Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Organize messy conference LaTeX template .zip files into clean Overleaf-ready structure. Extracts, analyzes, cleans up, and generates README with submission requirements. Triggers on "organize LaTeX template", "clean up template", "prepare Overleaf template", "整理LaTeX模板".
.claude/skills/catlog22-scholar-latex-organizer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 774% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 73% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 158% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 24% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 20% | 0% |
Transform messy conference LaTeX template .zip files into clean, Overleaf-ready submission templates. Official templates often contain excessive examples and disorganized files — this skill converts them into templates ready for writing.
.workflow/codebase/ARCHITECTURE.md exists, read for project contextmaestro load --type spec --category coding — load coding conventionsmaestro search "academic writing research paper" --json — top 5 entries as prior contextUser: "Organize this LaTeX template"
|
v
┌──────────────────────────────────────────────────────────────┐
│ SKILL.md (Orchestrator) │
│ Collect preferences → Dispatch phases → Track progress │
└──────────┬───────────────────────────────────────────────────┘
|
┌───────┼───────────┬──────────────┐
v v v v
┌──────┐┌──────────┐┌──────────────┐
│ P1 ││ P2 ││ P3 │
│Extract││ Cleanup ││ README & │
│ & ││ & ││ Finalize │
│Analyze││ Organize ││ │
└──┬───┘└────┬─────┘└──────┬───────┘
│ │ │
v v v
analysis clean output/
+ plan structure (Overleaf-ready)Before dispatching to any phase, collect these preferences:
Questions to ask the user:
1. Template Source
"Where is the template .zip file? (path)"
→ templatePath
2. Conference Info
"Conference name or submission page URL (for extracting requirements)"
→ conferenceInfo
3. Output Directory
"Where should the organized template go? (default: ./output/)"
→ outputDir
4. Section Structure
Options: Standard (intro/related/method/experiments/conclusion) | Custom
→ sectionStructureStore responses as organizerPreferences context for all phases.
When workflowPreferences.autoYes === true:
> COMPACT DIRECTIVE: Context compression MUST check TodoWrite phase status. > The phase currently marked in_progress is the active execution phase — preserve its FULL content. > Only compress phases marked completed or pending.
LaTeX Template Organization:
- [ ] Phase 1: Extract & Analyze — extract zip, identify files, diagnose issues
- [ ] Phase 2: Cleanup & Organize — create structure, clean main.tex, copy assets
- [ ] Phase 3: README & Finalize — generate README, verify compilationPhase 1: Extract & Analyze
└─ Ref: phases/01-extract-analyze.md
├─ Input: templatePath, conferenceInfo
└─ Output: analysisResult (file inventory, main file, issues, conference type)
Phase 2: Cleanup & Organize
└─ Ref: phases/02-cleanup-organize.md
├─ Input: analysisResult, sectionStructure, outputDir
└─ Output: organized template in outputDir/
Phase 3: README & Finalize
└─ Ref: phases/03-readme-finalize.md
├─ Input: analysisResult, conferenceInfo, outputDir
└─ Output: README.md + final verificationPhase Reference Documents (read on-demand when phase executes):
| Phase | Document | Purpose | Compact | |-------|----------|---------|---------| | 1 | phases/01-extract-analyze.md | Extract and analyze template | TodoWrite driven | | 2 | phases/02-cleanup-organize.md | Clean up and restructure | TodoWrite driven + sentinel | | 3 | phases/03-readme-finalize.md | Generate README, finalize | TodoWrite driven |
Compact Rules:
