What it does. DEFAULT poster pipeline — build an academic conference poster (ICML/NeurIPS/ICLR/CVPR/...) as a single HTML/CSS file with measurement-driven hard gates, real paper figures, a two-hue design-token system, and print-ready PDF via headless Chromium. Use when the user says "做海报", "poster", "conference poster", "paper poster", or asks to design/redo a research poster. Supersedes the retired LaTeX /paper-poster. Part of ARIS: Auto-claude-code-research-in-sleep (wanshuiyin/Auto-claude-code-research-in-sleep).

Upstream	wanshuiyin/Auto-claude-code-research-in-sleep
Skill file	skills/paper-poster-html/SKILL.md
License	MIT
Author	wanshuiyin
Fetched	2026-09-10

Install

• Clone the repo and run bash tools/install_aris.sh, or copy skills/paper-poster-html/ into ~/.claude/skills/paper-poster-html/; npx skills add wanshuiyin/Auto-claude-code-research-in-sleep --skill paper-poster-html also works.
• Raw file: curl -sL https://raw.githubusercontent.com/wanshuiyin/Auto-claude-code-research-in-sleep/HEAD/skills/paper-poster-html/SKILL.md

SKILL.md (verbatim)

name: paper-poster-html
description: "DEFAULT poster pipeline — build an academic conference poster (ICML/NeurIPS/ICLR/CVPR/...) as a single HTML/CSS file with measurement-driven hard gates, real paper figures, a two-hue design-token system, and print-ready PDF via headless Chromium. Use when the user says \"做海报\", \"poster\", \"conference poster\", \"paper poster\", or asks to design/redo a research poster. Supersedes the retired LaTeX /paper-poster."
argument-hint: "[paper-dir-or-pdf] [— venue: ICLR, canvas: 185x90cm landscape, venue-colors: true]"
allowed-tools: Bash(*), Read, Write, Edit, Grep, Glob, WebFetch, WebSearch, AskUserQuestion, mcp__codex__codex

Paper Poster (HTML): measurement-gated poster generation

One HTML file styled for an exact print canvas (@page { size: W H }), rendered to PDF via Playwright print emulation. Iterate by measuring, not eyeballing — the screen preview lies; only print emulation at the correct viewport tells the truth. Core gate machinery is adapted from posterly (MIT, © 2026 Ruishuo Chen — see NOTICE.md and LICENSES/posterly-MIT.txt); ARIS adds style discipline gates, figure-provenance gates, the cross-model review loop, and the anti-patch-loop fix vocabulary.

Why this skill exists (the failure it prevents)

A predecessor pipeline produced a poster with 30+ colors, zero real paper figures, a screen-pixel canvas, and tiny formulas floating in oversized boxes, then spent 12+ review rounds making it worse — each round added a new badge color or bespoke SVG patch. The cure is structural, not exhortative:

1. Hard gates run before any aesthetic opinion (alignment, style, assets must PASS first — a reviewer never sees an unmeasured poster).
2. A closed fix vocabulary — visual-review fixes can only touch design tokens, whole catalogued components, content rebalance, assets, or canvas choice. New inline styles / new hex values / bespoke decorations are structurally forbidden.
3. Two-hue discipline as a machine check, not a style suggestion.
4. Real paper figures with provenance manifest, or the gate fails.

Mental model

paper (.tex / PDF) ──► content plan + claim→evidence audit (codex, fresh)
                              │
   figures extracted ─────────┤  FIGURE_MANIFEST.json (provenance, sha256)
   (real paper figures ONLY)  ▼
   template scaffold ──► fill ──► run_gates.py            ◄─── HARD, loop here
                                  preflight → style → asset → measure → polish
                              │ all hard gates PASS
                              ▼
                    Claude visual review (≤3 issues × ≤3 rounds, fix-vocabulary only)
                              │ score ≥ 9
                              ▼
                    codex final cross-model review (fresh thread, full HTML+PDF)
                              │ pass
                              ▼
                    verify-final → poster.pdf + GATE_REPORT.json

Constants

• SKILL_SCRIPTS = ${CLAUDE_SKILL_DIR}/scripts — all helpers are single-owner and ship inside this skill (Arch C). If the directory is missing the install is broken: abort and tell the user to re-install the skill (Policy A — the gates ARE the skill; never improvise replacements).
• REVIEWER_MODEL = gpt-6-astra, reasoning xhigh, fresh thread per review call (mcp__codex__codex, never codex-reply across review boundaries).
• CANVAS — from the venue's official spec, looked up live in Phase 0. Never assume. (Known anchor: ICLR 2026 main = 185×90 cm landscape per its official printing service; ICML/NeurIPS commonly 60×36 in landscape; workshop posters often 61×91 cm portrait. Specs change yearly — verify.)
• PALETTE — default = templates/tokens/generic.json (slate-blue #2D5F8B accent
  • gold #C9A24A highlight + neutrals) for all venues. Venue packs are opt-in via — venue-colors: true. Purple-dominant accents (hue 250–285) are banned unless the user passes — allow-purple: true.
• AUTO_PROCEED = false — wait for explicit confirmation at every 🚦 checkpoint.
• OUTPUT_DIR = poster_html/ in the working directory.

Workflow

Phase 0 — Resume, dependencies, venue spec

1. Resume: if poster_html/POSTER_STATE.json exists with status: in_progress (< 24 h), resume from the saved phase.
2. Dependencies (degradation chain, in order):
   • Playwright + bundled Chromium → if missing, python3 -m playwright install chromium → if install fails but system Chrome exists, scripts fall back to channel="chrome" → if all fail: you may produce the content plan and scaffold only, label everything "not print verified", and must NOT emit a final PDF.
   • pdfinfo missing → PyMuPDF reads PDF dimensions. At least one of pdftoppm / PyMuPDF must exist for PNG review renders.
   • MathJax: download tex-svg.js once into poster_html/assets/mathjax/ and reference it locally in the HTML. CDN is acceptable only for drafts; the measure gate hard-fails on unrendered MathJax either way.
3. Venue spec lookup (live): consult the venue's official poster-instructions page (search + fetch). Extract dimensions, orientation, font floor, logo policy, anonymity rules, file format. Record {spec, source_url, retrieved} into POSTER_STATE.json — specs change yearly; never reuse a cached spec silently.

