a-green-hand-jack/appendix-organizer

Appendix Organizer

Structure and write supplementary material so it supports the main paper without duplicating it. A well-organized appendix proves reproducibility, satisfies checklists, and handles reviewer questions without weakening the main paper's argument.

Use this skill when:

• the main paper is drafting or nearly complete and appendix structure needs to be decided
• a venue checklist (NeurIPS, ICLR, ICML) requires specific supplementary sections
• appendix sections have grown organically and need consistent cross-reference structure
• reviewer questions about implementation details, full ablations, or additional results need structured homes
• main-paper claims and appendix claims need boundary enforcement to avoid scope creep

Do not use this skill to write main paper sections — use paper-writing-assistant. Do not use this skill to plan main paper structure — use paper-writing-contract-planner.

Pair this skill with:

• paper-writing-contract-planner to ensure appendix scope is consistent with the main paper's writing contract
• paper-writing-assistant to draft prose for appendix sections after structure is decided
• paper-reviewer-simulator to anticipate which appendix sections reviewers will ask for
• submit-paper to verify the appendix is included and formatted correctly in the submission package
• artifact-evaluation-prep for artifact-evaluation appendix sections and reproduction instructions

Skill Directory Layout

<installed-skill-dir>/
├── SKILL.md
└── references/
    └── venue-checklists.md

Progressive Loading

• Always read references/venue-checklists.md when the target venue has a mandatory checklist.
• Read paper/.agent/writing-contract.md when it exists.
• Read paper/.agent/paper-evidence-board.md to identify which evidence needs appendix homes.
• Read the main paper draft when auditing boundary alignment.

Core Principles

• Appendix sections exist to prove claims and satisfy checklist requirements, not to dump everything that did not fit.
• Every appendix section must have a clear job: proof, ablation, implementation detail, reproducibility record, or checklist item.
• The main paper must be self-contained and readable without the appendix.
• Cross-references should go from main → appendix, not appendix → main (supplementary material does not require the main paper to reference back).
• Reviewer simulation should inform what goes in the appendix: what questions will reviewers ask?
• Appendix figures and tables need captions as carefully as main-paper ones.

Step 1 — Gather Context

Find:

• paper draft: paper/main.tex, paper/paper.tex, or paper/sections/
• existing appendix: paper/appendix.tex, paper/sections/appendix/, or \appendix in the main file
• writing contract: paper/.agent/writing-contract.md
• evidence board: paper/.agent/paper-evidence-board.md
• venue: from writing contract or user argument

Record:

• target venue and its checklist requirements
• which main-paper sections already have appendix cross-references
• which claims have evidence that was deferred to the supplement
• which ablations were run but not included in the main paper

Step 2 — Select Mode

• plan: design the appendix structure from scratch or from a partial draft
• draft: write one or more appendix sections after structure is decided
• audit: check existing appendix for boundary violations, missing sections, and broken cross-references
• checklist: fill in or verify a venue-specific reproducibility/ethics checklist

Step 3 — Design Appendix Structure

Read references/venue-checklists.md for the target venue.

For each appendix section, record:

Section: <letter/title>
Job: proof / ablation / implementation-detail / additional-results / reproducibility / checklist / ethics
Main-paper claim it supports: <CLM-XXX or section reference>
Content summary: <1-2 sentences>
Status: planned / drafted / complete / deferred
Cross-reference in main paper: <section and line/paragraph>

Standard appendix section types for ML papers:

• Proofs: formal proofs of claims or theorems stated in the main paper
• Implementation details: architecture hyperparameters, training schedules, optimizer configs, hardware
• Additional ablations: variants that support the main result but would clutter the paper
• Additional results: more datasets, longer training, extra baselines, per-class breakdown
• Dataset details: data sources, statistics, preprocessing steps, license info
• Reproducibility statement: seeds, code, compute cost, expected variance
• Broader impact / limitations: required by some venues even if not in main paper
• Checklist sections: NeurIPS reproducibility, ICLR checklist, artifact evaluation

Step 4 — Enforce Claim Boundaries

For each appendix section, check:

• Does the appendix section make a claim that is stronger than the main paper's claim? If so, align the wording or move the claim to the main paper.
• Does the appendix introduce a new method or contribution not in the main paper? Flag for review — this is usually a scope problem.
• Are all cross-references from the main paper pointing to existing appendix sections? Fix broken references.
• Are appendix figures and tables cited in the main paper or in the appendix section that uses them?

Step 5 — Fill Venue Checklist

Read references/venue-checklists.md for the target venue.

For each required checklist item, classify as:

• yes: satisfied with a pointer to where
• no: not applicable, with reason
• partial: needs work, with a specific action
• na: explicitly not applicable

Do not leave checklist items blank. Reviewers treat unanswered items as potential red flags.

Step 6 — Write or Audit Appendix Sections

When drafting:

• Write each appendix section with a clear topic sentence that states its job.
• Use the same notation, terminology, and symbol conventions as the main paper.
• Refer back to the main paper section being supplemented at the start of each section.
• Captions for appendix figures/tables must be self-contained.

When auditing:

• Flag redundancy: content that repeats the main paper without adding proof, detail, or results.
• Flag scope creep: new claims or methods that belong in the main paper.
• Flag missing sections: evidence deferred to the appendix that has not been written.
• Flag broken cross-references.

Memory Writeback

• Update paper/.agent/paper-evidence-board.md when appendix sections fill evidence slots
• Update paper/.agent/writing-contract.md if appendix scope changes the main paper's claim boundaries

Final Sanity Check

Before finishing:

• every appendix section has a documented job and a main-paper connection
• venue checklist items are all answered
• no appendix section introduces a claim stronger than the main paper's allowed wording
• all main-paper \appendix cross-references point to actual sections
• appendix figures and tables have self-contained captions