pdf-creator
Converts Markdown to print-ready PDF with proper Chinese typography, themes and a visual self-check. Use for Markdown → PDF, including documents formatted for printing or mobile reading. Not for Word/WPS files or Word output (use docx-creator).
Install / Use
npx skills add daymade/claude-code-skills --skill pdf-creatorInstalls into whichever agent you are using.
SKILL.md
Installable skill definition
Quality Score
Category
Content & MediaSupported Platforms
Our assessment of pdf-creator
pdf-creator scores 93/100 on our quality scale, 219th of 1,155 Content & Media skills we index (top 19%).
Its SKILL.md is 23 KB long, well organised into 28 sections with 5 code examples: a thorough specification that gives an agent plenty to work with.
With 1,426 GitHub stars, it is one of the more widely adopted skills in the catalogue.
Maintenance, license and trust
- The repository was last updated about 2 months ago, so pdf-creator is actively maintained.
- Our last check on 2026-10-26 found the source still online.
- It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
- Its trust signals score 100/100, with no cautions. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Safety scan
No issues foundOur scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.
AI review by kimi-k2.7-code on 2026-10-04. Automated pattern scan on 2026-10-04. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.
pdf-creator compared with similar skills
All 4 of these similar skills score higher than pdf-creator; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| pdf-creator (this skill)by daymade | 93 | 1.4k | 53d ago | SKILL.md |
| siyuanby siyuan-note | 100 | 47.3k | today | MCP Server |
| x402by browser-use | 100 | 121.0k | 3d ago | SKILL.md |
| algorithmic-artby anthropics | 100 | 177.9k | 55d ago | SKILL.md |
| pptxby anthropics | 100 | 177.9k | 55d ago | SKILL.md |
Frequently asked questions
- How do I install pdf-creator?
- Run
npx skills add daymade/claude-code-skills --skill pdf-creator. The install tabs above show the steps for each supported agent. - Which AI agents does pdf-creator work with?
- It is written for Universal, as a SKILL.md file. Other agents that read the same format can often use it too.
- Is pdf-creator safe to use?
- Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It is MIT-licensed and scores 100/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is pdf-creator still maintained?
- The repository was last updated about 2 months ago, so pdf-creator is actively maintained.
Skill content
View source on GitHubname: pdf-creator disable-model-invocation: true description: >- Converts Markdown to print-ready PDF with proper Chinese typography, themes and a visual self-check. Use for Markdown → PDF, including documents formatted for printing or mobile reading. Not for Word/WPS files or Word output (use docx-creator).
PDF Creator
Create professional PDF documents from markdown with Chinese font support and theme system.
For an existing .docx source, use docx-creator/references/word-to-pdf.md instead.
Choose by the authoritative input as well as the output suffix.
Quick Start
Every scripts/… path below is relative to this skill's own directory — run the commands from there, or substitute the absolute path. Running them from the directory holding your markdown fails with Failed to spawn: 'scripts/md_to_pdf.py'. Omitting the output path is allowed and derives it from the input's basename (input.md → input.pdf, in the input's directory).
# Default theme (formal: Songti SC + black/grey, A4 print)
uv run --with weasyprint scripts/md_to_pdf.py input.md output.pdf
# Warm theme (training: PingFang SC + terra cotta)
uv run --with weasyprint scripts/md_to_pdf.py input.md --theme warm-terra
# Mobile theme (narrow page, large font — for phone reading / WeChat sharing)
uv run --with weasyprint scripts/md_to_pdf.py input.md --theme mobile
# Batch convert all markdown files with a specific theme
uv run --with weasyprint scripts/batch_convert.py *.md --theme warm-terra --no-preview
# No weasyprint? Use Chrome backend (auto-detected if weasyprint unavailable)
python scripts/md_to_pdf.py input.md --theme warm-terra --backend chrome
# List available themes
python scripts/md_to_pdf.py --list-themes dummy.md
Themes
Stored in themes/*.css. Each theme is a standalone CSS file.
| Theme | Page Size | Font | Color | Best for |
|-------|-----------|------|-------|----------|
| default | A4 | Songti SC + Heiti SC | Black/grey | Legal docs, contracts, formal reports |
| cjk-auto | A4 | Songti SC + Heiti SC | Black/grey | Tables with uneven column content (course schedules, itemized lists) |
| warm-terra | A4 | PingFang SC | Terra cotta (#d97756) + warm neutrals | Course outlines, training materials, workshops |
| warm-terra-menu | A4 | PingFang SC | Terra cotta (#d97756) + warm neutrals | warm-terra variant hardened for module menus/lists: 2-column long-text tables wrap without first-column overflow + Menlo unicode-range keeps CJK inline-code from rendering blank in Preview/Adobe |
| mobile | 148mm × 210mm | PingFang SC | Terra cotta + warm neutrals | Phone reading, WeChat sharing, on-the-go reference |
To create a new theme: copy themes/default.css, modify, save as themes/your-theme.css.