🚦 Checkpoint: echo the venue spec table (canvas, orientation, source URL) and the chosen template. Wait.

Phase 0.5 — Design discovery (one AskUserQuestion batch)

Ask once, ≤4 questions: layout template (from templates/README.md), palette (default generic pack / venue pack / custom within constraints), logos + venue mark (paths or "none" — never fabricate; check the venue's logo policy), QR target (paper / code / project page / none — generate offline with qrencode or python-qrcode; never a remote QR-service URL). Persist answers in POSTER_STATE.json as design_decisions — re-read before any later "improvement" so deliberate choices are never reverted.

Phase 1 — Paper ingest, content plan, claim audit

1. Read the paper source (.tex ideal; PDF otherwise). Extract: title/authors/affils, the 3–5 headline numbers, core method (equations verbatim), main results (tables/figures and what they show), takeaways. Build poster_html/POSTER_CONTENT_PLAN.md — what goes in which column, word budget per card. Target density (excluding table cells, captions, author line, footer): standard poster 550–850 words; dense theory+empirical poster 750–1050 words, allowed only when ≥2 compact components are used (eqn-anatomy, flow-strip, derived-col, claim-pills, keybox--4). Warn yourself below 500 words on a 4-column landscape (it will read as sparse next to professionally dense posters) unless the template is hero/visual-first; warn above 1100 unless the user asked for dense mode. Bullets ≤ 8 words when possible — density comes from structure, not long prose. Prefer compact structure over prose: if the paper contains an explicit objective, algorithm, theorem mechanism, or baseline comparison, extract at least two of: (1) empirical objective / loss stack; (2) term-by-term equation anatomy; (3) a method-flow strip grounded in paper variables; (4) a derived-Δ column for method-vs-baseline rows; (5) a 4-up implementation/theory keybox; (6) a claim/evidence pill table for numeric-heavy posters. Do not invent an algorithm. If the paper has only an objective, label the component "objective flow" or "loss anatomy", never "algorithm".
2. Cross-model content audit (fresh codex thread, xhigh): give it the content plan path + paper source path(s) — paths only, no summaries — and ask for a claim→evidence table: | claim on poster | paper file:line | paper says (verbatim) | match? | with match ∈ {OK, NUMERIC-MISMATCH, OVERCLAIM, MISSING-PRECONDITION, NOT-IN-PAPER, SCOPE-NARROWED}. Save to poster_html/CLAIM_EVIDENCE.md.
3. Fix every non-OK row or record it as a user-acknowledged tradeoff.

🚦 Checkpoint: content plan + audit summary. Wait.

Phase 2 — Real paper figures (provenance-gated)

Source preference chain:

1. Paper source figures/ (vector SVG/PDF → convert to SVG via inkscape/pdf2svg if available, else rasterize ≥ 2× rendered px).
2. PDF-only: extract_pdf_figures.py contact-sheet + auto to list candidate regions → pick crops (🚦 human confirms crop choices) → crop at 300–450 DPI.
3. Last resort: user supplies explicit page,x0,y0,x1,y1 bboxes.

Then preprocess_figures.py --autocrop every asset. Every paper-derived image gets a FIGURE_MANIFEST.json entry (source hash, page, bbox, dpi, sha256, natural_px) and is embedded as <img data-source="paper" data-asset-id="...">.

Hard rule: ≥ 2 paper-derived visuals or the asset gate fails. Theory-only papers may waive the total-area rule (--waive-total-area) at a human checkpoint — never silently. Never draw bespoke decorative SVG "figures" as substitutes.

Figure-area bands (asset gate, fractions of body): total target 14–22 % (warn < 12 % / > 24 %, hard < 10 % / > 28 %); per ordinary figure target 4–8 % (warn

10 %, hard > 13 %); figure--duo combined 8–12 %. Hero templates pass --hero (centerpiece may take 30–40 %). The failure mode is symmetric: too small reads as decoration, too big crowds out content. Sibling figures that share axes or tell a before→after story belong in one figure--duo card, not two cards.

Phase 3 — Scaffold + tokens

cp templates/<chosen>.html poster_html/poster.html; retarget @page + .poster dims to the venue canvas (two edits, same values); apply the chosen token pack onto the :root DESIGN TOKENS block; fill content per the plan; embed manifest figures. Run preflight + style_check — both must PASS before any layout iteration. (A fresh scaffold is expected to fail measure — that gate judges a filled poster.)

Phase 4 — Layout hard loop

After every layout change:

python3 "$SKILL_SCRIPTS/run_gates.py" poster_html/poster.html \
    --tokens <pack.json> --manifest poster_html/FIGURE_MANIFEST.json \
    --report poster_html/GATE_REPORT.json

Canonical order: preflight → style → asset → measure → polish. Targets: column-bottom spread < 5 px (aim < 3), footer gap ∈ [30, 50] px, intercard gap ∈ [12, 50] px, canvas-fill ∈ [95, 101] %, poster bbox aligned to page within ±2 px. Fix guidance for each failure mode lives in the gate output and templates/COMPONENTS.md. Do not proceed while any hard gate fails. Do not let a reviewer see an unmeasured poster. Balance under-filled columns with content from the paper (Gate C), never with whitespace, space-between, or stretched cards.

Phase 5 — Claude visual review (gated aesthetics)

Render and read the result yourself:

python3 "$SKILL_SCRIPTS/render_preview.py" poster_html/poster.html
pdftoppm -r 100 poster_html/poster_preview.pdf poster_html/review_full -png -f 1 -l 1
# plus 2-4 region crops at higher res (header / one column / equations) via PIL

Calibrate first (../shared-references/taste-calibration.md): if human-curated references/good/ + references/bad/ exist under this skill dir (or the project supplies its own pair), score those 3+3 reference posters on the axes below BEFORE the target, anchoring the scale. Never select, search for, or generate anchors yourself; if no reference sets exist, proceed uncalibrated and mark CALIBRATION: none — never fabricate anchor scores. Axes (weights sum 1.0): Design 0.35 · Craft 0.30 · Functionality 0.20 · Originality 0.15. Mapping: SCORE = min(round(1 + 9 × COMPOSITE), lowest triggered cap) — caps apply AFTER the mapping, and the loop's Score ≥ 9 threshold below always reads this final capped SCORE, never the raw composite.

