Skip to content

About

High-performance, zero-dependency CLI for the pdfnative engine. Automate JSON-to-PDF/A rendering, digital signatures (RSA/ECDSA), and compliance inspection within CI/CD pipelines via shell-pipe workflows.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

95 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

pdfnative-cli

CI CodeQL npm version npm downloads zero extra runtime dependencies TypeScript License: MIT npm provenance OpenSSF Scorecard

pdfnative website

Official CLI for the pdfnative engine — render JSON to PDF (32 Unicode scripts, right-to-left text, PDF/A, PDF/X-4, spot colours, JPEG 2000, charts, typography), sign / verify / archive signatures (PAdES, LTV), merge / split / extract pages, fill forms, encrypt, inspect, extract text and compare — from a terminal, a CI job or an AI agent (--json, 12 stable error codes, --dry-run, llms.txt). Zero extra runtime dependencies.

What's new in v1.6.0 — built on pdfnative 1.9.0. render speaks right-to-left (--direction auto|ltr|rtl, full UAX #9: Arabic and Hebrew paragraphs aligned and justified in visual order, extracted in logical order), adds five Indic scripts (gu pa kn ml or — Gujarati, Punjabi, Kannada, Malayalam, Odia: 32 Unicode scripts), spot colours (layout.print.spotColours, { spot, tint } and { inks } values → /Separation and /DeviceN), the PDF/A-2 + PDF/X-4 dual claim in one file (--tagged pdfa2b --pdfx pdfx4), JPEG 2000 images (.jp2 / .j2k passed through as /JPXDecode), per-value chart colours and four new diagnostics (13 under --strict). extract-text / compare gain --on-stream-error limit|throw|skip (an over-cap stream is E_INPUT, never a silently empty page); ltv embed composes /DSS from material you already hold (--cert / --ocsp / --crl, --match all|chain) with no network at all; annotate link and text-string decoding come from the engine. The repository now proves its compatibility promise on every pull request (the v1.5.0 suite and schema captures replayed against the current sources), runs the gate on Ubuntu, Windows and macOS, lints its workflows, fuzzes its parsers with ClusterFuzzLite and publishes through a four-job guard → build → publish → attest pipeline. Zero breaking changes. See release notes and docs/AGENT_CONTRACT.md.

⭐ Star pdfnative — the zero-dependency PDF engine that powers this CLI.

Highlights

  • Right-to-left & 32 scripts (v1.6.0, pdfnative ≥ 1.9.0) — render --direction auto|ltr|rtl (or a per-paragraph direction) lays Arabic, Hebrew and mixed paragraphs out under the full Unicode Bidirectional Algorithm — right-aligned, justified in visual order, punctuation on the right side — and extract-text returns them in logical order even from untagged output (/Span /ActualText; --actual-text false restores the bare runs). --font gu|pa|kn|ml|or add Gujarati, Gurmukhi, Kannada, Malayalam and Odia: 32 Unicode scripts, 36 bundled font modules.
  • Spot colours, the dual claim & JPEG 2000 (v1.6.0) — declare inks in layout.print.spotColours and paint with { "spot": "Pantone 300 C" }, { spot, tint } or an { inks } mix (/Separation / /DeviceN; an undeclared ink is E_INPUT); claim PDF/A-2b or 2u and PDF/X-4 in one file (--tagged pdfa2b --pdfx pdfx4, checked by inspect --check pdfa,pdfx); embed .jp2 / .j2k images and watermarks untouched (/JPXDecode, PDF 1.5; refused under PDF/A-1 by a diagnostic).
  • render — pipe a JSON document into a production-ready PDF. Encryption (AES-128/256), watermarks (text + image), page templates, PDF/A archival, 32 Unicode scripts + COLRv1 colour emoji + a bundled math font (--font math), native vector charts (bar, barH, line, pie, donut — pdfnative ≥ 1.6.0), PDF bookmarks (--outline), layout introspection (--inspect-layout / --debug-layout), streaming (single-pass, page-by-page, or true constant-memory --stream-true), and a hybrid flags + --layout file.json model.
  • Typography (v1.5.0, pdfnative ≥ 1.8.0) — layout.typography brings paragraph breaking (widows/orphans, keepWithNext, splittable paragraphs), align: "justify", optical margin alignment, soft hyphens, French punctuation spacing with unit binding (150 € never breaks), kerning, OpenType features (onum, smcp, tnum, …) and exact base-14 metrics. Four high-frequency flags — --split-paragraphs, --keep-headings-with-next, --kerning, --font-features <tag,…> — cover the common cases; the rest rides in --layout.
  • CMYK & PDF/X-4 (v1.5.0) — every colour accepts CMYK ("c m y k" or [c,m,y,k]), printer's marks gain colour bars, and render --pdfx pdfx4 --output-intent-icc <cmyk.icc> --trapped false writes a PDF/X-4 file whose claim inspect --check pdfx re-validates with pdfnative's structural validator (PDFX_* diagnostics, --strict gating). A 25-file conformance corpus (PDF/A via veraPDF, PDF/X via the validator) guards every release.
  • Reproducible output (v1.5.0) — the global --creation-date <iso8601> (or SOURCE_DATE_EPOCH) pins /CreationDate, xmp:CreateDate, the {date} placeholder and the trailer /ID, all in UTC, so the same input renders to byte-identical bytes on every host and in every timezone. The repository's own 99-sample baseline is held to that promise.
  • 27 scripts & custom fonts (v1.5.0) — --font lo|nod|khb|tdd|cjm add Lao, Tai Tham, New Tai Lue, Tai Le and Cham; ha/yo/ig/sw alias latin for Hausa, Yoruba, Igbo and Swahili; --font-file <path.ttf>[:name] registers your own TrueType/OpenType file (32 MiB cap, magic-byte and parser validation — never from a JSON payload). --variant table embeds fonts too, so table renders can claim PDF/A.
  • extract-text / fill / encrypt / decrypt (v1.3.0, pdfnative ≥ 1.6.0) — extract reading-order text as text/JSON/NDJSON for RAG & agents (no OCR); fill, flatten & export AcroForms with an incremental save (signatures stay valid; fill --export gives a read→edit→fill round-trip); re-secure with AES-128/256 or remove encryption. merge / split / extract gain --password, --encrypt, and constant-memory --stream.
  • doctor — an offline environment/capability preflight (CLI/Node/pdfnative versions, Web Crypto CSPRNG required by encrypt, command count, and — v1.5.0 — the bundled font inventory probed on disk, the Unicode version and the PDF/A + PDF/X conformance targets). Text or --json; exit 0/1 — an ideal agent pre-flight.
  • sign — CMS/PKCS#7 digital signatures with full metadata (--reason, --name, --location, --contact, --signing-time) and intermediate CA chains via --cert-chain (repeatable). Uses native node:crypto for constant-time signatures by default (--pure-crypto opts out). Keys loaded from env vars or files; never logged. v1.4.0 adds a functional RFC 3161 --timestamp <tsa-url> (PAdES B-T), PAdES profile (--profile pades — ETSI.CAdES.detached, ESS signing-certificate-v2), sha384/512 digests (RSA), multi-signatures (--allow-multiple / --field-name), and visible signature placement (--signature-rect / --signature-page); v1.5.0 adds --timestamp-timeout <ms> for the TSA round-trip.
  • PAdES long-term signatures (v1.4.0, pdfnative ≥ 1.7.0) — the full B-T → B-LT → B-LTA ladder: sign --timestamp <tsa> --profile pades (B-T) → ltv add --online (B-LT: OCSP + CRL validation data archived into /DSS + /VRI) → doc-timestamp --url <tsa> (B-LTA: a /DocTimeStamp revision, repeatable for renewal) → ltv add --online. The ltv collect / ltv embed split supports air-gapped pipelines — collect on a connected machine, embed fully offline; since v1.6.0 ltv embed --cert <pem|der> --ocsp <der> --crl <der> (repeatable, --match all|chain) composes the /DSS from material you already hold, with no collect step at all. All network access is explicit opt-in and SSRF-guarded.
  • metadata / compare (v1.4.0) — update /Info + XMP metadata with an incremental save (existing signatures stay valid); diff two PDFs by text + structure with CI-friendly exit codes (identical → 0, different → 1 / E_CHECK_FAILED).
  • batch --manifest (v1.4.0) — a declarative, sequential multi-command pipeline (render → sign → verify → …) in one JSON file, with @<id> output references and an --allow-network gate so untrusted manifests can never trigger network I/O.
  • Print production & charts v2 (v1.4.0, pdfnative ≥ 1.7.0) — render layout gains bleed/trim/art/crop boxes, printer marks, userUnit, ICC output intents, and viewer preferences (duplex, numCopies, …); charts grow to 9 types (stacked bars, area, scatter) with dual axes, log/time/category x-axes, and data labels. render --strict turns PDF/A conformance diagnostics into a hard error before the first output byte.
  • inspect — PDF version, page count, encryption, PDF/A and PDF/X conformance, signature count, metadata (--iso-dates normalises PDF dates to ISO 8601 — v1.5.0), page labels, markup/link annotations (--annotations), PDF/UA (ISO 14289-1) structural validation and a PDF/X-4 validation report (--pdfx). --verbose, --pages, and --check pdfa|signed|encrypted|pdfua|pdfx for CI assertions.
  • verify — verify every CMS/PKCS#7 signature: byte-range integrity, RSA/ECDSA signature value (RSA now also sha384/512), certificate chain, trust roots, RFC 3161 timestamp (PAdES-T), /DocTimeStamp revisions (PAdES B-LTA), and OCSP + CRL revocation (embedded /DSS offline by default, opt-in SSRF-guarded online). JSON & text output, --strict, --revocation, --revocation-policy; a SHA-1 timestamp imprint is reported as a weak digest and refused under --strict (v1.5.0).
  • merge / split / extract — page-tree operations (pdfnative ≥ 1.5.0): concatenate several PDFs, split one PDF into many (per-page or per-range), or pull selected pages into a new PDF. --drop-annotations and --max-output-size guards included.
  • annotate — attach markup annotations (highlight, text note, square, line, …) and — v1.5.0 — link annotations (url, validated: http/https/mailto only) to an existing PDF via a JSON spec, using an incremental save so the original bytes — and any existing signature — stay intact.
  • govern — expose pdfnative's AI-governance / HITL contract: govern rules, govern policy, and govern verify-issue <draft.md> to gate an issue/PR draft (exit 1 / E_POLICY on violation) before a human reviews and submits it.
  • batch — render every JSON file in a directory to PDF in parallel, reusing the full render pipeline, with a per-file summary and bounded --concurrency.
  • completion — emit bash, zsh, fish, or powershell shell-completion scripts.
  • schema — print a versioned JSON Schema (Draft 2020-12) for any CLI input/output shape, plus schema manifest (a machine-readable capability manifest) and llms.txt, so agents can self-validate and discover the CLI's tools.
  • Agent-native — a global --json status/error envelope, 12 stable E_* error codes, and a --dry-run validation mode let autonomous AI agents and CI drive the CLI deterministically; global flags may precede or follow the command name (v1.5.0). Token-economy levers — --summary (minimal verdict), --fields (dot-path projection), and compact JSON under --json — shrink agent output ~90 %. See docs/AGENT_CONTRACT.md.
  • Engineered like the engine (v1.5.0, aligned with 1.9.0 in v1.6.0) — one quality gate (npm run gate) with fast / CI / publish profiles, run as required checks on Ubuntu (Node 22 and 24), Windows and macOS; a byte-exact sample baseline (test-output/samples/ vs tests/regression/baselines/samples.sha256.json), a 25-file PDF/A + PDF/X conformance corpus, a compatibility promise proved on every pull request (the previous release's test suite and schema captures replayed against the current sources — SECURITY.md §Compatibility promise), ClusterFuzzLite over the CLI's own parsers, hardened and linted workflows (SHA-pinned actions, zizmor + actionlint, CodeQL), a guard → build → publish → attest release pipeline (Trusted Publishing, SBOMs, SLSA Build L2 attestations), a documentation verifier (npm run verify:docs) and a committed Claude Code layer with a human-in-the-loop guard hook — 1555 tests.
  • AI-governance / HITL — the govern command surfaces pdfnative's Human-in-the-Loop contract to agents: they act as draftsmen, never autonomous submitters. govern verify-issue gates a local draft; a human always reviews and submits.
  • .pdfnativerc.json — optional config file for default flags (global + per-command); precedence is CLI flags > env > config.
  • Zero extra dependencies — pdfnative is the sole runtime dependency.
  • Offline by default — no network access unless you explicitly opt in with verify --revocation online, sign --timestamp, ltv --online, doc-timestamp --url, or batch --allow-network — and every request passes an SSRF guard (no redirects).
  • Stdin / stdout by default — every command is shell-pipeline friendly.
  • Secret-safe — signing keys, certs, encryption passwords never appear in error output or stderr. PEM material redacted; layout-file attachments[].data injection blocked.
  • ESM-first, TypeScript strict — built with tsup, typed declarations included.
  • NPM provenance — signed builds via GitHub Actions OIDC.