Print vs Mobile: Choose the Right Theme
| Scenario | Recommended Theme | Why |
|----------|-------------------|-----|
| Print on A4 paper, handouts, contracts | default | Standard page size, formal typography |
| Training materials, course outlines | warm-terra | Warm accent color, readable for workshop contexts |
| Send via WeChat, read on phone | mobile | Narrow page (148mm), 15px font, 1.9 line-height — comfortable on small screens |
| Both print AND mobile needed | Run twice with different themes | The skill is fast; generate both versions |
Decision rule: If the user does not specify, default to warm-terra for training/course content and default for formal documents — but pick cjk-auto over default when the document's tables have uneven column content (schedules, itemized lists), which is what that theme's table-layout: auto exists for. Ask "是否需要手机版?" only when the output channel is unclear.
Backends
The script auto-detects the best available backend from content × theme — the theme half matters because the reason CJK ever needed Chrome is a property of the theme's font stack, not of the text being Chinese:
- CJK + a Songti/Heiti theme (
default,cjk-auto) → weasyprint. These themes embed CID TrueType, which every reader renders, so Chrome buys nothing and costs the clip described below. - CJK + a PingFang theme (
warm-terra,mobile,warm-terra-menu, and any theme not on the safe list) → Chrome. weasyprint subset-embeds PingFang SC as CID Type 0C OpenType, which macOS Preview / Adobe Reader fail to render — garbled text on the recipient's device even though Chrome's own PDF viewer looks fine. - Non-CJK content → weasyprint (faster, no browser startup).
Routing is pinned by scripts/tests/test_backend_routing.py. A theme you add yourself is treated as PingFang-class until you add it to _WEASYPRINT_SAFE_CJK_THEMES in md_to_pdf.py — do that only after confirming its CJK faces are CID TrueType.
| Backend | Install | Pros | Cons |
|---------|---------|------|------|
| weasyprint | pip install weasyprint | Precise CSS rendering, no browser needed, does not clip overflow | Subsets PingFang SC as CID Type 0C — unreadable in Preview/Adobe |
| chrome | Google Chrome installed | Zero Python deps, renders PingFang correctly | Clips anything past the @page box (see below) |
Override with --backend chrome or --backend weasyprint; an explicit flag always wins over auto-detection.
The theme does not tell you which backend ran. When weasyprint is not importable, a default / cjk-auto render falls back to Chrome — with a warning on stderr, and exit 0. Read the backend= field of the Generated: line to find out which one you got; that field, not the theme name, is what decides whether the Chrome-only step below applies.
Chrome clips, it does not merely overflow
Chrome wraps each page in a re W* n clip path at the @page content box. Content past that box is still in the PDF's object layer but is never painted. Measured on A4 with margin: 2.5cm 2cm 2cm 2cm: the clip path ends at 538.90pt, while the table's right border sits at 545.18pt — so the border is silently amputated.
This is not a "wide table" problem — under default and cjk-auto it is every table. Those themes set table { table-layout: fixed; width: 100% }, so a two-column table spans the same full content width as a ten-column one and lands its right border at the same 545.18pt. Measured: a trivial | 周一 | 周二 | table clips identically to a six-column fee schedule. Content width is irrelevant; the theme's own width: 100% plus cell padding is what crosses the line.
The overflow itself is by design: the CJK typography layer sets overflow-wrap: normal precisely so content overflows rather than breaking mid-token (see "CJK Typography" below). That trade-off is safe under weasyprint and destructive under Chrome.
The reason this survived repeated delivery is not that the preview lies — it is that the symptom looks deliberate. The last column's text is complete and correctly spaced; only a hairline border is gone, which reads as a styling choice. Meanwhile the visual checklist below primes you to look for text cut off, which is exactly what does not happen here.
Rasterisers do honour the clip, so the preview PNGs this script already generates (pdftoppm at 130 dpi) do show the defect, as does a 400 dpi render — at 400 dpi, zero ink across the 25 pixel columns straddling the expected border. What does not show it is any check that reads coordinates instead of pixels — pdfplumber still reports a rect at 545.18pt, because the object is genuinely there.