Score strictly 1–10. Critical caps (hard floors — a calibrated composite never overrides them): < 2 real paper figures → ≤ 3; broken canvas / clipped content / unreadable math → ≤ 4; ≥ 4 visible hue families or gradient-heavy header → ≤ 4; large blank cards or columns → ≤ 5; fabricated visual claim → ≤ 3. Checks: posterly-showcase gestalt (would this hang next to a professionally designed poster without looking like a patched dashboard?), single-accent discipline, real figures readable and central, print hierarchy (title → headline stats → figures → detail), column fill, equation prominence (no tiny math in oversized boxes), serif-body/sans-display pairing, no gradient kitsch, component consistency, 60-second narrative. Output format:

SCORE: N/10            (= min(round(1 + 9 × COMPOSITE), lowest cap); drives the loop)
COMPOSITE: 0.xx        (weighted; list the four per-axis scores)
CALIBRATION: anchored | none
GAP: <which reference poster the target falls short of / exceeds, on which axis, and why — one paragraph; omit only when CALIBRATION: none>
CAPS_TRIGGERED: ...
TOP_ISSUES: (max 3)
ALLOWED_FIX_TYPE per issue: token | component | rebalance | asset | template/canvas
PATCH_LOOP_RISK: low | medium | high

Loop: fix (fix vocabulary below) → re-run Phase 4 gates → re-score. ≤ 3 issues per round, ≤ 3 rounds. Score ≥ 9 → Phase 6. Still < 9 after 3 rounds → STOP patching; escalate to template / canvas / content re-choice (back to Phase 3) or a human decision. Never enter round 4 of cosmetic patching.

Fix vocabulary (closed set — the anti-patch-loop core)

Allowed: (a) edit a :root token value; (b) swap/remove/add a whole component instance from templates/COMPONENTS.md; (c) content rebalance (move a card across columns, trim/grow text from the paper, resize a figure within its AR band); (d) template/canvas re-choice; (e) global edits to an existing component's CSS that reference only tokens; (f) switching predefined variants (.eqn--large, .card--compact, .figure--wide, .nowrap, …); (g) asset fixes (re-crop, swap for a clearer figure from the same paper, re-preprocess).

Forbidden: new inline styles, new hex values anywhere, bespoke decorative SVG, per-element font-size overrides. A new component may not be born inside the visual loop — stop, get a human checkpoint, add it to COMPONENTS.md, re-run from Phase 3.

Phase 6 — Codex final review (fresh thread, cross-model)

All hard gates PASS + polish warnings zero-or-waived + visual ≥ 9 first. Then a fresh codex thread (xhigh) reviews the final artifacts (not the content plan): poster.html, the rendered PDF/PNG, the paper source, GATE_REPORT.json, CLAIM_EVIDENCE.md — paths only, no executor framing. It checks: (1) fidelity & overclaims re-checked on final text (polish introduces new claims), (2) residue (\ref{, TODO, raw < in math, missing images, remote URLs), (3) visual rhetoric (headline numbers prominent, banner readable from 2 m), (4) gate-log coherence. The reviewer recommends; it does not edit. Any fix → back through Phase 4/5 gates — never straight to re-review.

Phase 7 — Final verification + report

python3 "$SKILL_SCRIPTS/poster_check.py" verify-final poster_html/poster_preview.pdf \
    --from-html poster_html/poster.html --max-size-mb 20

Page count 1, dimensions match @page, size ≤ 20 MB, no TODO/residue, no remote assets. Report: PDF path, final spread px, footer-gap range, gate summary table, unresolved waivers, codex verdict. Update POSTER_STATE.json → done.

State persistence

poster_html/POSTER_STATE.json: {phase, venue, canvas{w,h,orientation,source_url, retrieved}, template, token_pack, design_decisions{...}, figures_selected[], visual_rounds, codex_threads{audit, final}, status, timestamp} — written after every phase; enables compact-recovery resume.

Key rules

• Measure, don't eyeball. No layout claim without run_gates.py output.
• Gates before aesthetics. Claude/codex review only ever sees a poster whose hard gates PASS. This ordering is what kills the patch-loop death spiral.
• Never invent paper numbers or figures. Numbers come from the paper source; visuals carry manifest provenance. Fabrication = critical cap ≤ 3.
• Two hues, one system. Accent + gold + neutrals. The style gate enforces it; don't negotiate with the gate.
• Real figures are the poster. A poster without the paper's own figures is a dashboard, not a poster.
• Fix vocabulary is closed. If a fix isn't expressible as token / component / rebalance / asset / canvas, it's the wrong fix.
• Cross-model verdicts. Claude drives the loop and scores visuals; acceptance of content fidelity comes from the fresh codex thread (a loop can drive, never acquit).
• Preserve user decisions. Re-read design_decisions before "improving" anything.
• Vendor boundary. poster_check.py, render_preview.py, _posterly/ are vendored from posterly — keep diffs minimal; ARIS-side logic goes in the new scripts, not in vendored files.

Review tracing

Save every codex reviewer call's trace per shared-references/review-tracing.md to .aris/traces/paper-poster-html/<date>_run<NN>/ (audit + final threads, raw responses).

Output contract

poster_html/
├── poster.html              # single-file source of truth
├── poster_preview.pdf       # print-emulated, verify-final-checked
├── poster_preview.png       # thumbnail
├── POSTER_STATE.json        # resume state
├── GATE_REPORT.json         # canonical gate ledger (schema v1)
├── POSTER_CONTENT_PLAN.md   # what-goes-where + word budgets
├── CLAIM_EVIDENCE.md        # codex claim→evidence audit
├── FIGURE_MANIFEST.json     # figure provenance (sha256, page, bbox, dpi)
└── assets/{paper_figures,logos,qr,mathjax}/

When NOT to use

• Slides, not a poster → /paper-talk / /slides-polish.
• The paper's headline isn't stable yet — fix the paper first; a poster amplifies whatever story it's given.