Supported Features

Feature Status Notes
Commands
render JSON → PDF ✅ Streaming, hybrid layout model, 32 scripts incl. right-to-left, bookmarks, layout introspection, PDF/A + PDF/X-4 (jointly since v1.6.0), spot colours, JPEG 2000
sign digital signatures ✅ RSA/ECDSA (CMS/PKCS#7), metadata fields, cert chains, native node:crypto
inspect PDF metadata ✅ --verbose, --pages, --pdfua, --pdfx, --iso-dates, --annotations, --form-fields, --encryption, --password, page labels, --check …
verify signature verification ✅ Integrity + chain + trust + timestamp + revocation; --strict
extract-text reading-order text ✅ --format text|json|ndjson, --runs, --pages, --password, --on-stream-error limit|throw|skip (v1.6.0); right-to-left and shaped text in logical order; RAG/agent-native (no OCR)
fill fill / flatten / export AcroForms ✅ --data <values.json>, --flatten, --export (values → --data map); incremental save
encrypt / decrypt ✅ AES-128/256 re-encryption & transparent decryption (--password, --stream)
doctor preflight ✅ CLI/Node/pdfnative versions, Web Crypto (CSPRNG), command count, font inventory, Unicode version, conformance targets; text or --json
merge concatenate PDFs ✅ Page-tree API; --password, --encrypt, --stream, --drop-annotations, --max-output-size
split one PDF → many ✅ Per-page (default) or per-range (--pages); --password, --encrypt, --stream
extract selected pages ✅ 1-based --pages list/range; --password, --encrypt, --stream
annotate markup + link annotations ✅ Incremental save (signatures preserved); JSON spec via --annotations; link type with URL validation (v1.5.0), its body built by the engine (v1.6.0)
govern AI-governance / HITL ✅ rules / policy / verify-issue; gates drafts with E_POLICY
batch parallel rendering ✅ Directory → PDFs, --concurrency, --fail-fast
completion shell scripts ✅ bash / zsh / fish / powershell
schema JSON Schema export ✅ Per-command schemas + summaries + status + manifest (capability manifest)
.pdfnativerc.json config file ✅ Global + per-command defaults; flags > env > config
Agent / automation
Global --json envelope ✅ Status on success, { ok, error: { code, message } } on failure
Stable error codes ✅ 12 codes: E_USAGE, E_INPUT, E_PARSE, E_IO, E_SIGN, E_VERIFY_FAILED, E_CHECK_FAILED, E_POLICY, E_UNSUPPORTED, E_PASSWORD, E_NETWORK, E_RUNTIME
Global flags before the command ✅ pdfnative --json --dry-run render … ≡ pdfnative render … --json --dry-run (v1.5.0)
Reproducible output ✅ Global --creation-date <iso8601> / SOURCE_DATE_EPOCH → byte-identical PDFs across hosts and timezones (v1.5.0)
Capability manifest ✅ schema manifest (JSON) + llms.txt — for agent tool discovery
--dry-run validation ✅ render / sign / batch / merge / split / extract / annotate / fill / encrypt / decrypt / metadata / ltv / doc-timestamp
Document Blocks
Headings, paragraphs, lists ✅ Full text styling support
Tables ✅ Headers, rows, multi-page
Barcodes ✅ QR, Code 128, EAN-13, Data Matrix, PDF417
Charts ✅ chart block: bar, barH, line, pie, donut (native PDF paths, pdfnative ≥ 1.6.0)
Hyperlinks ✅ URL validation, blue underlined text
Form fields ✅ Text, checkbox, radio, dropdown, listbox
Page breaks, spacers ✅ Explicit pagination control
Table of contents ✅ Auto-generated with /GoTo links
Advanced Layouts (v0.2.0)
PDF/A archival (1b, 2b, 2u, 3b) ✅ --tagged pdfa<level> (preferred) or --conformance (deprecated); validated against the veraPDF reference validator in CI (blocking)
Streaming output ✅ --stream (single-pass) for large documents
Compression ✅ --compress flag
Encryption (AES-128/256) ✅ --encrypt-* flags + env-var precedence
Watermarks (text + image) ✅ --watermark-text, --watermark-image, --watermark-position
Headers / footers with placeholders ✅ --header-{l,c,r}, --footer-{l,c,r}, {page}/{pages}/{date}/{title}
Custom page sizes ✅ --page-size A4|Letter|… or WxH in points
Custom margins ✅ --margin <N> or --margin <t,r,b,l>
PDF/A-3 attachments ✅ --attachment <path>:<mime>:<rel>:<desc> (repeatable)
Multilingual fonts ✅ 32 Unicode scripts via --font <code> --lang <code> (e.g. th, ja, ar, te, si, km, lo, cjm); Latin built-in
Table-centric variant (PdfParams) ✅ --variant table
Full PdfLayoutOptions ✅ --layout <file.json>
Signing (v0.2.0)
RSA signatures (rsa-sha256) ✅ Default algorithm
ECDSA signatures (ecdsa-sha256) ✅ P-256 SEC1 / PKCS#8 keys (v0.3.0)
Auto signature-placeholder injection ✅ One-command sign of any rendered PDF (v0.3.0)
Signature metadata ✅ --reason, --name, --location, --contact, --signing-time
Cert chains (intermediate CAs) ✅ --cert-chain <pem> (repeatable) or PDFNATIVE_SIGN_CHAIN env
Verification (v0.2.0+)
Byte-range integrity (SHA-256) ✅ Recomputed and compared with CMS messageDigest attribute
CMS signature-value verification ✅ RSA-SHA256 + ECDSA-SHA256 (v0.3.0)
Certificate chain verification ✅ Via pdfnative verifyCertSignature
Trust roots ✅ --trust <root.pem> (repeatable) + self-signed acceptance
RFC 3161 timestamp recognition ✅ Reported as timestampPresent
RFC 3161 timestamp validation (PAdES-T) ✅ TSA signature, messageImprint binding, chain, genTime
OCSP revocation (RFC 6960) ✅ Embedded /DSS + opt-in online via AIA (SSRF-guarded)
CRL revocation (RFC 5280) ✅ Embedded /DSS + opt-in online via CDP (SSRF-guarded)
Revocation policy ✅ --revocation offline|online|disabled, --revocation-policy soft-fail|strict
Sign-side LTV (timestamp embedding / DSS) ✅ v1.4.0: sign --timestamp, ltv collect|embed|add, doc-timestamp (see below)
Render iteration
Smart tables ✅ --table-wrap, --repeat-header, --zebra, --cell-padding, --min-row-height
Page-by-page streaming ✅ --stream-page-by-page (TOC- and {pages}-compatible)
True constant-memory streaming ✅ --stream-true (parts freed as emitted; byte-identical output)
Configurable block cap ✅ --max-blocks <n> (default 100 000)
PDF/UA structural validation ✅ inspect --pdfua / --check pdfua (ISO 14289-1) — developer-time gate, not a substitute for veraPDF
--watch re-render on file change ✅ 200 ms debounce, requires file --output
--template <file.json> ✅ Deep-merge base under input (caller wins)
--font bundled shortcuts ✅ Repeatable allow-list: latin, emoji, color-emoji, math, 32 script codes + 4 latin aliases (ha, yo, ig, sw)
Page-tree & annotations (v1.2.0)
Merge PDFs ✅ merge <a.pdf> <b.pdf> … (or repeatable --input) → --output
Split PDF ✅ split --output-dir <dir> per-page, or --pages 1-2,3-4 per-range
Extract pages ✅ extract --pages 4,1-2 (1-based; order preserved, repeats allowed)
Drop annotations on copy ✅ --drop-annotations on merge / split / extract
Output-size guard ✅ --max-output-size <bytes> on merge / split / extract
Markup annotations ✅ annotate --annotations <spec.json> — highlight, text, square, circle, line, freetext, …
Incremental save ✅ annotate preserves original bytes (existing signatures stay valid)
Inspect annotations ✅ inspect --annotations lists markup + link annotations
Page labels ✅ inspect reports /PageLabels when present
Bookmarks & layout (v1.2.0)
PDF bookmarks (outline) ✅ render --outline auto (from headings) or --outline <tree.json>
Math / technical symbols ✅ render --font math (Noto Sans Math; auto-routed code points)
Layout inspection report ✅ render --inspect-layout → LayoutInspection JSON (no PDF)
Layout debug overlay ✅ render --debug-layout [margins,content,cells]
Signing crypto (v1.2.0)
Native constant-time crypto ✅ sign uses node:crypto by default (side-channel-resistant)
Pure-JS crypto fallback ✅ sign --pure-crypto forces pdfnative's portable bignum path
AI-governance / HITL (v1.2.0)
Governance rules / policy ✅ govern rules (protocol) / govern policy (machine-readable JSON)
Draft verification gate ✅ govern verify-issue <draft.md> → exit 1 / E_POLICY on violation
Text, forms, encryption & charts (v1.3.0, pdfnative ≥ 1.6.0)
Text extraction ✅ extract-text --format text|json|ndjson (reading order, --runs, --password); no OCR
Fill AcroForms ✅ fill --data <values.json> (incremental save; encrypted PDFs via --password)
Flatten AcroForms ✅ fill --flatten (stamp appearances, remove interactive fields)
Export form values ✅ fill --export → --data-shaped JSON map (read → edit → fill round-trip)
Encrypt / decrypt ✅ encrypt --owner-password … (AES-128/256) / decrypt --password … (--stream)
Encrypted page-tree sources ✅ --password on merge / split / extract / inspect
Re-encrypt page-tree output ✅ --encrypt [aes-128|aes-256] on merge / split / extract
Constant-memory streaming ✅ --stream (+ --chunk-size) on merge / split / extract / encrypt / decrypt
Native vector charts ✅ chart document block: bar, barH, line, pie, donut (pure PDF paths, tagged /Figure)
List form fields / encryption ✅ inspect --form-fields / inspect --encryption
Unified render encryption flags ✅ render --encrypt [aes-128|aes-256] --owner-password … (same vocab as merge/split/extract)
Environment preflight ✅ doctor (versions, Web Crypto/CSPRNG, command count; text or --json)