Rather than rely on noticing a missing hairline, run the check. It takes two forms, and for a Chrome-rendered PDF only the second one is a verdict.
# Form 1 — one file. Compares ink against the PDF's own object layer.
uv run --with pdfplumber --with pillow --with numpy scripts/check_table_borders.py out.pdf
Form 1 takes each detected table's column boundaries, counts the ones that actually have ink in a pdftoppm raster, and exits non-zero naming any boundary present in the PDF but absent on paper. It reports how many tables it measured, and says NOTHING CHECKED rather than PASS when it found none — a document whose table it could not detect has not been cleared, it has been skipped.
The boundaries come from the cell grid rather than from raw vertical edges, because a <hr>'s end-caps and an inline <code> span's background look exactly like column rules to an edge-based reader. Measured on 49 WeasyPrint-produced PDFs from a real knowledge base, 46 of which had geometry the edge-based version examined: it failed 34 of those 46. Residual limit: on PDFs from other tools, pdfplumber sometimes assembles decoration into a table that was never there, and the check will report rules for it. Treat a failure on a PDF this skill did not produce as a prompt to look, not a verdict.
It cannot clear a Chrome render on its own, because the clip destroys evidence two different ways. Chrome keeps geometry that straddles the clip — form 1 catches that, since the rule is promised and unpainted — but drops geometry lying entirely outside it, and a rule that was never written into the object layer is never looked for. Measured: a table styled with vertical rules and no cell fills prints 5/5 promised rules painted — PASS while its right border is genuinely gone. The bundled themes escape that only by accident; their cell background fills straddle the clip and leave an edge at the border's position for form 1 to find.
So when the theme routes to Chrome and the document has tables, render the same source with the other backend and compare:
# Form 2 — render a reference with the other backend, then compare.
# Use the SAME --theme as the deliverable; only the backend differs.
uv run --with weasyprint scripts/md_to_pdf.py doc.md /tmp/ref.pdf \
--theme warm-terra --backend weasyprint --no-preview
uv run --with pdfplumber --with pillow --with numpy \
scripts/check_table_borders.py out.pdf --reference /tmp/ref.pdf
Exit codes: 0 passed, 1 at least one check did not pass, 2 the check could not run, 3 no table was detected so nothing was measured. 3 is not a pass: if the document does have a table, its style drew no rules pdfplumber could find, so this check cannot speak for it — fall back to reading the raster at the table's right edge yourself, and say in your handoff that the gate did not run. pdfplumber prints Could not get FontBBox from font descriptor for subset CJK fonts on the way; that is parser noise, not a finding.
The reference is a measuring stick, never a deliverable — it may well have the CID Type 0C problem that sent this theme to Chrome in the first place, which does not affect its geometry.
Both files are ink-checked, and the rule counts are compared in both directions, so the order you pass them in cannot decide whether the damaged file gets examined. That matters more than it sounds: for a border that Chrome clipped rather than dropped, the two renders have the SAME rule count — the geometry is still there, merely unpainted — so the count comparison sees nothing, and only the ink check on the right file finds it. Before the reference was ink-checked too, passing the clipped file as --reference produced an unqualified PASS.
The count comparison, in both directions: fewer rules than the reference means the renderer dropped some; more means the arguments are swapped or the two files are not the same document. Counts rather than positions, and document-wide rather than per page, because the two backends break the same source differently — measured on one 60-row CJK table, whose row content decides where the breaks land: 5 pages vs 3 for default, 3 vs 15 for warm-terra-menu, with the rule counts identical throughout.
Calibrated across 5 themes × single- and multi-page × both argument orders, 20 runs: the 12 pairs containing no clipped file pass in both orders, and the 8 that do are caught in both orders. Separately, 14 real markd
Truncated for display — read the full file on GitHub.
Related Skills
siyuan
47.3kAn open-source, privacy-first, self-hosted knowledge workspace where humans and AI agents work together 开源、隐私优先、自托管的知识工作空间,让人与智能体在此协作
x402
121.0kSet up Browser Use Cloud payments with x402 — pay per request from a crypto wallet (USDC on Base mainnet), no signup or API key.
algorithmic-art
177.9kCreating algorithmic art using p5.js with seeded randomness and interactive parameter exploration. Use this when users request creating art using code, generative art, algorithmic art, flow fields, or particle systems.
pptx
177.9kUse this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an em…
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