Other files in this skill

• DESIGN_FINAL.md
• IMPLEMENTATION_CONVENTIONS.md
• LICENSES/posterly-MIT.txt
• NOTICE.md
• scripts/_posterly/init.py
• scripts/_posterly/canvas.py
• scripts/_posterly/measure.py
• scripts/_posterly/polish.py
• scripts/_posterly/preflight.py
• scripts/_posterly/render.py
• scripts/_posterly/textutil.py
• scripts/_posterly/verify_final.py
• scripts/asset_check.py
• scripts/extract_pdf_figures.py
• scripts/poster_check.py
• scripts/preprocess_figures.py
• scripts/render_preview.py
• scripts/run_gates.py
• scripts/style_check.py
• templates/COMPONENTS.md
• templates/README.md
• templates/landscape_4col.html
• templates/landscape_hero.html
• templates/portrait_2col.html
• templates/tokens/acl.json
• templates/tokens/cvpr.json
• templates/tokens/generic.json
• templates/tokens/iclr.json
• templates/tokens/icml.json
• templates/tokens/neurips.json

DESIGN_FINAL.md (verbatim)

paper-poster-html — 收敛后的最终设计(codex 3 轮讨论产物)

日期:2026-06-05。讨论 thread:019e96dc-8350-70f0-a595-c1ad15919aa8(gpt-5.5 xhigh)。 状态:Round 2 全点收敛,本文档为实现规格。

0. 定位

• HTML+CSS 是新的默认 poster 路径;旧 /paper-poster(LaTeX tcbposter)已退役为 重定向 stub(2026-06-07 落库决定,较本规格的 legacy 共存方案更进一步; 旧实现仅存于 git history)。
• posterly(MIT, github.com/Chenruishuo/posterly)的 tools vendor 进 skill,不外部依赖 不重写;LICENSES/posterly-MIT.txt + NOTICE.md 注明来源与 ARIS 修改。

1. 目录布局

skills/paper-poster-html/
├── SKILL.md
├── LICENSES/posterly-MIT.txt
├── NOTICE.md
├── templates/
│   ├── README.md
│   ├── COMPONENTS.md          # 组件契约目录(Q6)
│   ├── landscape_4col.html    # fork 自 posterly,de-gradient + token 化
│   ├── landscape_hero.html
│   ├── portrait_2col.html
│   └── tokens/
│       ├── generic.json       # 默认:slate-blue #2D5F8B + gold #C9A24A
│       ├── iclr.json … cvpr.json   # opt-in venue 包
└── scripts/
    ├── poster_check.py        # vendored(measure/preflight/polish/verify-final)
    ├── render_preview.py      # vendored(Playwright print render)
    ├── _posterly/…            # vendored 内部模块
    ├── style_check.py         # 新:风格硬门(12 条规则)
    ├── asset_check.py         # 新:真图溯源门
    ├── run_gates.py           # 新:canonical 顺序跑全门,写 GATE_REPORT.json
    ├── extract_pdf_figures.py # 新:PDF→contact sheet→候选裁剪
    └── preprocess_figures.py  # 新:autocrop/转格式/分辨率检查

工作目录输出:

poster_html/
├── poster.html / poster.pdf / poster_preview.png
├── POSTER_STATE.json / GATE_REPORT.json
├── CLAIM_EVIDENCE.md / FIGURE_MANIFEST.json
└── assets/{paper_figures,logos,qr,mathjax}/

2. 设计 token 纪律

• 默认色卡 = generic(所有 venue):accent #2D5F8B 族 + gold #C9A24A 族 + 中性色。 venue 色卡 opt-in(— venue-colors: true),约束:accent S≤0.55、L∈[0.25,0.45]; gold 族固定 H∈[38,48]、S≤0.65、L∈[0.42,0.65];主 accent 禁紫(H 250–285), 除非 — allow-purple: true。venue identity 默认文字 badge。
• 字号必须走 --fs-* token scale(≤9 档,超出 warn)。
• serif 正文(Charter/Source Serif Pro/Georgia/Times New Roman)+ sans 标题(Inter/Aptos/Helvetica Neue/Arial);mono 仅代码(Menlo/Consolas)。

3. style_check.py 源门规则(codex 定稿,逐条实现)

#	严重度	规则
1	HARD	颜色字面量只许出现在 token 文件 / :root token block;例外:data-color-exempt="logo" 的 SVG 内部
2	HARD	禁 inline style 含颜色/字体/字号/布局关键值(豁免同上 + paper asset 内部)
3	HARD	组件 CSS 颜色必须 var(--…)
4	HARD	渲染后非中性色相聚类 ≤2(聚类半径 18°,须落在 accent/gold hue ±22°;非中性=alpha≥0.10 且 S≥0.18;豁免 <img>、logo、data-source="paper"、QR)
5	HARD	禁 linear-gradient;radial-gradient 仅许 .poster 背景且所有 color stop alpha≤0.06
6	HARD	字体配对:正文 serif 栈、标题/表头 sans 栈
7	HARD	字体白名单(§2)
8	HARD	字号必须用 --fs-* token 或组件 class,禁任意 px 漂移
9	WARN	字号 token >9 档
10	HARD	契约属性:论文图必须 data-source="paper" + data-asset-id;logo 豁免必须显式标注
11	HARD	禁自造装饰 SVG;inline SVG 仅许 logo / QR fallback / COMPONENTS.md 已收录的结构图
12	WARN	大面积深色(L<0.18 且 >8% poster 面积)→ 土嗨预警

4. asset_check.py 真图门

• ≥2 张 data-source="paper" 图;每张面积 ≥ poster 1.5%;paper-image 总面积 ≥ body 12%。
• raster natural size ≥ rendered size 1.5×(目标 2×)。
• FIGURE_MANIFEST.json 必填:source PDF hash、page、bbox、crop dpi、asset sha256、是否来自论文。
• 真图获取链:论文源 figures/(SVG/PDF→SVG 转换优先)→ PDF-only 时 PyMuPDF 300–450 DPI 渲染 contact sheet → 自动候选 + 人工选 → 用户给 page,x0,y0,x1,y1 bbox → 不足 2 张硬失败 (除非 human checkpoint 显式 waiver)。

5. 公式门(半硬)

• EQN/BROKEN(MathJax 没渲染出来)= HARD(vendored measure 已有)。
• EQN/UNDERSIZED:.eqn inner box >80px 高且 math bbox 面积 <15% → HARD; <25% 或底部空白 >35% → WARN(final 前必须修复或记录 waiver)。

6. MathJax 本地化

Phase 0 下载 tex-svg.js 到 poster_html/assets/mathjax/(缓存复用),HTML 引本地路径; 下载失败 → 询问后 CDN 仅供草稿;final 的 measure 门对 MathJax 失败保持硬失败。