Capability manifest ✅ schema manifest + llms.txt for agent tool discovery
PowerShell completion ✅ completion powershell
Long-term signatures & document ops (v1.4.0, pdfnative ≥ 1.7.0)
RFC 3161 signing timestamp (PAdES B-T) ✅ sign --timestamp <tsa-url> (+ --timestamp-digest, --timestamp-nonce); opt-in network, SSRF-guarded
PAdES signature profile ✅ sign --profile pades (ETSI.CAdES.detached, ESS signing-certificate-v2)
CMS digest selection ✅ sign --digest sha256|sha384|sha512 (RSA; ECDSA is sha256-only)
Multiple signatures ✅ sign --allow-multiple + --field-name (default stays idempotent single-signature)
Visible signature placement ✅ sign --signature-rect "x1,y1,x2,y2" + --signature-page (+ --placeholder-bytes)
LTV validation data (PAdES B-LT) ✅ ltv collect|embed|add — OCSP + CRL into /DSS + /VRI; collect needs --online, embed is fully offline (air-gapped)
Document timestamp (PAdES B-LTA) ✅ doc-timestamp --url <tsa> — /DocTimeStamp (ETSI.RFC3161) incremental revision, repeatable for renewal
Verify B-LTA / sha384-512 ✅ verify validates /DocTimeStamp revisions as RFC 3161 tokens; accepts rsa-sha384 / rsa-sha512; reports fieldName + isDocTimestamp
Signature inventory ✅ inspect --signatures (fieldName, subFilter, byteRange, isDocTimestamp, isPlaceholder — never signature bytes) + --check "signatures>=N"
Metadata editing ✅ metadata --title/--author/--subject/--keywords/--mod-date or --from-json — incremental save, signatures stay valid
PDF comparison ✅ compare a.pdf b.pdf --mode text|structure|both — exit 0 identical, exit 1 / E_CHECK_FAILED on differences; no visual diff
Manifest pipelines ✅ batch --manifest tasks.json — sequential fail-fast tasks, @<id> references, --allow-network gate, --continue-on-error
PDF/A strict gating ✅ render --strict — conformance diagnostics become E_CHECK_FAILED before any output byte (else stderr warnings + diagnostics[] under --json)
Charts v2 ✅ 9 types (+ stackedBar, stackedBarH, area, scatter), series.xValues, yAxis: right + axis2, log scale, category/linear/time x-axis, data labels, label stride/rotation
Print production ✅ layout.print (bleed, trimBox, bleedBox, artBox, cropBox, marks, userUnit 1–75000), layout.outputIntent (ICC RGB), layout.viewerPreferences (duplex, pickTrayByPDFSize, printPageRange, numCopies)
Image blocks by path ✅ { "type": "image", "src": "logo.png" } (relative to the input JSON) or dataBase64
Encrypted annotate ✅ annotate --password — appended objects re-encrypted under the existing scheme
Page-box preservation ✅ merge / split / extract now preserve BleedBox/TrimBox/ArtBox/UserUnit (pdfnative ≥ 1.7.0; previously dropped)
Anti zip-bomb cap ✅ Global --max-inflate-size <bytes> — cap on any decompressed stream while parsing (default 100 MiB)
Network error code ✅ Stable E_NETWORK — opt-in network operation failed (TSA / OCSP / CRL fetch)
Typography, PDF/X-4, CMYK & reproducible output (v1.5.0, pdfnative ≥ 1.8.0)
Typography engine ✅ layout.typography — widows/orphans, keepWithNext, splittable paragraphs, paragraph align: "justify", optical margins, soft hyphens (U+00AD, always honoured), punctuation spacing ("fr" / "fr-CA" preset or a rules array + unit binding), kerning, OpenType fontFeatures, exact base-14 metrics
Typography flags ✅ render --split-paragraphs, --keep-headings-with-next, --kerning, --font-features <tag,…> (merged into layout.typography; flags win over --layout)
CMYK colours ✅ Every colour field and flag accepts "c m y k" (0–1) or [c,m,y,k] (percent) beside hex / RGB; DeviceCMYK operators emitted
Printer's colour bars ✅ layout.print.marks.colourBars: true | { tints, size }
PDF/X-4 output ✅ render --pdfx pdfx4 --output-intent-icc <cmyk.icc> [--output-intent-id <s>] [--trapped true|false|unknown]; coherence errors → E_INPUT, PDFX_* diagnostics → --strict / E_CHECK_FAILED; envelope carries pdfx
PDF/X check ✅ inspect --pdfx (full report) and --check pdfx (exit 0/1); pdfxConformance always reported; --summary adds pdfx
32 Unicode scripts ✅ + lo (Lao), nod (Tai Tham), khb (New Tai Lue), tdd (Tai Le), cjm (Cham); ha / yo / ig / sw alias latin
Custom fonts ✅ render --font-file <path.ttf>[:name] (repeatable; 32 MiB cap, magic bytes, parseFontData + validateFontData; never from JSON)
Table variant fonts ✅ --variant table --font latin --lang latin embeds fonts, so table renders can claim PDF/A
Reproducible dates ✅ Global --creation-date <iso8601> (or SOURCE_DATE_EPOCH) pins /CreationDate, xmp:CreateDate, {date}, trailer /ID, in UTC; envelope carries creationDate
Layout revival ✅ --layout files and inline layout accept creationDate (ISO string), outputIntent.iccProfile (number[]), nested typography / outputIntent merges one level
ISO dates in inspect ✅ inspect --iso-dates normalises metadata.creationDate / modDate from D:YYYY… to ISO 8601
Link annotations ✅ annotate type link (rect + url; http / https / mailto, validated)
TSA timeout ✅ sign --timestamp-timeout <ms> (with --timestamp); envelope timestamp.timeoutMs
Weak timestamp digest ✅ verify reports timestampDigest and notes a SHA-1 messageImprint; refused under --strict
Global flags first ✅ pdfnative --json --dry-run <command> … — every global flag is accepted before the command name
doctor capabilities ✅ fonts (36 modules / 32 scripts probed on disk), unicode (USE_UNICODE_VERSION), conformance (pdfa1b,pdfa2b,pdfa2u,pdfa3b,pdfx4)
SSRF ranges ✅ Benchmarking 198.18.0.0/15, TEST-NET-1 192.0.2.0/24 and NAT64 64:ff9b::/96 join the blocked ranges
Layout file cap ✅ --layout files share the 50 MB JSON cap; ICC profiles 16 MiB (acsp signature checked)
Right-to-left, 32 scripts, spot colours & the dual claim (v1.6.0, pdfnative ≥ 1.9.0)
Right-to-left (UAX #9) ✅ render --direction auto|ltr|rtl → layout.direction; a paragraph's own direction wins; Arabic / Hebrew paragraphs right-aligned and justified in visual order, punctuation and numbers placed per UAX #9; envelope carries direction
Logical-order extraction ✅ Untagged shaped and right-to-left runs carry /Span /ActualText (--actual-text, default on; false restores the 1.8.0 runs and the envelope says actualText: false), so extract-text returns the source string; subs / sups glyphs extract as their source characters
Five Indic scripts ✅ --font gu|pa|kn|ml|or (Gujarati, Gurmukhi / Punjabi, Kannada, Malayalam, Odia) — 32 Unicode scripts, 36 font modules; doctor probes them all
Spot colours ✅ layout.print.spotColours: [{ name, alternate: cmyk }]; colour values { spot }, { spot, tint }, { inks: { name: tint } } on text, tables, charts (colors per value); /Separation and /DeviceN with /Colorants; an undeclared ink → E_INPUT
PDF/A-2 + PDF/X-4 dual claim ✅ --tagged pdfa2b|pdfa2u --pdfx pdfx4 --output-intent-icc <cmyk.icc> in one file (two OutputIntents, pdfaid + pdfxid in the XMP); pdfa1b / pdfa3b with --pdfx are refused by the engine (E_INPUT); inspect --check pdfa,pdfx verifies both; envelope carries tagged and pdfx
JPEG 2000 images ✅ image blocks and --watermark-image accept .jp2 (JP2 container) and .j2k (raw codestream): embedded untouched as /JPXDecode, PDF 1.5; PDFA1_JPX_IMAGE (refused under PDF/A-1) and PDFA_JPX_IMAGE (a JPX with its own colour space under PDF/A-2/3) diagnostics
Chart colours & labels ✅ series[].colors (one colour per value, spot colours included); a series label is optional (an unlabelled series renders as Series N)
Thirteen diagnostics ✅ + PDFA_GLYPH_WITHOUT_UNICODE (PDF/A-2u / 3b text with no Unicode mapping), TEXT_GLYPH_NOT_COVERED (a character no registered font covers — the base-14 fallback would draw a box); all escalated by --strict
Stream-limit policy ✅ extract-text / compare --on-stream-error limit|throw|skip: a page or font stream past --max-inflate-size is E_INPUT (default), every undecodable stream fails (throw), or the page reads as empty text (skip, the 1.8.0 lenience) — classified by the engine's PdfStreamLimitError, never by its message
Offline LTV material ✅ ltv embed --cert <pem|der> --ocsp <der> --crl <der> (repeatable; PEM bundles and DER accepted, 16 MiB cap each) and --match all|chain compose /DSS + /VRI through the engine's composeLtvData; exclusive with --data; envelope carries match; no network, ever
Engine-built /Link bodies ✅ annotate type link is buildAnnotationBody({ type: 'link', rect, url }) — the CLI builds no PDF dictionary itself; inspect / compare decode /Info strings with the engine's decodePdfTextString (PDFDocEncoding and UTF-16)
Table-variant guard ✅ --variant table refuses a malformed PdfParams (infoItems must be { label, value } objects, rows { cells }) with E_INPUT instead of an engine TypeError
Compatibility promise ✅ Every command, flag, error code, schema subject and envelope field of 1.x is kept; the v1.5.0 suite and schema captures are replayed on every pull request (compat-previous), see SECURITY.md

Note: everything listed works today. Planned work is tracked in ROADMAP.md.

