using existing paper data ...
plt.savefig('figures/fig_N_revised.pdf', bbox_inches='tight')

Run the script: python3 figures/fig_N_revised.py Verify output exists: ls figures/fig_N_revised.pdf

Architecture/pipeline diagrams: Invoke /paper-illustration for this figure: Use Skill: /paper-illustration "[figure description from figure_todo.md]"

Minor fixes (axis labels, captions, colors): Write specific LaTeX/matplotlib instructions in figure_report.md.

Step 4 · Write figure_report.md

# Figure Revision Report

## Figure 1 · [caption excerpt]
**Issues**: color-only distinction, no axis label on Y
**Action taken**: generated figures/fig_1_revised.py
**Run**: `python3 figures/fig_1_revised.py`
**Suggestion**: unified tSNE panel combining Figures 1 and 3

## Figure 2 · [caption excerpt]
**Issues**: routed to /paper-illustration
**Action taken**: invoked /paper-illustration for architecture diagram

## Author Todo
- Fig 3 requires new ablation data → added to author_todo.md

Step 5 · Output files

• figure_report.md — per-figure analysis and actions
• figures/fig_N_revised.py — new plot scripts (data-driven only)
• author_todo.md — updated with any figure items needing new data

Print when done:

=== Figure Advisor Complete ===
✅ Generated: N plot scripts → figures/fig_N_revised.py
🎨 Routed to /paper-illustration: N diagrams
⚠️  Author action needed: author_todo.md (N items)
📋 Report: figure_report.md
```37:["$","$L3f",null,{"content":"$40","frontMatter":{"name":"figure-advisor","description":"Analyze and improve figures in an academic paper based on reviewer comments in figure_todo.md. Gives improvement suggestions and generates new matplotlib code for data-driven plots. Invokes /paper-illustration for architecture diagrams. Use when figure_todo.md exists, or when the user says \"improve figures\", \"修改图表\", \"analyze figures\", or after /writing-reviser completes.\n","argument-hint":["figure_todo.md + paper.pdf"],"allowed-tools":"Bash(*), Read, Grep, Glob, Write, Edit, Skill"}}]