7. run_gates.py + GATE_REPORT.json

• canonical order:preflight → style_check → asset_check → measure → polish。
• 默认 accumulate(一次给全修复面),--fail-fast 可选。
• style/asset 保持独立 CLI(vendor diff 干净),run_gates.py 做编排。
• GATE_REPORT.json schema:schema_version/skill/timestamp/poster_html/canvas{source,width_cm, height_cm,orientation,source_url}/overall/hard_failures/warnings/gates[{name,severity,status, command,summary,artifacts}]。
• polish 的 WARN 在 Phase 6 前必须清零或显式 waiver。

8. Workflow phases(SKILL.md 主结构)

Phase	内容	门	Checkpoint
0	resume + deps(Playwright 链)+ venue spec 实时调研(WebSearch/WebFetch 官方页,URL+date 入 state)	—	🚦确认 venue/canvas
0.5	设计问卷(layout/palette/logo/QR/source 一轮 AskUserQuestion)	—	🚦确认设计输入
1	paper ingest + content plan + claim→evidence 表	codex fresh xhigh 内容审计	🚦全 claim OK 或用户接受 tradeoff
2	真图提取/预处理	asset_check + FIGURE_MANIFEST	🚦PDF-only 时人工选裁剪
3	scaffold + token patch	preflight + style_check	—
4	布局硬循环	preflight+style+asset+measure(spread<5 aim<3;footer gap 30–50;intercard 12–50;fill 95–101%;position≤2px)	—
5	渲染 + Claude 视觉审(rubric §9)	≤3 issue×≤3 轮;fix 限定词汇表 §10	—
6	codex 终审(fresh xhigh,审 final HTML+PDF 不是 plan;fidelity/overclaim/residue/叙事/gate logs;不直接改文件)	任何 fix 回 Phase 4/5	—
7	verify-final + 报告	PDF 1 页/尺寸/≤20MB/无 TODO/无 remote asset	完成

Playwright 降级链:bundled Chromium → python -m playwright install chromium → channel="chrome" → 仍失败则只产 content plan/scaffold,标注 "not print verified", 不许产出最终 PDF。pdfinfo 缺 → PyMuPDF 读尺寸;pdftoppm/PyMuPDF 至少一个用于 PNG。

9. Claude 视觉 rubric(Phase 5)

1–10 分;critical cap:无真图或 <2 张 →≤3;画布坏/裁切/公式不可读 →≤4; ≥4 个色相家族或重渐变 header →≤4;大空白卡/列 →≤5;捏造视觉 claim →≤3。 检查项:posterly-showcase gestalt / 单 accent 纪律 / 真图居中可读 / 打印层级 (title→headline→figures→details)/ 列底对齐无半空卡 / 公式占框 / serif+sans 配对 / 无渐变 kitsch / 组件一体感 / 60 秒叙事。 输出格式:SCORE: N/10、CAPS_TRIGGERED、TOP_ISSUES(≤3)、 ALLOWED_FIX_TYPE: token|component|rebalance|asset|template/canvas、PATCH_LOOP_RISK。 校准:旧 poster ≤3 分;posterly showcase ≥9 分。

10. Fix 词汇表(反补丁循环核心)

视觉审循环内只允许: (a) 改 :root token 值; (b) 整组件实例的换/删/加(组件集来自 COMPONENTS.md); (c) 内容再平衡(卡片跨列移动 / 从论文取材增删文字 / AR 门带宽内调图); (d) 画布/模板重选(升级路径); (e) 组件 stylesheet 的全局改动(只许引用 token,禁新 hex); (f) 预定义 variant 切换(.figure--wide、.card--compact、.eqn--large、.nowrap 等, 必须已录入 COMPONENTS.md); (g) asset fix(重裁剪/换同论文更清晰图/重跑 preprocess)。 禁止:新 inline style、新 hex、自造装饰 SVG、单元素字号 override。 新组件禁止在视觉循环内诞生——需要新组件 → 停,human checkpoint 录入 COMPONENTS.md, 从 Phase 3 重跑。

11. COMPONENTS.md 契约

每组件:purpose / allowed variants / required data attributes / token usage / which gates inspect it / allowed fix operations / anti-patterns。 首发组件:card、numbered-card、figure-card、hero-figure、eqn、result-table、 claim-evidence、keybox、takeaways、qr-block、venue-badge、footer。

12. 已知失败模式防御(codex round 1 §7)

remote 资源 networkidle 假死→本地化;logo 豁免走私颜色→显式 data-color-exempt; 低清裁剪→1.5×/2× 检查;截图导致 PDF 爆体积→verify-final 20MB; 视觉审诱发新组件/新色→fix 词汇表;venue 规格过期→每次实时查+记录 URL/date; 改写引入新 claim→Phase 6 审 final HTML 不是 content plan。

12.5 Round-3 ACK nits(已采纳)

1. style_check 规则 8:calc(var(--fs-*) * …) 仅许预定义组件 variant 使用,否则成漏洞。
2. asset_check 的"paper-image 总面积 ≥ body 12%"对纯理论论文可 waiver——但首个验收案例不许 waiver。
3. Phase 0 的 venue 调研在 skill 文案里写成泛化的 "official venue page lookup"(跨工具栈映射,codex 镜像兼容)。

13. 验收

首个验收案例:为一篇公开的 ICLR 2026 OpenReview 论文(理论+实验混合型)重做 poster: 画布 185×90cm 横版(ICLR 官方打印服务规格),generic 色卡, 真图来自 OpenReview PDF,目标执行方视觉 rubric ≥9 + 跨模型终审通过。 (已达成:全 gate PASS、列底 spread <1px、两轮跨模型终审 PRINT-READY。)

IMPLEMENTATION_CONVENTIONS.md (verbatim)

paper-poster-html 实现约定(所有实现 agent 必读)

配合 DESIGN_FINAL.md(规格)使用。本文档定死跨文件契约——实现时逐字遵守, 有疑问按本文档,不要自由发挥。

A. CSS Token 契约(templates + tokens/*.json + style_check 三方共享)

:root 中的 token block 必须被注释 /* ===== DESIGN TOKENS ===== */ 和 /* ===== END DESIGN TOKENS ===== */ 包围(style_check 靠这对注释定位 token block)。

颜色 token(只有这些地方允许出现颜色字面量):