PDF/A & PDF/X status (v1.6.0)

The CLI's PDF/A outputs are validated against the veraPDF reference validator in CI (blocking), and its PDF/X-4 outputs against pdfnative's structural PDF/X validator: a corpus of 25 files produced by the CLI itself — 20 claiming PDF/A (renders across all four levels plus attachments, headers/footers, outline, watermark, a table-variant render with embedded fonts, an AcroForm, Gray and CMYK output intents, an incremental PAdES signature, a metadata update, a JPEG 2000 image under PDF/A-2b and four negative canaries — unembedded fonts, an ICC v4 profile under PDF/A-1) and 6 claiming PDF/X-4 (a CMYK and a Gray print render, a signed print render, a spot-colour render, a negative canary broken by an annotation outside the page boxes, and the PDF/A-2b + PDF/X-4 dual claim, counted in both groups) — is checked against the profile each file claims in XMP. It includes negative canaries that the validators must reject — so a validator that accepts everything fails the run instead of turning it green. The PDF/A recipe is --tagged pdfa<level> --font latin --lang latin (fonts must be embedded per ISO 19005; the sRGB OutputIntent is emitted automatically by the engine); the PDF/X-4 recipe is --pdfx pdfx4 --output-intent-icc <cmyk.icc> --font latin --lang latin --trapped false --strict. Run the same gates locally with npm run corpus:pdfa && npm run validate:pdfx && npm run validate:pdfa — without veraPDF installed the PDF/A step prints install hints and exits 0 as a skip, not a pass (npx tsx scripts/gate.ts --publish --require-all turns the skip into a failure). See CONTRIBUTING.md for details. Not a certification — validation evidence against a specific veraPDF version (1.30.2) and the engine's own PDF/X-4 validator (not a certified preflight tool).

Installation

npm install --global pdfnative-cli

Or run without installing:

npx pdfnative-cli render --input doc.json --output report.pdf

Requirements: Node.js ≥ 22 (Node 20 reached end-of-life 2026-04-30). The gate runs on Node 22 and 24 on Linux and on Node 22 on Windows and macOS, all as required checks; Node 26 is advisory until the first minor after a green history (SECURITY.md §Supported runtimes). Bun and Deno are not tested.

Documentation

  • 📘 Quick Start (below) — Get rendering in 5 minutes
  • 🏛️ KNOWLEDGE_BASE.md — Full CLI reference, architecture, integration patterns
  • 🤖 AGENT_CONTRACT.md — The process contract for autonomous agents (envelopes, error codes, token economy)
  • 📚 samples/README.md — runnable samples organized by feature
  • 🔧 pdfnative library — Underlying PDF engine docs
  • ❓ FAQ — Common questions & troubleshooting

Quick Start

Render a PDF from JSON

# From a file
pdfnative render --input document.json --output report.pdf

# From stdin
cat document.json | pdfnative render --output report.pdf

# Streaming (large documents)
pdfnative render --input big-doc.json --output report.pdf --stream

# True constant-memory streaming (lowest peak memory; byte-identical)
pdfnative render --input big-doc.json --output report.pdf --stream-true

# PDF/A conformance (embed the bundled Latin font — ISO 19005 requires embedded
# fonts; this recipe is what the blocking veraPDF CI gate validates)
pdfnative render --input document.json --output archived.pdf \
  --tagged pdfa2b --font latin --lang latin

document.json is a DocumentParams object (pdfnative schema render prints the full shape, typography and PDF/X included):

{
  "title": "Monthly Report",
  "blocks": [
    { "type": "heading", "text": "Monthly Report", "level": 1 },
    { "type": "paragraph", "text": "Summary for April 2026." },
    { "type": "list", "style": "bullet", "items": ["Revenue: +18%", "NPS: 72"] }
  ],
  "footerText": "Confidential",
  "metadata": { "author": "Finance Team", "subject": "April 2026 Report" }
}

Typography, PDF/X-4 & reproducible output (v1.5.0)

# Typography: split long paragraphs across pages, keep headings with their body,
# kern the bundled Latin font and turn on old-style numerals
pdfnative render --input report.json --output report.pdf --font latin --lang latin \
  --split-paragraphs --keep-headings-with-next --kerning --font-features onum,smcp

# The full typography surface lives in layout.typography (JSON or --layout file):
# widows/orphans, justify, optical margins, soft hyphens, French spacing + unit binding
pdfnative render --input report.json --output report.pdf --layout typography.json

# PDF/X-4 for print: CMYK output intent, trapped flag, strict diagnostics — then re-check
pdfnative render --input brochure.json --output brochure.pdf \
  --pdfx pdfx4 --output-intent-icc press.icc --font latin --lang latin --trapped false --strict
pdfnative inspect --input brochure.pdf --check pdfx          # exit 0 when the claim holds

# Reproducible bytes: pin the creation instant (or export SOURCE_DATE_EPOCH=1767225600)
pdfnative render --input doc.json --output a.pdf --creation-date 2026-01-01T00:00:00Z
TZ=Asia/Tokyo pdfnative render --input doc.json --output b.pdf --creation-date 2026-01-01T00:00:00Z
sha256sum a.pdf b.pdf                                        # identical

Right-to-left, Indic scripts, spot colours & the dual claim (v1.6.0)

# Arabic and Hebrew paragraphs under the full Unicode Bidirectional Algorithm — right-aligned,
# justified in visual order — then read back in logical order (no --tagged needed)
pdfnative render --input letter.json --output letter.pdf --direction auto \
  --font ar --font he --font latin --lang ar,he,latin
pdfnative extract-text --input letter.pdf --format ndjson

# Five more Indic scripts: Gujarati, Punjabi (Gurmukhi), Kannada, Malayalam, Odia
pdfnative render --input notice.json --output notice.pdf --font gu --font pa --font kn --font ml --font or --lang gu,pa,kn,ml,or

# Spot colours under PDF/X-4 (inks declared in layout.print.spotColours, used as { "spot": "…" })
pdfnative render --input brochure.json --output brochure.pdf \
  --pdfx pdfx4 --output-intent-icc press.icc --font latin --lang latin --trapped false --strict

# PDF/A-2b and PDF/X-4 claimed in ONE file, then both claims re-checked
pdfnative render --input brochure.json --output archive.pdf \
  --tagged pdfa2b --pdfx pdfx4 --output-intent-icc press.icc --font latin --lang latin --trapped false
pdfnative inspect --input archive.pdf --check pdfa,pdfx

# An over-cap content stream is a refusal, not an empty page (default: limit)
pdfnative extract-text --input untrusted.pdf --max-inflate-size 8000000 --on-stream-error throw

# PAdES B-LT with no network at all: the DSS from material you already hold
pdfnative ltv embed --input signed.pdf --output lt.pdf --cert chain.pem --ocsp signer.ocsp.der --crl ca.crl --match chain

Sign a PDF

# Keys from environment variables (recommended for CI/CD)
export PDFNATIVE_SIGN_KEY="$(cat private.pem)"
export PDFNATIVE_SIGN_CERT="$(cat cert.pem)"
pdfnative sign --input document.pdf --output signed.pdf

# Keys from files
pdfnative sign --input document.pdf --output signed.pdf \
  --key private.pem --cert cert.pem

Inspect a PDF

# JSON output (default)
pdfnative inspect --input report.pdf

# Human-readable
pdfnative inspect --input report.pdf --format text

# PDF/UA (ISO 14289-1) structural validation report
pdfnative inspect --input report.pdf --pdfua

# CI accessibility gate (exit 1 if not PDF/UA-structurally-valid)
pdfnative inspect --input report.pdf --check pdfua

# PDF/X-4 validation report, ISO 8601 dates (v1.5.0)
pdfnative inspect --input brochure.pdf --pdfx --iso-dates --fields pdfx.valid,metadata.creationDate

# From stdin
cat report.pdf | pdfnative inspect

Example output:

{
  "version": "1.7",
  "pageCount": 3,
  "encrypted": false,
  "pdfaConformance": "2b",
  "signatures": 1,
  "metadata": {
    "title": "Monthly Report",
    "author": "Nizoka",
    "creationDate": "D:20260427120000+00'00'"
  }
}

Merge, split & extract pages (v1.2.0)

# Concatenate several PDFs (sources as positional args or repeated --input)
pdfnative merge a.pdf b.pdf c.pdf --output combined.pdf

# Split one PDF into one file per page
pdfnative split --input report.pdf --output-dir pages/ --prefix page

# Split into ranges — one output per comma-separated segment
pdfnative split --input report.pdf --output-dir out/ --pages "1-2,3-4"

# Extract selected pages (1-based; order preserved, repeats allowed)
pdfnative extract --input report.pdf --output cover.pdf --pages "4,1-2"

Extract text, fill forms & encrypt (v1.3.0)

# Extract reading-order text as NDJSON (one object per page — ideal for RAG/agents)
pdfnative extract-text --input report.pdf --format ndjson > pages.ndjson

# Export a form's current values, edit, then fill (read → edit → fill)
pdfnative fill --input form.pdf --export > values.json
pdfnative fill --input form.pdf --data values.json --output filled.pdf
pdfnative fill --input filled.pdf --flatten --output flat.pdf

# Preflight the environment (agents: gate `encrypt` on this)
pdfnative doctor --format json

# Encrypt with AES-256, confirm the scheme, then decrypt
pdfnative encrypt --input report.pdf --output secure.pdf \
  --owner-password "$OWNER" --user-password "$USER" --algorithm aes-256
pdfnative inspect --input secure.pdf --encryption --password "$USER"
pdfnative decrypt --input secure.pdf --output plain.pdf --password "$USER"

# Render native vector charts from a document with a `chart` block
pdfnative render --input dashboard.json --output dashboard.pdf

Long-term signatures, compare & metadata (v1.4.0)

The canonical PAdES ladder — each rung is one command, network access is always an explicit opt-in (--timestamp / --online / --url), SSRF-guarded, no redirects:

# B-T  — sign with a PAdES profile and an RFC 3161 trusted timestamp
pdfnative sign --input doc.pdf --output signed.pdf \
  --profile pades --timestamp https://tsa.example.com/tsr

# B-LT — archive the OCSP/CRL validation data into /DSS + /VRI
pdfnative ltv add --input signed.pdf --output lt.pdf --online

# B-LTA — append a document timestamp covering every byte (repeat to renew)
pdfnative doc-timestamp --input lt.pdf --output lta.pdf --url https://tsa.example.com/tsr
pdfnative ltv add --input lta.pdf --output archived.pdf --online

# Air-gapped variant: collect on a connected machine, embed fully offline
pdfnative ltv collect --input signed.pdf --output ltv.json --online
pdfnative ltv embed   --input signed.pdf --data ltv.json --output lt.pdf
# Diff two PDFs by text + structure (exit 0 identical, exit 1 on differences)
pdfnative compare original.pdf revised.pdf --mode both --format json

# Update /Info + XMP metadata without breaking existing signatures (incremental save)
pdfnative metadata --input signed.pdf --output retitled.pdf \
  --title "Q3 Report (final)" --author "Finance Team"

# Verify the whole ladder — /DocTimeStamp revisions are validated as RFC 3161 tokens
pdfnative verify --input archived.pdf --strict
pdfnative inspect --input archived.pdf --signatures --check "signatures>=1"

Annotate a PDF (v1.2.0)

# Attach markup annotations from a JSON spec (incremental save — signatures stay intact)
pdfnative annotate --input report.pdf --output annotated.pdf \
  --annotations notes.json

# List them back out
pdfnative inspect --input annotated.pdf --annotations --format text

notes.json is a JSON array (or { "annotations": [...] }), each entry a markup or link annotation plus a 1-based page:

{
  "annotations": [
    { "page": 1, "type": "highlight", "rect": [72, 700, 320, 715], "color": "#FFD400", "contents": "Check this figure." },
    { "page": 1, "type": "text", "rect": [330, 700, 350, 720], "icon": "Comment", "contents": "Needs a citation." },
    { "page": 1, "type": "link", "rect": [72, 640, 300, 655], "url": "https://pdfnative.dev" }
  ]
}

Render bookmarks & introspect layout (v1.2.0)

# Add a bookmark tree derived from the document's headings
pdfnative render --input doc.json --output book.pdf --outline auto

# …or supply an explicit OutlineItem[] tree
pdfnative render --input doc.json --output book.pdf --outline outline.json

# Register the bundled math font so ∑ ∫ √ π render as real glyphs
pdfnative render --input math.json --output math.pdf --font latin --font math

# Emit a LayoutInspection JSON report instead of a PDF
pdfnative render --input doc.json --output layout.json --inspect-layout

# Render a PDF with debug guides overlaid
pdfnative render --input doc.json --output debug.pdf --debug-layout margins,content,cells

AI-governance / Human-in-the-Loop (v1.2.0)

Agents act as draftsmen: they may draft an issue/PR locally, but a human must review and submit it. Nothing here touches the network.

