What problem does it solve? Academic poster generation often produces unprintable results: fabricated figures, dozens of clashing colors, misaligned columns, and endless cosmetic patch loops. This Skill replaces eyeballing with hard measurement gates, a closed fix vocabulary, and provenance-checked real paper figures, producing a print-ready PDF on the first disciplined pass. ## Core Features & Use Cases - Measurement-gated pipeline: preflight, style, asset, measure, and polish gates run in canonical order via run_gates.py, enforcing column alignment, canvas fill, two-hue palette discipline, and figure resolution before any visual review. - Real figure provenance: extract_pdf_figures.py crops figures from the paper PDF at 300-450 DPI and FIGURE_MANIFEST.json tracks sha256, page, and bbox so every image traces to the source paper. - Cross-model review loop: a Claude visual rubric with critical caps plus a fresh codex final review audit claims against the paper, with a closed fix vocabulary preventing patch-loop regressions. - Use Case: Given an ICLR paper PDF, produce a 185x90cm landscape poster with venue-correct canvas, real figures, MathJax equations, and a verified single-page PDF under 20MB. ## Quick Start Ask the agent to build a conference poster from your paper PDF by invoking the paper-poster-html skill with the paper path and venue, for example: create a poster for paper.pdf targeting ICLR with a 185x90cm landscape canvas.