--accent: #2D5F8B;  --accent-deep: #1F4566;  --accent-light: #E8F1F8;  --accent-soft: #D7E5F0;
--gold: #C9A24A;    --gold-soft: #FFF7E0;
--text-primary: #1A1A1A;  --text-secondary: #555555;  --text-muted: #888888;
--bg-page: #F6F2F0;  --bg-card: #FFFFFF;  --bg-card-tint: #FAFAFB;  --bg-emphasis: var(--accent-light);
--border-soft: #D8D8D8;  --border-strong: var(--accent);

字号 scale(9 档,模板内所有 font-size 必须引用之一;calc(var(--fs-N) * k) 仅许 COMPONENTS.md 预定义 variant 使用):

--fs-1: calc(9 * var(--u));   /* 微标签 */      --fs-2: calc(10 * var(--u));  /* 小 caption */
--fs-3: calc(11 * var(--u));  /* caption/表格 */ --fs-4: calc(12 * var(--u));  /* 正文 */
--fs-5: calc(13 * var(--u));  /* 公式/强调 */    --fs-6: calc(15 * var(--u));  /* 副标题 */
--fs-7: calc(16 * var(--u));  /* 节标题 */       --fs-8: calc(22 * var(--u));  /* banner 数字 */
--fs-9: calc(32 * var(--u));  /* 主标题 */

单位:--u: 1.6px(screen)/ @media print { :root { --u: 1mm } }。其余尺寸一律 calc(N * var(--u)),hairline 可用裸 px(≤2px)。

B. HTML 属性契约

属性	用途	谁检查
`data-measure-role="poster	header	banner
data-source="paper" + data-asset-id="<manifest id>"	标记来自论文的图	asset_check + style_check 豁免
data-color-exempt="logo"	logo/印章 SVG 的调色豁免	style_check
data-fig-layout="beside-text"	图文并排的 AR 门 opt-out(vendored)	poster_check polish

模板和最终 poster 中禁止任何 style= 属性(零容忍,style_check 规则 2 实现成这样, 简单可靠)。例外:data-color-exempt="logo" 元素的内部 SVG 标记、data-source="paper" 的 <img> 上仅允许 style="width: NN%"(AR 调宽)。 为此模板必须自带 utility classes(替代 posterly 模板里的 inline style):

.fs-1 … .fs-9        /* font-size: var(--fs-N) */
.mt-1 … .mt-6        /* margin-top: calc(N * var(--u)) */
.mb-1 … .mb-4        /* margin-bottom */
.w-45 .w-50 … .w-100 /* 图宽 45%…100%,步长 5 */
.text-secondary .text-muted .nowrap .text-center

C. CLI 契约(scripts/)

全部 Python 3.10+,只用 stdlib + 已确认可用的 PyMuPDF(fitz)/PIL/playwright(lazy import, 缺失时给可读错误+降级指引)。每个脚本 --help 完整。exit code:0=pass,1=hard fail,2=用法/环境错误。

style_check.py

python3 style_check.py POSTER.html [--tokens TOKENS.json] [--json OUT.json]
                       [--no-render]  # 跳过渲染门(规则4、12 标 SKIPPED)

实现 DESIGN_FINAL §3 的 12 条规则 + §12.5 nit 1。源门(规则 1-3,5-11)纯静态解析 (html.parser + 正则提 CSS);渲染门(规则 4、12)用 playwright 取 computed style。 色相聚类:rgba→HSL;非中性= alpha≥0.10 且 S≥0.18;greedy 聚类半径 18°(色环距离); 聚类数 ≤2 且每类中心落在 tokens 的 accent/gold hue ±22°(hue_centers 来自 --tokens JSON, 缺省从 :root 解析 --accent/--gold 算)。 JSON 输出:{"gate":"style","status":"PASS|FAIL|WARN","rules":[{"id":1,"severity":"hard","status":"PASS","detail":"..."}]}

asset_check.py

python3 asset_check.py POSTER.html --manifest FIGURE_MANIFEST.json [--json OUT.json]
                       [--min-paper-figs 2] [--min-fig-area 0.015] [--min-total-area 0.12]
                       [--waive-total-area]   # 纯理论论文 waiver(DESIGN_FINAL §12.5 nit 2)
                       [--no-render]          # 面积检查降级为 natural-size 估算

检查:≥N 张 data-source="paper" 且 manifest 里 from_paper=true;每张渲染面积 ≥ poster 1.5%; 总面积 ≥ body 12%(可 waive);natural_px ≥ rendered px 1.5×(WARN 在 <2×); manifest 必填字段齐全(见 D);文件存在且 sha256 匹配。

run_gates.py

python3 run_gates.py POSTER.html [--report GATE_REPORT.json] [--fail-fast]
                     [--strict-polish] [--tokens TOKENS.json] [--manifest FIGURE_MANIFEST.json]
                     [--waive-total-area] [--no-render]

canonical order:preflight → style → asset → measure → polish。默认 accumulate。 子门以 subprocess 调同目录脚本(sys.executable;poster_check.py 子命令用其 CLI)。 GATE_REPORT.json 严格按 DESIGN_FINAL §7 schema(canvas 信息从 POSTER_STATE.json 读, 读不到则从 @page 解析,source 标 "page-rule")。汇总 overall=PASS/FAIL + hard_failures + warnings。

extract_pdf_figures.py

python3 extract_pdf_figures.py PAPER.pdf --out DIR [--dpi 350]
        contact-sheet                      # 整页缩略 contact sheet + 自动候选框
        crop --page P --bbox x0,y0,x1,y1 --name ID [--caption-hint "..."]
        auto                               # 自动检测大图块候选(图/表),输出候选列表

bbox 单位 = PDF points(72dpi 坐标,fitz 默认)。crop 模式渲染该区域至 --dpi,写 PNG 到 DIR,并 upsert FIGURE_MANIFEST.json(同目录上级)。contact-sheet 写 DIR/contact_sheet_pNN.png。

preprocess_figures.py

python3 preprocess_figures.py IMG... [--autocrop] [--pad 6] [--min-px 1200 700] [--manifest M.json]

PIL autocrop 白边(ImageChops.difference vs 白底,留 --pad px),报告 natural size, 低于 --min-px 给 WARN;改动后同步更新 manifest 的 natural_px/sha256。