# Print the human/agent protocol and the machine-readable policy
pdfnative govern rules
pdfnative govern policy --pretty

# Gate a locally-authored draft (exit 1 / E_POLICY on a violation)
pdfnative govern verify-issue ./draft.md

Examples

Ready-to-run examples are in samples/, organized by feature category:

Category Examples Description
render/document/ 6 files Minimal, report, all-blocks reference, invoice, technical spec, --max-blocks guard
render/table/ 2 files Project status, financial summary
render/table-smart/, table-variant/ 1 + 1 files Smart-table flags; --variant table (PdfParams, fonts embedded since v1.5.0)
render/barcode/ 3 files QR code, Code 128 shipping label, EAN-13 product
render/form/ 2 files Contact form, survey
render/toc/ 2 files Document with auto-generated table of contents, (v1.6.0) a 160-entry TOC that paginates
render/link/ 1 file Resource directory with hyperlinks
render/watermark/ 3 files Draft watermark, confidential watermark, CLI-flag styling
render/layout/ 3 files US Letter, A5 portrait, A4 landscape
render/headers-footers/, template/ 1 + 2 files Page-number placeholders; --template deep-merge
render/pdfa/, attachments/ 5 + 1 files PDF/A-1b, PDF/A-2b, PDF/A-2u, PDF/A-3b archival conformance; (v1.5.0) an AcroForm under PDF/A-2b; PDF/A-3b XML attachment
render/encryption/ 2 files AES-128 / AES-256 protected renders
render/outline/ 2 files PDF bookmarks — --outline auto + explicit tree
render/math/ 1 file Math/technical symbols via --font math
render/inspect-layout/ scripts --inspect-layout report + --debug-layout guides
render/font/ 5 files Bundled font presets: Latin, the 1.3.0 scripts, emoji, (v1.5.0) the five 1.8.0 scripts, --font-file, colour-emoji skin tones / ZWJ sequences / flags
render/multilang/ 13 files Thai, Japanese, multilingual drivers, (v1.5.0) Lao, Tai Tham / New Tai Lue / Tai Le / Cham, Hausa / Yoruba / Igbo / Swahili, and four families (European & Caucasian, right-to-left, Indic, Chinese & Korean) — every one of the 32 script codes is rendered by the corpus
render/chart/ 6 files Native vector charts — bar/line/pie/donut plus (v1.4.0) stacked bars, area/scatter, dual axes, log & time axes
render/print/ 9 files Print production — bleed/trim boxes + printer's marks, viewer preferences, (v1.5.0) CMYK colours, colour bars, PDF/X-4 with a synthetic CMYK profile
render/images/ 1 file (v1.6.0) JPEG 2000 image block passed through as /JPXDecode (pdfnative ≥ 1.9.0)
render/typography/ 7 files (v1.5.0) Paragraph breaking, justify + optical margins + soft hyphens, French (fr, fr-CA) and custom punctuation spacing, unit binding, kerning + OpenType features, per-block keepWithNext / splittable
render/base14/ 1 file (v1.5.0) layout.typography.metrics: "exact" on the base-14 path — rendered with no --font flag, the only place the option acts
render/reproducible/ 2 files (v1.5.0) --creation-date / SOURCE_DATE_EPOCH — the double-render script proves byte identity across timezones
render/watch/ scripts --watch re-render on change
merge/ scripts Concatenate PDFs (page-tree)
split/ scripts Split one PDF per-page or per-range
extract/ scripts Pull selected pages into a new PDF
extract-text/ scripts Reading-order text (text / json / ndjson)
fill/ scripts Fill, flatten & export AcroForms
encrypt/ scripts Encrypt / decrypt round-trip (AES-256, --stream)
doctor/ 2 pairs Environment / capability preflight; (v1.5.0) fonts, Unicode and conformance targets
metadata/ scripts Update /Info + XMP metadata (incremental save)
compare/ scripts Diff two PDFs by text + structure
annotate/ 2 pairs Attach markup annotations (incremental save); (v1.5.0) link annotations
govern/ scripts AI-governance / HITL: rules, policy, verify-issue
sign/ 11 pairs Digital signature incl. timestamp (B-T), LTV ladder, multi-signatures, native vs pure-JS crypto, (v1.5.0) --timestamp-timeout (Bash + PowerShell)
verify/ 7 pairs Signature verification: trust roots, revocation, strict mode, (v1.5.0) weak-digest note
inspect/ 10 pairs JSON & text inspection incl. --annotations, --signatures, (v1.5.0) --check pdfx and --iso-dates (Bash + PowerShell)
batch/ 3 pairs Directory render and --manifest pipelines
agent/ 5 pairs --json + --dry-run, schema, the error envelope, token economy, (v1.5.0) global flags before the command
completion/, config/ 1 + 1 pairs Shell completions; .pdfnativerc.json defaults
streaming/ 2 pairs Streaming render (single-pass, page-by-page, true constant-memory)

Render every sample at once (the same run the CI baseline uses):

npm run build && npm run test:generate     # → test-output/samples/ (99 sample PDFs, byte-stable)
npx tsx scripts/verify-samples.ts          # compare with tests/regression/baselines/samples.sha256.json

PDFNATIVE_CLI=$(which pdfnative) npm run test:generate drives a globally installed binary instead.

See samples/README.md for full descriptions, block type reference, and integration patterns (GitHub Actions, Docker, TypeScript).


Command Reference

The 21 commands are grouped by purpose (the global pdfnative --help shows the same grouping):

Group Commands
Create & edit render, fill, annotate, metadata
Page tree merge, split, extract
Security sign, verify, ltv, doc-timestamp, encrypt, decrypt
Read & extract inspect, extract-text, compare
Automation & meta batch, doctor, schema, completion, govern

pdfnative render

Flag Default Description
--input <file> stdin Path to a JSON file (DocumentParams or PdfParams if --variant table)
--output <file> stdout Output PDF path
--stream false Single-pass streaming output (AsyncGenerator); no TOC, no {pages}
--stream-page-by-page false Stream at PDF object boundaries (TOC- and {pages}-compatible)
--stream-true false True constant-memory streaming; parts freed as emitted; byte-identical; no TOC, no {pages}
--chunk-size <bytes> 65536 Chunk size for --stream / --stream-true (not --stream-page-by-page)
--variant <kind> document document (default) or table (selects buildPDFBytes)
--layout <file.json> — Load a Partial<PdfLayoutOptions> (CLI flags override)
--page-size <size> from layout file or pdfnative default Named (a4, letter, legal, a3, tabloid, a5) or WxH in points
--margin <N> or --margin <t,r,b,l> from layout / default Page margins in points
--compress false Enable FlateDecode compression
--max-blocks <n> 100000 Maximum document blocks before pdfnative aborts (large-report guard)
--tagged <level> none PDF/A: none, pdfa1b, pdfa2b, pdfa2u, pdfa3b; pdfa2b / pdfa2u may be combined with --pdfx pdfx4 since v1.6.0 (pdfnative ≥ 1.9.0), pdfa1b / pdfa3b may not (E_INPUT); the --json envelope carries tagged
--pdfx <target> none (v1.5.0) PDF/X: pdfx4 (ISO 15930-7). Requires a prtr ICC output intent (--output-intent-icc or layout.outputIntent), refuses encryption (exit 2); the --json envelope carries pdfx
--output-intent-icc <file.icc> — (v1.5.0) ICC profile for layout.outputIntent.iccProfile (16 MiB cap, acsp signature checked; prtr CMYK required for PDF/X)
--output-intent-id <s> ICC file basename (v1.5.0) outputIntent.outputConditionIdentifier
--trapped true|false|unknown — (v1.5.0) /Trapped + pdf:Trapped (PDF/X-4 requires true or false)
--strict false Escalate the 13 diagnostic codes — PDF/A (PDFA_NO_FONT_ENTRIES, PDFA_UNEMBEDDED_FORM_FONT, PDFA_DEVICE_CMYK_IMAGE, PDFA_DEVICE_CMYK_CONTENT, PDFA_ICC_PROFILE_VERSION, PDFA_GLYPH_WITHOUT_UNICODE, PDFA_JPX_IMAGE, PDFA1_JPX_IMAGE), PDF/X (PDFX_NO_FONT_ENTRIES, PDFX_DEVICE_CMYK, PDFX_ANNOTATIONS), TYPOGRAPHY_FEATURE_INEFFECTIVE and TEXT_GLYPH_NOT_COVERED — into an error (exit 1, E_CHECK_FAILED) before any output byte; without it they are stderr warnings + a diagnostics[] array in the --json envelope
--split-paragraphs false (v1.5.0) layout.typography.splitParagraphs — long paragraphs may break across pages (widow/orphan rules apply)
--keep-headings-with-next false (v1.5.0) layout.typography.keepHeadingsWithNext — a heading never ends a page alone
--kerning false (v1.5.0) layout.typography.kerning — GPOS pair kerning for embedded fonts
--font-features <tag,…> — (v1.5.0) layout.typography.fontFeatures — OpenType feature tags (onum, smcp, tnum, …; ligature features are not applied by the engine; four alphanumerics each)
--direction auto|ltr|rtl auto (engine) (v1.6.0) layout.direction — the base direction of every paragraph (full UAX #9, pdfnative ≥ 1.9.0): auto detects it from the first strong character, aligns right-to-left paragraphs to the right and justifies them in visual order; a paragraph block's own direction wins; the envelope carries direction
--actual-text [true|false] true (engine) (v1.6.0) layout.actualText — untagged shaped and right-to-left runs carry their logical text as /Span /ActualText, so extract-text returns the source string; false restores the bare 1.8.0 runs (the envelope then carries actualText: false)
--conformance <1b|2b|3b> — Deprecated — use --tagged pdfa<level>
--watermark-text <s> / --watermark-image <path> — Text or image watermark (PNG, JPEG and — v1.6.0 — JPEG 2000 .jp2 / .j2k; the engine sniffs the format)
--watermark-opacity <0-1> / --watermark-angle <deg> / --watermark-color <#hex> / --watermark-font-size <pt> / --watermark-position background|foreground — Watermark styling
--watermark-position background|foreground background Render order
--header-{left,center,right} <tpl> — Header template; placeholders {page}, {pages}, {date}, {title}
--footer-{left,center,right} <tpl> — Footer template; same placeholders
--encrypt [aes-128|aes-256] — Enable encryption (bare = aes-128). Same vocabulary as merge/split/extract
--owner-password <s> $PDFNATIVE_ENCRYPT_OWNER_PASS Owner password (required to encrypt)
--user-password <s> $PDFNATIVE_ENCRYPT_USER_PASS Optional user (open) password
--permissions <list> all denied Comma list: print,copy,modify,extract
(legacy aliases) — --encrypt-algorithm, --encrypt-owner-pass, --encrypt-user-pass, --encrypt-permissions still work
--attachment <path>[:mime[:rel[:desc]]] (repeatable) — PDF/A-3 file attachment
--lang <code,code> — Activate registered font loaders for non-Latin scripts (th, ja, ar, te, si, km, lo, …); Latin is built-in. Aliases ha, yo, ig, sw resolve to latin
--font <name> (repeatable) — Register a bundled font shortcut. Allow-list: latin, emoji, color-emoji, math, and the 32 script codes ar hy bn ru hi am ka el he ja km ko my pl zh si ta te th bo tr vi + (v1.5.0) lo nod khb tdd cjm + (v1.6.0) gu pa kn ml or; ha yo ig sw alias latin. The name doubles as the --lang code. --variant table embeds them too
--font-file <path.ttf>[:name] (repeatable) — (v1.5.0) Register a custom TrueType/OpenType font from disk (name defaults to the file stem, [a-z0-9-]; added to --lang). 32 MiB cap, magic bytes (00 01 00 00 / true / OTTO — collections and WOFF refused), parsed and validated by pdfnative before use; a name colliding with the allow-list is a usage error. Never loadable from a JSON payload
--outline auto|<file.json> — Add a PDF bookmark tree: auto derives it from headings; a path loads an explicit OutlineItem[] tree
--inspect-layout false Emit a LayoutInspection JSON report instead of a PDF (document variant only)
--debug-layout [margins,content,cells] — Overlay layout debug guides on the rendered PDF (bare flag = all)

