AAAI Revision Detemplater

A Codex skill for rebuilding AAAI revision packages into cleaner, more defensible submissions.

This skill is aimed at a specific failure mode that shows up in real paper revisions:

• the story becomes over-templated;
• contributions are forced into artificial symmetry;
• critical content gets pushed into the supplement;
• abstracts and TL;DR fields become numerically dense;
• conclusions sound stronger than the actual control or robustness evidence;
• authors rerun packages unnecessarily instead of reusing valid frozen artifacts.

aaai-revision-detemplater fixes that layer.

What it does

The skill helps Codex:

• rebuild the paper's narrative spine from the real review question;
• reduce template feel in titles, contributions, section names, and experiment framing;
• decide what must stay in the main paper versus what belongs in supplementary material;
• tighten title, abstract, keywords, TL;DR, and OpenReview registration text;
• soften over-strong claims so they match the actual evidence scope;
• prefer package reuse over unnecessary recomputation.

Why this is different

Most paper-writing prompts optimize for fluent prose.

This skill optimizes for revision quality under AAAI constraints:

• reviewer-facing clarity;
• evidence-calibrated claims;
• main-paper-first content allocation;
• lower "AI smell" from mirrored structure and repetitive rhetorical framing.

Best use cases

Use it when you already have:

• a working AAAI .tex submission;
• figures, tables, and experiment artifacts;
• a supplementary document or upload package;
• reviewer comments or internal critique;
• a draft that feels technically solid but rhetorically too uniform.

Install

Codex reads personal skills from $CODEX_HOME/skills when CODEX_HOME is configured, otherwise from ~/.codex/skills. Clone this repository into that directory. The GitHub repository is named AAAI-Writer, but the destination folder must be aaai-revision-detemplater so it matches the skill name.

The commands below use the default ~/.codex/skills location.

PowerShell:

New-Item -ItemType Directory -Force "$HOME\.codex\skills" | Out-Null
git clone https://github.com/cxcx52/AAAI-Writer.git "$HOME\.codex\skills\aaai-revision-detemplater"

macOS or Linux:

mkdir -p "$HOME/.codex/skills"
git clone https://github.com/cxcx52/AAAI-Writer.git "$HOME/.codex/skills/aaai-revision-detemplater"

Open a new Codex task after installation so the skill catalog is refreshed.

Invoke

Example prompt:

Use $aaai-revision-detemplater to rebuild this AAAI paper and supplement while reusing valid experiment artifacts.

For a useful first pass, provide the main .tex entry point, supplementary source, current PDF or build folder, reviewer comments if available, and any existing upload ZIPs. State the page budget and whether experiment artifacts are frozen.

See 中文使用说明 and the copyable example request.

Expected output

A package-level run returns:

• a revision diagnosis and one-sentence narrative spine;
• specific main-paper and supplement moves;
• title, abstract, keyword, TL;DR, and OpenReview edits when present;
• an artifact-reuse plan that identifies what can remain frozen;
• residual claim, page-budget, and upload-package risks.

Repository contents

• SKILL.md: core workflow and operating boundaries
• agents/openai.yaml: UI metadata and default invocation prompt
• references/narrative-rebuild.md: how to reconstruct the story
• references/main-vs-supplement.md: main-paper versus supplement allocation
• references/artifact-governance.md: fact sheets, claim boundaries, and package records
• references/upload-audit.md: source ZIP, PDF preview, checksum, and portal checks
• references/metadata-surfaces.md: abstract, TL;DR, title, and keywords
• references/evidence-calibration.md: how to avoid overclaiming
• docs/usage_zh.md: concise Chinese usage guide
• examples/example_input.md: realistic request template
• CHANGELOG.md: public release history
• LICENSE: MIT license

Scope

This skill revises arguments, prose, source layout, metadata, and submission packaging. It does not invent evidence, guarantee acceptance, or rerun frozen experiments unless a requested change invalidates the existing artifacts.