in_progress → preserve full content, do not compresscompleted → may compress to summaryRead() to recover| Conference | Document Class | Key Config | |------------|---------------|------------| | KDD (ACM) | acmart | nonacm option for anonymous submission | | ACM Conferences | acmart | anonymous,review options | | CVPR/ICCV | cvpr | Two-column, strict page limits | | NeurIPS | neurips_20XX | Anonymous review | | ICLR | iclr20XX_conference | Two-column | | AAAI | aaaiXX | Two-column, 8 pages + refs |
Phase 1 ──analysisResult──→ Phase 2
Phase 1 ──analysisResult──→ Phase 3
Output structure:
outputDir/
├── main.tex (cleaned main file)
├── text/ (section files)
│ ├── 01-introduction.tex
│ ├── 02-related-work.tex
│ ├── 03-method.tex
│ ├── 04-experiments.tex
│ └── 05-conclusion.tex
├── figures/ (image files)
├── tables/ (table files + example)
├── styles/ (.sty/.cls files)
├── references.bib (bibliography)
└── README.md (usage guide)| Error | Action | |-------|--------| | Zip extraction fails | Check file integrity, ask user to re-download | | Main file not identifiable | List all .tex files, ask user to choose | | Missing dependency (.sty/.cls) | Warn user, suggest CTAN download | | Conference info unavailable | Fall back to template comments + documentclass detection | | Compilation fails after cleanup | Revert to minimal cleanup, report what needs manual fix |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 20,604 | 7,175 | -65% | 1 | 1 | 0% | 3,823 | 2,072 | -46% | 0 | 0 | — |
case-02 | fail→pass | 5,765 | 7,089 | +23% | 1 | 1 | 0% | 328 | 2,868 | +774% | 0 | 0 | — |
case-03 | fail→fail | 17,902 | 7,233 | -60% | 1 | 1 | 0% | 3,313 | 2,465 | -26% | 0 | 0 | — |
case-04 | pass→pass | 11,908 | 11,436 | -4% | 1 | 1 | 0% | 1,726 | 3,221 | +87% | 0 | 0 | — |
case-05 | fail→fail | 2,571 | 4,162 | +62% | 1 | 1 | 0% | 280 | 2,271 | +711% | 0 | 0 | — |
case-06 | pass→fail | 17,952 | 6,073 | -66% | 1 | 1 | 0% | 2,753 | 1,993 | -28% | 0 | 0 | — |
case-07 | fail→fail | 19,088 | 10,988 | -42% | 1 | 1 | 0% | 2,622 | 3,337 | +27% | 0 | 0 | — |
case-08 | pass→pass | 10,319 | 6,860 | -34% | 1 | 1 | 0% | 1,583 | 2,774 | +75% | 0 | 0 | — |
case-19 | fail→pass | 9,561 | 3,152 | -67% | 1 | 1 | 0% | 1,302 | 2,257 | +73% | 0 | 0 | — |
case-09 | fail→pass | 6,303 | 4,942 | -22% | 1 | 1 | 0% | 996 | 2,568 | +158% | 0 | 0 | — |
case-10 | pass→pass | 5,980 | 5,210 | -13% | 1 | 1 | 0% | 932 | 2,473 | +165% | 0 | 0 | — |
case-11 | fail→fail | 4,659 | 5,145 | +10% | 1 | 1 | 0% | 687 | 2,513 | +266% | 0 | 0 | — |
case-12 | fail→pass | 15,760 | 10,348 | -34% | 1 | 1 | 0% | 2,683 | 3,321 | +24% | 0 | 0 | — |
case-13 | fail→pass | 14,891 | 5,883 | -60% | 1 | 1 | 0% | 2,160 | 2,592 | +20% | 0 | 0 | — |
case-14 | pass→pass | 11,577 | 6,096 | -47% | 1 | 1 | 0% | 1,748 | 2,605 | +49% | 0 | 0 | — |
case-15 | fail→pass | 11,398 | 7,266 | -36% | 1 | 1 | 0% | 1,652 | 2,906 | +76% | 0 | 0 | — |
case-16 | pass→pass | 13,103 | 7,391 | -44% | 1 | 1 | 0% | 1,968 | 2,878 | +46% | 0 | 0 | — |
case-17 | fail→pass | 11,922 | 7,805 | -35% | 1 | 1 | 0% | 1,670 | 2,819 | +69% | 0 | 0 | — |
case-18 | fail→fail | 12,164 | 10,892 | -10% | 1 | 1 | 0% | 1,878 | 3,211 | +71% | 0 | 0 | — |
case-20 | pass→pass | 11,395 | 6,371 | -44% | 1 | 1 | 0% | 1,743 | 2,739 | +57% | 0 | 0 | — |
case-21 | fail→fail | 6,812 | 1,885 | -72% | 1 | 1 | 0% | 881 | 1,974 | +124% | 0 | 0 | — |
case-22 | fail→pass | 13,798 | 6,188 | -55% | 1 | 1 | 0% | 2,271 | 2,802 | +23% | 0 | 0 | — |
case-23 | fail→pass | 14,046 | 3,080 | -78% | 1 | 1 | 0% | 2,094 | 2,190 | +5% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 20 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +35 percentage points is the difference between those two pass rates over the 20 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.