Document & layout JSON (v1.6.0, pdfnative ≥ 1.9.0) — layout.direction: "auto" | "ltr" | "rtl" and a paragraph-level direction (UAX #9); layout.actualText (default on); layout.print.spotColours: [{ name, alternate }] declares the inks a document may use, and every colour field then also accepts { "spot": "<name>" }, { "spot": "<name>", "tint": 0.5 } or { "inks": { "<name>": 0.8, "<other>": 0.2 } } (/Separation with the CMYK alternate, /DeviceN with /Colorants for a mix; an undeclared ink is E_INPUT); chart series gain colors (one per value) and an optional label; image blocks accept JPEG 2000 bytes (src: "…jp2" or dataBase64) and are embedded as /JPXDecode without decoding (PDF 1.5; a JPX that carries its own colour space raises PDFA_JPX_IMAGE under PDF/A-2/3 and PDFA1_JPX_IMAGE under PDF/A-1); layout.pdfx: "pdfx4" may now sit next to tagged: "pdfa2b" | "pdfa2u" (the dual claim: two OutputIntents, both identifications in the XMP; pdfa1b / pdfa3b stay refused). A long table of contents paginates. Validate with pdfnative schema render.

Document & layout JSON (v1.5.0, pdfnative ≥ 1.8.0) — layout.typography (widows, orphans, splitParagraphs, keepHeadingsWithNext: true | { minLines }, opticalMargins, punctuationSpacing: "fr" | "fr-CA" | [{ char, side, space }], unitBinding, bindShortWords, hyphenationLanguage, kerning, fontFeatures: [tag, …], metrics: "exact" — base-14 path only, inert once --font registers a font; soft hyphens U+00AD are honoured unconditionally), paragraph align: "justify", block-level keepWithNext / splittable; every colour accepts CMYK ("c m y k" 0–1 or [c,m,y,k] percent) beside hex / RGB; layout.print.marks.colourBars: true | { tints, size }; layout.pdfx: "pdfx4" with a prtr CMYK layout.outputIntent (iccProfile as a number array in JSON); layout.creationDate (ISO string) — flags win over the file, nested typography / outputIntent objects merge one level deep. --strict also covers the TYPOGRAPHY_FEATURE_INEFFECTIVE diagnostic (a requested feature the font cannot honour).

Document & layout JSON (v1.4.0, pdfnative ≥ 1.7.0) — no new flags, richer JSON: image blocks accept src (a path resolved relative to the --input JSON's directory) as an alternative to dataBase64; chart blocks grow to 9 types (bar, barH, stackedBar, stackedBarH, line, area, scatter, pie, donut) with series[].xValues, series[].yAxis: "right" + axis2, axis.scale: "log", xAxis: category|linear|time, dataLabels, labelStride, and labelRotation; the --layout file gains print (bleed, trimBox, bleedBox, artBox, cropBox, printer marks, userUnit 1–75000), outputIntent (ICC RGB) and viewerPreferences (duplex, pickTrayByPDFSize, printPageRange, numCopies); params.metadata accepts author/subject/keywords/trapped. Validate with pdfnative schema render.

See samples/render/ for a working example of every category.

pdfnative sign

Flag Default Description
--input <file> stdin Path to the input PDF
--output <file> stdout Output signed PDF path
--key <file> $PDFNATIVE_SIGN_KEY Path to PEM private key (env var takes precedence)
--cert <file> $PDFNATIVE_SIGN_CERT Path to PEM certificate (env var takes precedence)
--cert-chain <file> (repeatable) $PDFNATIVE_SIGN_CHAIN Intermediate CA PEMs
--algorithm rsa-sha256|ecdsa-sha256 rsa-sha256 Signature algorithm (RSA or P-256 ECDSA)
--digest sha256|sha384|sha512 sha256 CMS digest algorithm (RSA only; ecdsa-sha256 is sha256-only)
--profile pkcs7|pades pkcs7 pades = ETSI.CAdES.detached (PAdES B-B: ESS signing-certificate-v2, omits signing-time)
--reason <s> — Reason for signing (PDF metadata)
--name <s> — Signer name (PDF metadata)
--location <s> — Signing location (PDF metadata)
--contact <s> — Signer contact (PDF metadata)
--signing-time <ISO 8601> now Explicit signing timestamp
--timestamp <tsa-url> — Embed a verified RFC 3161 timestamp token at signing time (PAdES B-T with --profile pades). Opt-in network, SSRF-guarded. TSA failure → E_NETWORK; malformed response → E_PARSE; never a silent fallback
--timestamp-digest sha256|sha384|sha512 sha256 Digest for the TSA message imprint
--timestamp-nonce <hex> random 8 bytes Explicit TSA request nonce
--timestamp-timeout <ms> 10000 (v1.5.0) Timeout for the TSA round-trip (positive integer; usage error without --timestamp); reported as timestamp.timeoutMs
--allow-multiple false Allow signing an already-signed PDF (appends a signature field); default stays idempotent single-signature (1.x behaviour)
--field-name <name> auto Signature form-field name
--signature-rect "x1,y1,x2,y2" invisible Visible signature widget rectangle (PDF points)
--signature-page <n> 1 1-based page for the signature widget
--placeholder-bytes <n> auto Explicit /Contents placeholder size (overrides the estimate)
--pure-crypto false Force pdfnative's pure-JS RSA/ECDSA math instead of the default native node:crypto (constant-time) provider

Without --timestamp the command performs no network I/O, and --dry-run never touches the network even when --timestamp is present. Under --json, a timestamped signature adds timestamp: { url, digest, timeoutMs? } to the success envelope. --signing-time is a deliberate legal instant and is not pinned by the global --creation-date.

pdfnative inspect

Flag Default Description
--input <file> stdin Path to the PDF to inspect
--output <file> stdout Output report path
--format json|text json Output format
--verbose false Add trailer keys, catalog keys, object count, XMP
--pages false Add per-page metadata array (width/height/rotation + cropBox/trimBox/bleedBox/artBox/userUnit when present)
--annotations false List markup + link annotations per page (page labels are reported automatically when present)
--form-fields false List AcroForm fields (name, type, value, required/read-only, options)
--encryption false Report the encryption scheme (algorithm, revision, authenticatedAs), or null
--signatures false Signature-field inventory: fieldName, subFilter, byteRange, isDocTimestamp, isPlaceholder, sigObjNum, contentsLength — never the signature bytes
--password <s> — Password for an encrypted PDF (env: PDFNATIVE_PASSWORD)
--pdfua false Add a PDF/UA (ISO 14289-1) structural validation report (valid + errors + warnings)
--pdfx false (v1.5.0) Add a PDF/X-4 validation report (pdfx: { valid, target, errors, warnings }, pdfnative's validatePdfX); pdfxConformance (XMP GTS_PDFXVersion) is always reported
--iso-dates false (v1.5.0) Normalise metadata.creationDate / modDate from D:YYYYMMDD… to ISO 8601 (unparsable strings are left as is)
--check pdfa|signed|encrypted|pdfua|pdfx|"signatures>=N" (repeatable, or a comma list — v1.6.0) — CI-friendly assertion; AND semantics; sets exit code (0 = pass, 1 = fail). signatures>=N counts real signatures (placeholders and /DocTimeStamp fields excluded); pdfx requires a PDF/X-4 claim that validates

v1.4.0 also fixes the top-level signatures and formFields counters, which previously always reported 0, and reports metadata.trapped when present. --summary returns { pages, encrypted, signatures, pdfa, pdfx }.

pdfnative verify

Flag Default Description
--input <file> stdin Path to the (possibly signed) PDF
--format json|text json Output format
--strict false Exit 1 on any failure or zero signatures (CI-friendly)
--trust <root.pem> (repeatable) self-signed only Trusted root certificates (PEM)
--revocation offline|online|disabled offline Revocation source: embedded /DSS only, also fetch online (SSRF-guarded), or skip
--revocation-policy soft-fail|strict soft-fail strict fails the signature on any non-good status; soft-fail only fails on explicit revoked

Scope: byte-range integrity (SHA-256), full CMS signature value (RSA-PKCS#1 v1.5 — SHA-256/384/512, reported as rsa-sha256/rsa-sha384/rsa-sha512 — plus ECDSA-SHA256 over P-256), certificate chain + trust, RFC 3161 timestamp validation (PAdES-T), and OCSP (RFC 6960) + CRL (RFC 5280) revocation — embedded from the PDF /DSS offline by default, with opt-in online fetching through an SSRF-guarded HTTP client. v1.4.0 additions: each signature also reports its fieldName and isDocTimestamp, and /DocTimeStamp revisions (PAdES B-LTA) are validated as RFC 3161 timestamp tokens. v1.5.0: each timestamp reports its messageImprint algorithm as timestampDigest; a SHA-1 imprint adds the note weak digest: RFC 3161 messageImprint uses SHA-1 (refused under --strict) and, under --strict, fails the timestamp (E_VERIFY_FAILED). Sign-side LTV lives in sign --timestamp, ltv and doc-timestamp.

pdfnative ltv

PAdES B-LT: archive the certificates, OCSP responses and CRLs needed to validate the document's signatures long after certificates expire, into the PDF's /DSS + /VRI dictionaries (incremental save — existing signatures stay valid).

pdfnative ltv collect --input signed.pdf --online [--output ltv.json]  # fetch → replayable JSON
pdfnative ltv embed   --input signed.pdf --data ltv.json [--output out.pdf]  # 100 % offline
pdfnative ltv embed   --input signed.pdf --cert chain.pem --ocsp r.der [--crl c.der] \
                      [--match all|chain] [--output out.pdf]           # (v1.6.0) from material you hold
pdfnative ltv add     --input signed.pdf --online [--output out.pdf]   # collect + embed

collect requires --online (explicit network opt-in, SSRF-guarded, no redirects) and emits a replayable JSON file (schema subject: ltv-data). embed never performs network I/O — the collect/embed split supports air-gapped pipelines: collect on a connected machine, embed offline. Since v1.6.0 (pdfnative ≥ 1.9.0) embed also takes the material directly — certificates (PEM bundles or DER), DER OCSP responses and DER CRLs you obtained by any means — and composes the /DSS + /VRI through the engine's composeLtvData: --match all (default) references every item from every signature's VRI (the Adobe-tolerant superset), --match chain references only what validates each signature (unmatched material still ships in /DSS). --data and the material flags are exclusive; a composeLtvData refusal (no signature, an OCSP response matching no certificate) is E_INPUT. add does collect + embed in one pass.

