AgentSkillsCN

tex-figure-table-section-fix

当需要针对图表、表格或命名章节进行精准的质量修复时使用此功能，包括先验证再修复、限定范围的补丁修复，以及可选的视觉验证环节。

SKILL.md
--- frontmatter
name: tex-figure-table-section-fix
description: Use when targeted quality fixes are needed for figures, tables, or named sections, including verify-first remediation, scoped patching, and optional visual verification.

Tex Figure Table Section Fix

Overview

Use this skill for localized quality remediation. Keep scope tight: one target area at a time, evidence first, patch second.

Executable Entry Points

  • •scripts/verify_content_targets.py: verifies one target scope at a time (figures, tables, or section) and outputs JSON issues.

Example commands:

bash
python scripts/verify_content_targets.py --project-root . --main-tex main.tex --target figures --pretty
python scripts/verify_content_targets.py --project-root . --main-tex main.tex --target tables --pretty
python scripts/verify_content_targets.py --project-root . --main-tex main.tex --target section --section-name Introduction --pretty

Project Adaptation

  1. •Identify commands or scripts used to run figure/table/section checks.
  2. •Confirm local conventions (caption length, label format, booktabs policy, section style requirements).
  3. •Confirm whether visual verification is available (PDF render + vision review).

Workflow

  1. •Choose one target scope:
    • •figures
    • •tables
    • •specific section
  2. •Run verify-first checks via scripts/verify_content_targets.py and capture explicit issues.
  3. •Typical figure/table checks:
    • •missing labels/captions
    • •unreferenced labels
    • •overflow width or placement issues
    • •table formatting problems (for example missing booktabs style)
  4. •Typical section checks:
    • •weak structure or flow
    • •unsupported claims / missing citations
    • •TODO markers and placeholder text
  5. •Generate minimal unified diffs with precise context.
  6. •Apply through approval flow.
  7. •If layout risk remains, run visual verification loop and patch only confirmed issues.
  8. •Re-analyze and stop when residual issues are acceptable.

Inputs and Outputs

  • •Input: parsed structure facts, target scope, optional user instruction.
  • •Output: scoped patch set and post-fix quality findings.

Common Mistakes

  • •Skipping verify and jumping straight to free-form edits.
  • •Combining figures/tables/sections into one broad patch batch.
  • •Ignoring visual verification for layout-sensitive issues.
  • •Continuing auto-fix after repeated quality regression.