D. FIGURE_MANIFEST.json schema

{
  "schema_version": 1,
  "source_pdf": {"path": "…", "sha256": "…"},
  "figures": [
    {"asset_id": "fig_method", "file": "assets/paper_figures/fig_method.png",
     "from_paper": true, "page": 3, "bbox": [72.0, 100.0, 520.0, 380.0], "dpi": 350,
     "sha256": "…", "natural_px": [2178, 1362], "caption_hint": "Figure 2: …"}
  ]
}

E. 模板改造配方(posterly → ARIS fork)

对 3 个模板各做(以上游 posterly 仓库的 templates/*.html 为底):

1. 文件头注释:保留原说明,追加 "Adapted from posterly (MIT, © 2026 Ruishuo Chen) — see LICENSES/ & NOTICE.md; ARIS modifications: flat de-gradient, --fs token scale, zero-inline-style utilities, data-source/data-color-exempt contracts."
2. 去渐变:.poster::before 顶条 → 纯色 var(--accent);.framework-banner、 .takeaways-strip 背景 → 纯 var(--bg-emphasis);.callout.gold → 纯 var(--gold)。 .poster 的 radial tint(alpha≤0.06)保留。
3. token block:按 §A 注释包围;加 --fs-1..9;所有 font-size 改 var(--fs-N)。
4. 消灭 inline style:模板正文里所有 style="…" 换成 §B utility classes(在 CSS 段新增)。
5. 图组件:.figure img 的 TODO 注释里写明契约:<img src="assets/paper_figures/x.png" data-source="paper" data-asset-id="x" class="w-95">。
6. logo 槽注释写明 data-color-exempt="logo"。
7. data-measure-role 一律保留。
8. @page 默认值保留(60×36in / 24×36in),在头注释加"画布重定位:同步改 @page 与 .poster 的 width/height(各一处),ICLR 2026 main = 185cm 90cm landscape 示例"。
9. eqn 组件加 variant .eqn--large(font-size: calc(var(--fs-5) * 1.25),预定义 calc 豁免)。
10. 自检:改完后模板里 grep 不到 linear-gradient、style="(除 §B 两个例外注释示例)、 裸 #hex(token block 与 logo SVG 例外)。

F. tokens/*.json schema

{"name": "generic",
 "accent": {"base": "#2D5F8B", "deep": "#1F4566", "light": "#E8F1F8", "soft": "#D7E5F0"},
 "gold": {"base": "#C9A24A", "soft": "#FFF7E0"},
 "neutrals": {"text_primary": "#1A1A1A", "text_secondary": "#555555", "text_muted": "#888888",
              "bg_page": "#F6F2F0", "bg_card": "#FFFFFF", "bg_card_tint": "#FAFAFB",
              "border_soft": "#D8D8D8"},
 "hue_centers": {"accent": 210, "gold": 43}}

venue 包约束:accent S≤0.55、L∈[0.25,0.45]、禁 H 250–285;gold 全包固定 generic 值; light/soft 从 base 推(同 hue 低饱和高亮度)。bg_page 可随 accent 微调暖/冷但 ΔE 要小。

G. 测试基线

上游 posterly 仓库的 examples/hello_world/poster.html 是全门 PASS 的参照(vendored 四门)。 新脚本写完后:style_check 对 hello_world 允许 FAIL(它有 inline style——posterly 原版风格), 但对我们改造后的模板(填充前)源门必须 PASS;run_gates 对脚手架预期 measure FAIL(未填充), 这是正常的(模板=脚手架)。

templates/README.md (verbatim)

Template gallery — paper-poster-html

Three neutral HTML scaffolds, forked from the posterly templates (MIT, © 2026 Ruishuo Chen — see ../NOTICE.md) and adapted for ARIS: flat de-gradient, --fs-* token scale, zero inline-style utility classes, and the data-source / data-color-exempt contracts. Class names are unchanged, so COMPONENTS.md (the component contract catalog) applies to all three.

Each template is self-contained and neutral: no lab branding, no paper content — only TODO placeholders. The authoring loop is: copy one to your working dir as poster.html → apply a token pack → fill TODOs with paper content + real figures → run the gates (run_gates.py) and balance until they pass.

Every layout-critical element carries data-measure-role so the measurement gate can locate columns / hero / footer regions across templates. Do not remove these attributes — the measure gate depends on them.

Picking a template

Template	Canvas	Layout	Use when
landscape_4col.html	60 × 36 in landscape	header → optional banner → 4 columns → optional takeaways → footer	The default. Standard ML conference poster (ICML / NeurIPS / generic landscape) with ~3–5 content cards per column; mix of figures, equations, and tables.
landscape_hero.html	60 × 36 in landscape	header → hero panel (~60%) + supporting column (~40%) → takeaways → footer	ONE figure / table / system diagram is the main message. One big illustration left, 3–4 short cards right. No framework banner — the hero is the banner.
portrait_2col.html	24 × 36 in portrait	header → 2 columns → footer (no banner, no takeaways strip)	Portrait venues / sub-A0 sizes. Vertical space is precious, so banner + takeaways are dropped; the final card in the right column carries the conclusion/takeaways.

(File names follow DESIGN_FINAL §1: the ARIS forks drop the posterly _neutral suffix.)

Scaffolds, not finished posters

A template is a scaffold, not a poster. Figures are commented out and copy is TODO stubs, so each column only fills the top of the canvas. That has gate consequences you should expect — do not "fix" them on a fresh scaffold:

• preflight passes out of the box. It checks structure (valid data-measure-role values, no LaTeX residue, no bare < inside $…$ math, the root data-measure-role="poster"), which the scaffold already satisfies.
• style_check source rules (1–3, 5–11) pass on the forked scaffold. The ARIS fork has no inline styles, no linear-gradient, no stray hex outside the token block — so the static source gate is green on the empty template. (The upstream posterly originals keep inline styles and would FAIL style_check — that is expected; the fork exists precisely to fix that.)
• measure and polish are gates for your finished poster. They check that columns bottom-align to within 5 px, that the gap to the footer sits in a tight band (30–50), that intercard gaps stay in [12, 50], and that the canvas fills 95–101% — properties only a filled poster can have. An unfilled scaffold is expected to fail them (huge column-bottom spread, a large gap to the footer). That is the gate telling you the poster is not finished, not a bug in the template.
• asset_check fails until you embed ≥2 real paper figures. A scaffold has none.