Flag Default Description
--input <file> stdin Signed source PDF
--output <file> stdout Output: JSON (collect) or PDF (embed / add)
--online — (required for collect / add) Explicit opt-in for network fetches (SSRF-guarded, no redirects)
--prefer ocsp|crl ocsp Preferred revocation source
--extra-cert <pem> (repeatable) — Extra chain certificates (PEM)
--data <file> — (embed: this or the material below) Previously collected ltv-data JSON
--cert <file> (repeatable) — (v1.6.0, embed) A certificate — PEM bundle or DER: signer chains, TSA and OCSP responder certificates (16 MiB cap each)
--ocsp <file> (repeatable) — (v1.6.0, embed) A DER OCSPResponse
--crl <file> (repeatable) — (v1.6.0, embed) A DER CRL
--match all|chain all (v1.6.0, with material only) all: every item referenced from every signature's VRI; chain: each VRI references only what validates that signature; the envelope carries match
--timeout <ms> 10000 Network timeout
--dry-run false Validate inputs; no output, no network

The canonical PAdES ladder:

sign --timestamp <tsa> --profile pades   →  B-T
ltv add --online                         →  B-LT   (air-gapped: ltv embed --cert … --ocsp … --crl …)
doc-timestamp --url <tsa>                →  B-LTA
ltv add --online                         →  LTV for the doc-timestamp itself

pdfnative doc-timestamp

PAdES B-LTA: append a /DocTimeStamp signature field (SubFilter /ETSI.RFC3161, ISO 32000-2 §12.8.5) covering every byte of the document, as an incremental revision — earlier revisions stay byte-identical. Repeat periodically to renew LTA protection.

Flag Default Description
--input <file> stdin Signed source PDF
--output <file> stdout Output PDF
--url <tsa-url> — (required) RFC 3161 TSA URL (explicit network opt-in; SSRF-guarded, no redirects)
--digest sha256|sha384|sha512 sha256 Timestamp message-imprint digest
--field-name <name> DocTimeStamp1 Timestamp field name (auto-suffixed on collision)
--placeholder-bytes <n> 12288 /Contents placeholder size
--nonce <hex> random Explicit TSA request nonce
--timeout <ms> 10000 Network timeout
--dry-run false Validate inputs; no output, no network

TSA failures map to E_NETWORK; a malformed TSA response maps to E_PARSE. verify validates /DocTimeStamp revisions as RFC 3161 tokens (isDocTimestamp: true).

pdfnative batch

Two modes: directory mode (render every *.json in a directory, in parallel) and manifest mode (v1.4.0 — run a declarative multi-command pipeline). The modes are mutually exclusive.

Directory mode:

Flag Default Description
--input-dir <dir> required Directory of *.json document definitions
--output-dir <dir> required Output directory for the rendered *.pdf (created if absent)
--concurrency <n> 4 Maximum parallel renders
--fail-fast false Stop at the first failure (default: render all, then report)
--format json|text text Summary format

All other flags are forwarded to each render. Exit code 1 if any file fails.

Manifest mode (v1.4.0):

Flag Default Description
--manifest <tasks.json> — Declarative pipeline (schema subject: batch-manifest): { "version": 1, "tasks": [ { "id", "command", "flags" } ] }
--allow-network false Required for any network flag inside the manifest (--timestamp, --url, --online, --revocation online) — an untrusted manifest can never trigger network I/O on its own
--continue-on-error false Keep running after a failure; tasks depending (via @) on a failed task are skipped
--format json|text text Summary format
--dry-run false Validate the manifest / inputs without executing

Tasks run sequentially, fail-fast by default. A flag value "@<id>" references the output of an earlier task; relative paths resolve against the manifest's directory. The whitelist holds 14 of the 21 commands — render, sign, verify, inspect, merge, split, extract, extract-text, fill, encrypt, decrypt, annotate, metadata, doc-timestamp (ltv and compare need positional arguments and are not yet manifest-callable) — never govern, schema, completion, doctor, or batch itself. Under --json the summary envelope adds mode: "manifest", skipped, and a per-task tasks[] array. Exit code 1 if any task fails.

A manifest has the filesystem access of the user who invokes batch — the same trust level as command-line flags; only network access is additionally gated behind --allow-network. The manifest file is size-capped (50 MB) and bounded to 1 000 tasks, and path values undergo the same anti-traversal check as direct CLI flags.

pdfnative merge

Concatenate several PDFs into one (pdfnative ≥ 1.5.0 page-tree API).

Flag Default Description
positional paths — Source PDFs, in order (2–50). May be combined with --input
--input <file> (repeatable) — Additional source PDF (appended after positionals)
--output <file> stdout Output combined PDF
--drop-annotations false Strip annotations from the merged output
--max-output-size <bytes> 256 MiB Fail if the merged PDF would exceed this size (0/none = unlimited)
--password <s> — Password for encrypted source PDFs (env: PDFNATIVE_PASSWORD)
--encrypt [aes-128|aes-256] — Re-encrypt the output (bare = aes-128); needs --owner-password
--owner-password / --user-password — Passwords for --encrypt (env: PDFNATIVE_ENCRYPT_OWNER_PASS / _USER_PASS)
--permissions <list> — print,copy,modify,extract for --encrypt
--stream false Constant-memory streaming output (--chunk-size <bytes>)

Multiple encrypted sources: a single --password is applied to every source. Merging encrypted sources that use different passwords is not supported and fails with E_PASSWORD. Decrypt the outliers first (pdfnative decrypt), then merge.

The same --password / --encrypt / --owner-password / --user-password / --permissions / --stream flags apply to split and extract (which take a single input, so their --password is unambiguous).

pdfnative split

Split one PDF into several (pdfnative ≥ 1.5.0 page-tree API).

Flag Default Description
--input <file> stdin Source PDF
--output-dir <dir> — (required) Directory for the output parts (created if absent)
--pages <ranges> one per page 1-based ranges; each comma-separated segment (e.g. 1-2,3-4) becomes one output
--prefix <name> input stem or part Filename prefix; outputs are <prefix>-<n>.pdf (zero-padded)
--drop-annotations false Strip annotations from each part
--max-output-size <bytes> 256 MiB Per-part size cap (0/none = unlimited)

pdfnative extract

Pull selected pages into a new PDF (pdfnative ≥ 1.5.0 page-tree API).

Flag Default Description
--input <file> stdin Source PDF
--output <file> stdout Output PDF
--pages <list> — (required) 1-based page list/range (e.g. 4,1-2); order preserved, repeats allowed
--drop-annotations false Strip annotations from the output
--max-output-size <bytes> 256 MiB Output size cap (0/none = unlimited)

(also supports --password / --encrypt / --stream — see merge above.)

pdfnative extract-text

Extract reading-order Unicode text (pdfnative ≥ 1.6.0 extractText). No OCR — image-only pages yield empty text. Tagged documents carry /ActualText, so the source text comes back even where the typography engine inserted narrow no-break spaces or soft hyphens (v1.5.0); since v1.6.0 (pdfnative ≥ 1.9.0) untagged shaped and right-to-left runs carry it too, so Arabic, Hebrew, Khmer, Myanmar or Devanagari text comes back in logical order, as typed (unless the file was rendered with --actual-text false). A content or font stream past the global --max-inflate-size is reported instead of silently yielding an empty page: --on-stream-error decides.

Flag Default Description
--input <file> stdin Source PDF
--format text|json|ndjson text Plain text (form-feed between pages), a JSON array, or NDJSON (one object per page)
--pages <list> all 1-based selector to limit pages
--runs false Include positioned runs { text, x, y, fontSize, fontName }
--password <s> — Password for an encrypted PDF (env: PDFNATIVE_PASSWORD)
--max-length <n> 16000000 Hard cap on total characters (0/none disables)
--on-stream-error limit|throw|skip limit (v1.6.0) limit: a stream over --max-inflate-size fails with E_INPUT naming both remedies (raise the cap or skip); throw: any undecodable stream fails; skip: the page reads as empty text (the 1.8.0 lenience)
--summary / --fields / --pretty — Token-economy controls (json format)

pdfnative compare

Diff two PDFs by extracted reading-order text and/or document structure — page count, page/print boxes, metadata, form fields, annotations, encryption, signatures (v1.4.0). Built for CI and agents: identical documents exit 0; any difference exits 1 with the stable code E_CHECK_FAILED (the report is written to stdout first). Visual/rasterised diffing is out of scope — pdfnative has no rasteriser.

pdfnative compare a.pdf b.pdf --mode both --format json
Flag Default Description
positional paths — (required) The two PDFs to compare, a.pdf b.pdf
--mode text|structure|both both What to compare
--format text|json text Report format (stdout)
--tolerance <pt> 0 Geometric tolerance in points for page/box sizes
--ignore-whitespace false Collapse runs of whitespace before the text diff
--pages <list> all 1-based selector limiting the text diff (e.g. 1,3-5)
--password-a <s> — Password for the first PDF
--password-b <s> — Password for the second PDF
--on-stream-error limit|throw|skip limit (v1.6.0) The same policy as extract-text, applied to both PDFs: an over-cap or undecodable stream is E_INPUT (limit / throw) or read as empty text (skip) — never a silent "equal"
--pretty — Force indented JSON even under --json

pdfnative fill

Fill, flatten, and/or export an existing AcroForm using an incremental save (existing signatures stay valid for their revision). Discover field names with inspect --form-fields, or dump the current values with --export for a read → edit → fill round-trip.

Flag Default Description
--input <file> stdin Source form PDF
--output <file> stdout Filled PDF / exported values
--data <file> — JSON object name → string|boolean|string[] (or { "values": {…} })
--flatten false Also flatten (or flatten existing values when --data is omitted)
--export false Read-only: emit current values as a --data-shaped JSON map (unset choice fields omitted); ignores --data/--flatten
--force false Flatten even if a signed signature field is present
--on-unknown throw|ignore throw Behaviour for value keys that match no field
--need-appearances false Allow non-WinAnsi values by setting /NeedAppearances
--password <s> — Password for an encrypted PDF (env: PDFNATIVE_PASSWORD)

pdfnative encrypt

Re-secure a PDF with AES-128/256 (page-tree re-encryption). Rebuilds the page tree like merge, so signatures and form fields are dropped. Requires a Web Crypto CSPRNG; RC4 is never emitted.

