Paper2Poster Poster

Use this skill when the goal is to turn a paper into a readable, credible, poster-ready package instead of a long prose summary.

Core behavior

Work from the source paper first. Prefer grounded extraction over creative rewriting.

Do not invent:

• metrics that do not appear in the paper
• figure content you have not seen
• ablation claims that are not supported
• author affiliations, funding, or venue details unless supplied

If key information is missing, say so explicitly and continue with labeled assumptions.

Default workflow

1. Gather inputs.

Collect the paper source plus any constraints that are already available:

• paper.pdf, arXiv link, markdown draft, or pasted paper text
• venue, audience, poster dimensions, branding, deadline
• whether the user wants poster copy only or also Paper2Poster execution assets

If the user gave no dimensions, use the defaults in references/poster-defaults.md.

2. Choose the output mode.

• Draft mode: produce poster copy, layout guidance, figure recommendations, and a concise runbook.
• Execution mode: additionally prepare a poster.yaml draft, a Paper2Poster folder layout, and the exact command the user can run.

3. Extract poster-worthy content from the paper.

Capture:

• title and subtitle candidate
• one-sentence paper claim
• problem and motivation
• method overview
• core figures, tables, and diagrams worth featuring
• strongest quantitative results
• limitations, deployment notes, or future work if they matter

4. Compress for poster reading.

Rewrite for scanability:

• short headers
• bullets over paragraphs
• one idea per bullet
• clear result statements with units and baselines when available
• captions that can stand alone

5. Build the poster package.

Unless the user asks for a different structure, produce:

• poster_brief.md: audience, poster goal, assumptions, content priorities
• poster_copy.md: final section copy ready for layout
• poster_layout.md: panel plan, figure placement, hierarchy, and visual notes
• poster_runbook.md: exact next steps for Paper2Poster or manual assembly

If execution mode is requested, also produce poster.yaml.

6. Finish with a quality pass.

Check that:

• the poster has a visible narrative from problem to result
• text is shorter than the source summary would naturally be
• the main figure is obvious
• every number can be traced back to the source material
• there is a clear takeaway and contact/QR placeholder if relevant

Paper2Poster execution guidance

When the user wants a repo-ready Paper2Poster setup, follow the upstream folder and command pattern:

• place the paper at {dataset_dir}/{paper_name}/paper.pdf
• place a per-poster poster.yaml next to paper.pdf if custom styling is needed
• use python -m PosterAgent.new_pipeline with poster path, text model, vision model, and poster dimensions

Common options to surface when relevant:

• --poster_width_inches
• --poster_height_inches
• --conference_venue
• --institution_logo_path
• --conference_logo_path
• --use_google_search
• --max_workers

If the environment is not ready to run Paper2Poster, still provide the exact command and clearly mark it as unexecuted.

Output style

Prefer confident but compact deliverables.

• Keep section headers short.
• Prefer bullets to paragraphs.
• Preserve technical meaning.
• Surface assumptions at the top, not buried at the end.
• If the paper is dense, move secondary details into a small "Backup / Appendix" section suggestion rather than bloating the main poster.

When to read references

• Read references/inputs-and-deliverables.md when you need a checklist for missing materials or expected outputs.
• Read references/poster-defaults.md when dimensions, panel count, copy density, or layout choices are unspecified.
• Read references/example-prompts.md when the user wants a ready-to-paste invocation for Codex or Claude.