So the loop is: copy → apply token pack → fill content + drop in real figures → run run_gates.py and balance until measure/polish/asset go green. See DESIGN_FINAL §8 for the full phase structure and the worked ICLR 2026 acceptance case (§13).

Applying a token pack (tokens/*.json → :root)

The palette lives in two synchronized places: the tokens/*.json packs (machine-readable, what style_check/run_gates read for the accent/gold hue_centers) and the :root DESIGN TOKENS block in each template's <style> (what the browser renders). generic.json is the default for every venue. Venue packs are opt-in (— venue-colors: true).

To apply a pack, manually copy its values into the template's :root token block (the block fenced by /* ===== DESIGN TOKENS ===== */ … /* ===== END DESIGN TOKENS ===== */ — style_check locates the token block by exactly this comment pair, so keep it intact). The JSON-to-CSS field mapping:

JSON path	CSS token
accent.base	--accent
accent.deep	--accent-deep
accent.light	--accent-light
accent.soft	--accent-soft
gold.base	--gold
gold.soft	--gold-soft
neutrals.text_primary / text_secondary / text_muted	--text-primary / --text-secondary / --text-muted
neutrals.bg_page / bg_card / bg_card_tint	--bg-page / --bg-card / --bg-card-tint
neutrals.border_soft	--border-soft

--bg-emphasis: var(--accent-light) and --border-strong: var(--accent) are derived tokens — leave them as var(--…) references; they follow the accent automatically.

When you pass a pack to the gates (run_gates.py --tokens tokens/<venue>.json, style_check.py --tokens …), the hue_centers in the JSON are the source of truth for the hue-cluster check (style rule 4): the two allowed non-neutral hue families are hue_centers.accent ± 22° and hue_centers.gold ± 22°. If you copied the JSON values into :root correctly, the rendered hues will land inside those windows. (If you omit --tokens, style_check derives the centers from the :root --accent / --gold instead.)

Available packs (all share the generic gold family #C9A24A / #FFF7E0; all venue accents satisfy S ≤ 0.55, L ∈ [0.25, 0.45], hue ∉ [250, 285]):

Pack	Accent identity	accent hue center
generic.json	slate-blue #2D5F8B (the default for every venue)	210
iclr.json	deep green #2E6048	151
icml.json	deep maroon #8B3A4A	348
neurips.json	steel/slate blue #3A5A7A	210
cvpr.json	deep azure/indigo #27407A	222
acl.json	deep teal #256E72	183

Venue packs are opt-in identity, not a license to deviate from the discipline: a single accent family + the shared gold, deep and desaturated, never purple. The default text venue-badge (COMPONENTS.md) is the primary venue cue; the color pack is secondary.

Retargeting the canvas

A template ships with a default @page size and a matching .poster size. To change the canvas you must edit both in the template (each in exactly one place), keeping them identical — measure's canvas-fill / position-align gate compares the rendered .poster bounding box against the @page viewport, so a mismatch fails the gate.

1. @page in the <style> block — e.g. @page { size: 185cm 90cm; margin: 0; }.
2. .poster print dimensions in @media print { .poster { … width: 185cm; height: 90cm; } }. (The screen .poster uses calc(N * var(--u)) with --u: 1.6px; print uses --u: 1mm, so the screen preview scales automatically — you only hardcode the print width/height.)

The canvas parser accepts in / mm / cm / pt units. ICLR 2026 main conference example (the §13 acceptance case): the official print service spec is 185 × 90 cm landscape, so start from landscape_4col.html and set both @page and the print .poster to 185cm 90cm / width: 185cm; height: 90cm;.

Safe-area / margin design belongs as internal padding on a full-bleed .poster, never as a smaller poster — a smaller poster fails the position-align gate (the .poster bbox must align to (0,0)–(viewport_w, viewport_h)).

Zero inline-style + utility-class policy

The templates and any finished poster carry no style= attribute (zero tolerance — style_check rule 2; IMPLEMENTATION_CONVENTIONS §B). To make that possible the templates ship their own utility classes in the <style> block, replacing every inline style the posterly originals used:

.fs-1 … .fs-9        font-size: var(--fs-N)
.mt-1 … .mt-6        margin-top: calc(N * var(--u))
.mb-1 … .mb-4        margin-bottom: calc(N * var(--u))
.w-45 .w-50 … .w-100 figure <img> width 45%…100% (5% steps)
.text-secondary .text-muted .nowrap .text-center

The only sanctioned style= survivors:

• the internal markup of a data-color-exempt="logo" element (a logo / seal SVG), and
• style="width: NN%" on a data-source="paper" <img> for aspect-ratio width tuning when the value is off the 5% grid (prefer the .w-NN class when it lands on a step).

Likewise: no linear-gradient anywhere; the only gradient allowed is the .poster background radial-gradient tint with all color stops at alpha ≤ 0.06 (style rule 5). All font-sizes go through the --fs-* scale; calc(var(--fs-N) * k) is allowed only for the predefined component variants catalogued in COMPONENTS.md (e.g. .eqn--large). All colors are tokens — no hex literal outside the :root token block (and the exempted logo SVG).

Adding a new template

A new template MUST:

1. Set @page { size: <W> <H> } (in / mm / cm / pt) inside a <style> block — the canvas parser fails if absent — and a matching print .poster size.
2. Carry data-measure-role="poster" on the root poster element.
3. Use the roles consistently: header, banner (optional), body, column, card, hero (mutually exclusive with banner), footer-strip (optional), footer.
4. Use the --u unit system (1.6px screen, 1mm print) for ALL sizing via calc(N * var(--u)) — never bare px except hairlines (≤ 2px).
5. Carry the :root DESIGN TOKENS block fenced by the /* ===== DESIGN TOKENS ===== */ … /* ===== END DESIGN TOKENS ===== */ comment pair, the full --fs-1 … --fs-9 scale, and the §B utility classes.
6. Use only components from COMPONENTS.md; no inline style=, no linear-gradient, no hex outside the token block.
7. Keep all paper-specific content as TODO placeholders — neutral templates only.

Then add a row to the gallery table above, a catalog note if it introduces any new component (which requires the COMPONENTS.md new-component checkpoint), and link it in SKILL.md Phase 3.

Back to ARIS: Auto-claude-code-research-in-sleep or Agent skills.