Make pdf

Nutrient Make PDF

Turn a Markdown or HTML file into a finished PDF through the Nutrient DWS Build API. The differentiator over local chromium --print-to-pdf or Pandoc is the output stage: the same call can emit tagged, accessible PDF/UA or archival PDF/A, and stamp a diagonal text watermark — compliance work a plain HTML-to-PDF pipeline cannot do.

Quick Start

All paths below are relative to this skill's directory (the directory containing scripts/ and assets/).

1. Export the API key before generating (not needed for --html-only or the smoke test):
   • export NUTRIENT_API_KEY="nutr_sk_..." (the same variable the sibling document-processor-api scripts use; NUTRIENT_DWS_API_KEY is accepted as a fallback)
2. Generate:

uv run scripts/make-pdf.py --input report.md --out report.pdf

The markdown-it-py and nutrient-dws packages are fetched automatically on first use.

Common Tasks

• Standard PDF from Markdown: uv run scripts/make-pdf.py --input notes.md --out notes.pdf
• Accessible PDF/UA (tagged, for WCAG/ADA/EAA requirements): add --accessible
• Archival PDF/A: add --pdfa (conformance via --pdfa-level, one of pdfa-1a/1b/2a/2b/2u/3a/3u, default pdfa-2b; only valid together with --pdfa)
• Draft watermark: add --watermark DRAFT
• Memo layout instead of the default document layout: --template memo
• Dark theme: --theme dark
• Title-page metadata: --title, --subtitle, --author, --date (or set them in Markdown frontmatter; CLI flags win)
• Inspect the rendered HTML without calling the API (offline, no key): --html-only (Markdown input only — errors on .html input, which already is the HTML)
• Debug layout alongside the PDF: --output-html writes the intermediate HTML next to the PDF
• Batch: point --input at a directory (with --out <dir>, required) to convert every .md/.html in it — see Batch Conversion below

--accessible and --pdfa are mutually exclusive (one output type per build). Check exact arguments with uv run scripts/make-pdf.py --help.

Input Rules

• .md input is converted to HTML with a CommonMark+GFM parser (tables, strikethrough, task lists), then wrapped in the selected template with the theme CSS inlined — one self-contained HTML file is uploaded.
• .html input is uploaded as-is; template, theme, and metadata flags do not apply and a warning names any that were passed. Use it when you already have styled HTML.
• YAML frontmatter in Markdown (title, subtitle, author, date, template, theme) controls the cover block; CLI flags override frontmatter. When no title is given anywhere, the first # heading (or the filename) becomes the document title, so accessible output always has a real title.
• External images require network fetches on the DWS side and referenced local images are not uploaded in v1 — embed images as data: URIs if they must appear.

Verification (built in)

Compliance outputs are verified, not just labeled. With --accessible or --pdfa, the generated PDF is automatically checked after the build (--no-verify opts out); exit code 3 means "generated but failed verification" — the PDF is kept on disk so you can inspect it. If the verifier itself cannot run, the build still exits 3 rather than silently passing; --no-verify is the explicit way to accept unverified output. Passing --verify without a compliance flag is rejected up front — a standard PDF has no conformance claim to check.

Two assurance levels, and be honest about which one ran:

• Structural checks (always available, via scripts/verify-pdf.py): the PDF/UA-1 claim marker (pdfuaid XMP part 1 — PDF/UA-2 documents are out of scope), tagged-structure signals (MarkInfo, StructTreeRoot), language, DisplayDocTitle, and a non-empty document title — or the PDF/A identification (pdfaid part + conformance, matched against the requested level). These catch real failures but are not a full conformance audit.
• Full audit (optional): if verapdf is installed on PATH, the verifier also runs veraPDF for a genuine conformance validation and reports its verdict. Recommend installing veraPDF when the user's requirement is regulatory (ADA/EAA deadlines, records retention).

The verifier also works standalone on any PDF, including ones this skill did not create:

uv run scripts/verify-pdf.py --input existing.pdf --profile pdfua
uv run scripts/verify-pdf.py --input archive-dir/ --profile pdfa --pdfa-level pdfa-2b

Batch Conversion

--input <directory> converts every .md, .markdown, .html, and .htm file directly in that directory (non-recursive), up to 4 concurrently:

uv run scripts/make-pdf.py --input reports/ --out pdfs/ --accessible

• --out <dir> is required for directory input; outputs are named <stem>.pdf.
• Per-document metadata comes from each file's frontmatter, first # heading, or filename — global --title/--subtitle are rejected for directory input. Shared flags (--template, --theme, --author, --date, --watermark, compliance and verify flags) apply to every file.
• Failures don't stop the batch: a summary on stderr lists converted/verified/failed counts, stdout still prints only created file paths. Exit 0 = all good, 1 = at least one build failed, 3 = builds succeeded but at least one verification failed.
• Batch mode never overwrites existing outputs (and refuses inputs that would collide on the same output name) — point --out at a fresh directory or remove stale files when regenerating.

Layout Honesty (Chromium limits)

DWS renders HTML with headless Chromium (printToPDF). That means standard print CSS works (@page size and margins, page-break-*/break-* rules, @media print), but CSS Paged Media features Chromium does not implement will silently not render: position: running() running headers/footers, counter(page), and target-counter() TOC page numbers. Do not promise page-numbered TOCs or running headers from CSS. If the user needs those, say so and offer the trade-offs (front-matter TOC without page numbers, or a different toolchain).

Chaining Other Document Operations

Generation is one step. Related operations live in sibling plugins from the same marketplace:

• Sign, redact, or merge the produced PDF: the document-processor-api skill — /plugin install nutrient-dws@nutrient-skills. Its scripts read the same NUTRIENT_API_KEY; run them from that skill's own directory, e.g. uv run scripts/sign.py --input report.pdf --out report-signed.pdf.
• Make an EXISTING PDF accessible (auto-tag for PDF/UA): the remediate-pdf skill — /plugin install remediate-pdf@nutrient-skills. This skill generates new PDFs; that one fixes old ones (and bundles the same verify-pdf.py).

Implementation Rules

• Fail fast with a clear message when NUTRIENT_API_KEY (or the NUTRIENT_DWS_API_KEY fallback) is missing (unless --html-only).
• Never log the API key or request headers.
• Write outputs to explicit paths and print each created file path (one path per line on stdout; progress on stderr). A debug HTML path prints as soon as the file exists, even if a later build step fails.
• Non-zero exit on any failure; do not claim success unless the output file exists and is non-empty.

Validation

Before calling this skill done after changes:

• uv run scripts/smoke-test.py — offline (no network, no key): every template x theme combination, Markdown feature survival (headings, tables, code blocks), verifier checks against constructed fixtures, and batch discovery.
• Generate one real PDF if a key is available; for compliance outputs let the built-in verification run (or run verify-pdf.py explicitly) rather than only checking the file is non-empty.