Flag Default Description
--input <file> stdin Source PDF
--output <file> stdout Encrypted PDF
--owner-password <s> — (required) Owner password (env: PDFNATIVE_ENCRYPT_OWNER_PASS)
--user-password <s> — User (open) password (env: PDFNATIVE_ENCRYPT_USER_PASS)
--algorithm aes-128|aes-256 aes-128 Encryption algorithm
--permissions <list> defaults print,copy,modify,extract
--password <s> — Open an already-encrypted source (password rotation)
--stream false Constant-memory streaming output (--chunk-size <bytes>)

pdfnative decrypt

Remove encryption, emitting a plaintext copy. Rebuilds the page tree like merge (signatures/forms dropped).

Flag Default Description
--input <file> stdin Encrypted source PDF
--output <file> stdout Decrypted PDF
--password <s> — Document password (env: PDFNATIVE_PASSWORD)
--stream false Constant-memory streaming output (--chunk-size <bytes>)

pdfnative doctor

Environment / capability preflight — for humans (onboarding) and agents (pre-flight before encrypt). Fully offline. Exit code 0 when all checks pass, 1 otherwise.

Flag Default Description
--format json|text text Output format (global --json also selects JSON)
--pretty — Force indented JSON even under --json

Checks: cli version, node version (≥ 22), webcrypto (CSPRNG) availability, resolved pdfnative version, registered commands count, and (v1.5.0) fonts — the bundled font inventory (36 modules / 32 scripts, each module probed on disk; error when one is missing), unicode — the engine's USE_UNICODE_VERSION, and conformance — the claimable targets (pdfa1b,pdfa2b,pdfa2u,pdfa3b,pdfx4).

pdfnative annotate

Attach markup annotations to an existing PDF using an incremental save (the original bytes — and any existing signature — are preserved).

Flag Default Description
--input <file> stdin Source PDF
--output <file> stdout Annotated PDF
--annotations <file> — (required) JSON array (or { "annotations": [...] }); each entry is a markup annotation plus a 1-based page
--password <s> — Password for an encrypted PDF (env: PDFNATIVE_PASSWORD) — appended objects are re-encrypted under the existing scheme (v1.4.0)

Annotation types: text, highlight, underline, strikeout, squiggly, square, circle, line, freetext and (v1.5.0) link. Every entry needs a page and a rect [x1,y1,x2,y2]; line also needs start and end; link needs a url (http, https or mailto — javascript: and control characters are refused with E_INPUT) and is emitted as a /Link annotation with a /URI action built by the engine's buildAnnotationBody({ type: 'link' }) (v1.6.0, pdfnative ≥ 1.9.0 — the CLI builds no PDF dictionary itself). Colours accept hex, RGB or CMYK. Only known fields are forwarded — unknown keys are dropped, so nothing can leak into the emitted dictionary. A link added to a PDF/X-4 file breaks its claim (annotations must stay outside the page boxes) — re-check with inspect --check pdfx.

pdfnative metadata

Update /Info + XMP metadata (v1.4.0). The update is an incremental save: the original bytes are preserved as a prefix, so existing digital signatures remain valid for their revision. The XMP packet is kept in sync (xmp:ModifyDate, pdf:Keywords, …). Reading metadata stays in inspect.

Flag Default Description
--input <file> stdin Source PDF
--output <file> stdout Updated PDF
--title <s> — Document title
--author <s> — Author
--subject <s> — Subject
--keywords <s> — Keywords (single string)
--mod-date <ISO 8601> now Modification date (pass a fixed value for reproducible output)
--from-json <file> — JSON file { title?, author?, subject?, keywords?, modDate? } — mutually exclusive with the per-field flags
--password <s> — Password for an encrypted PDF (env: PDFNATIVE_PASSWORD)
--dry-run false Validate inputs without writing output

At least one metadata field is required.

pdfnative govern

Expose pdfnative's AI-governance / Human-in-the-Loop (HITL) contract. Agents act as draftsmen: a human must always review and submit.

pdfnative govern rules                 # human/agent protocol (AGENT_RULES) on stdout
pdfnative govern policy                # machine-readable policy JSON on stdout
pdfnative govern verify-issue draft.md # validate a draft; exit 1 / E_POLICY on violation

verify-issue fails a draft that proposes an external runtime dependency or omits a reproduction code block; missing recommended fields (environment, expected behaviour) are surfaced as warnings. Add --json for a machine-readable { ok, errors, warnings } report. A passing check is necessary but not sufficient — the human review gate always applies.

pdfnative completion

pdfnative completion bash > /etc/bash_completion.d/pdfnative
pdfnative completion zsh  > "${fpath[1]}/_pdfnative"
pdfnative completion fish > ~/.config/fish/completions/pdfnative.fish
pdfnative completion powershell >> $PROFILE   # Register-ArgumentCompleter

pdfnative schema

Print a versioned JSON Schema (Draft 2020-12) for a CLI input/output shape, so an agent can self-validate before invoking a command.

pdfnative schema            # render input schema (default)
pdfnative schema render     # render input (document | table variant)
pdfnative schema inspect    # inspect --format json output
pdfnative schema verify     # verify  --format json output
pdfnative schema batch      # batch   --format json output
pdfnative schema annotate       # annotate --annotations input
pdfnative schema extract-text   # extract-text --format json output
pdfnative schema fill           # fill --data values input
pdfnative schema form-export    # fill --export output
pdfnative schema govern-verify  # govern verify-issue --json output
pdfnative schema metadata       # metadata --from-json input (v1.4.0)
pdfnative schema ltv-data       # replayable JSON emitted/consumed by ltv collect/embed (v1.4.0)
pdfnative schema compare        # compare --format json output (v1.4.0)
pdfnative schema batch-manifest # batch --manifest pipeline file (v1.4.0)
pdfnative schema status         # the --json success envelope (write commands)
pdfnative schema manifest       # capability manifest: commands, flags, error codes
pdfnative schema doctor         # doctor --format json output
pdfnative schema list       # list the available subjects (19 subjects)

The manifest (schema manifest) is a machine-readable capability document — every command, its flags, the global flags, and the stable error codes — for AI-agent tool discovery. A prose/LLM-facing version ships as llms.txt at the package root.

Global options

Global options may be placed before or after the command name (v1.5.0): pdfnative --json --dry-run render --input doc.json and pdfnative render --input doc.json --json --dry-run are equivalent.

Flag Description
--config <file> Use a specific .pdfnativerc.json (default: nearest upward from cwd)
--no-config Ignore any .pdfnativerc.json
--quiet, -q Suppress progress output on stderr
--no-color Disable ANSI colour (also respects the NO_COLOR env var)
--json Agent mode: emit a JSON status/error envelope on stderr (data stays on stdout)
--dry-run Validate inputs and exit without writing output (render / sign / batch / merge / split / extract / annotate / fill / encrypt / decrypt / metadata / ltv / doc-timestamp). render pre-flights the real build in memory and discards the bytes, so every engine coherence error (E_INPUT), --strict escalation (E_CHECK_FAILED) and diagnostic surfaces exactly as on a real run (v1.5.0). Never performs network I/O, even when a network flag is present
--max-inflate-size <bytes> Cap the decompressed size of any single PDF stream while parsing untrusted input (anti zip-bomb; default 100 MiB) — v1.4.0
--creation-date <iso8601> (v1.5.0) Pin the creation instant of every PDF written in this run (render, batch): /CreationDate, xmp:CreateDate, the {date} placeholder and the trailer /ID derive from it, in UTC, so output is byte-identical on every host. Falls back to $SOURCE_DATE_EPOCH (integer seconds); an invalid value is a usage error. Encrypted output is never byte-reproducible (CSPRNG keys); sign --signing-time and metadata --mod-date are separate instants. The envelope carries creationDate
--version --json Machine-readable version output

Driving from AI agents

pdfnative-cli is designed so an autonomous agent (or any program) can drive it deterministically — no MCP server, no daemon, just the process contract:

  • stdout = the artifact (PDF, JSON report, schema, completion script); stderr = diagnostics.
  • Pass --json to get a single machine-readable envelope on stderr. On failure: { "ok": false, "command": "...", "error": { "code": "E_*", "message": "..." } }. On success for the write commands (render / sign / batch / merge / split / extract / annotate / fill / encrypt / decrypt / metadata / ltv / doc-timestamp): a { "ok": true, ... } status line.
  • Branch on the stable error code (E_USAGE, E_INPUT, E_PARSE, E_IO, E_SIGN, E_VERIFY_FAILED, E_CHECK_FAILED, E_POLICY, E_UNSUPPORTED, E_PASSWORD, E_NETWORK, E_RUNTIME) rather than the message text. Numeric exit codes stay 0 (success), 1 (runtime), 2 (usage).
  • Use --dry-run to validate input without producing output.
  • Fetch a schema (or schema manifest / llms.txt) to discover and validate before calling, and run doctor --format json as a capability pre-flight.
  • Pin --creation-date (or SOURCE_DATE_EPOCH) so a re-run of the same job yields the same bytes, and let compare or a SHA-256 be the assertion.
  • Rely on the compatibility promise: within 1.x no command, flag, error code, exit code, schema subject or envelope field is removed or renamed — a minor only adds — and every pull request replays the previous release's suite and schema captures against the current sources (SECURITY.md §Compatibility promise).

See docs/AGENT_CONTRACT.md (the full contract) and the samples/agent/ scripts. Agents working on this repository read AGENTS.md instead.

Security

  • Offline by default — no network access unless you explicitly opt in with verify --revocation online, sign --timestamp <url>, ltv collect|add --online, doc-timestamp --url, or batch --allow-network (which gates network flags inside a manifest). Every request passes an SSRF guard (scheme allow-list; private / loopback / link-local / CGNAT / multicast / benchmarking / TEST-NET / NAT64 address blocking; no redirects; timeout + size caps). A failed opt-in network operation maps to the stable E_NETWORK code — never a silent fallback.
  • Anti zip-bomb cap — the global --max-inflate-size <bytes> bounds the decompressed size of any single PDF stream while parsing untrusted input (default 100 MiB).
  • Signing keys are never logged — not in error messages, not in debug output.
  • Path traversal protection — all file path arguments are validated against ../ sequences.
  • Bounded, validated inputs — JSON documents and --layout files are capped at 50 MB before parsing; ICC profiles (--output-intent-icc, 16 MiB) and custom fonts (--font-file, 32 MiB) are checked by magic bytes and by pdfnative's parsers before use, and can only be named on the command line — never from a JSON payload.
  • Weak digests surfaced — verify names a SHA-1 timestamp imprint and refuses it under --strict; the CLI itself always requests SHA-256+ imprints.
  • Reproducible builds — --creation-date / SOURCE_DATE_EPOCH make output byte-identical across hosts; the repository's sample corpus is held to a committed SHA-256 baseline.
  • Supply chain — a four-job release pipeline (guard → build → publish → attest, v1.6.0): the tarball is packed once, verified by digest before an npm client pinned by SHA-512 uploads it through Trusted Publishing (OIDC, --provenance), then fetched back from the registry and attested with CycloneDX, SPDX and toolchain SBOMs (SLSA Build L2; verify an install with npm audit signatures); SHA-pinned, harden-runner and zizmor/actionlint-linted workflows, CodeQL, Dependabot with a cooldown, ClusterFuzzLite over the CLI's parsers; the gate as a required check on Ubuntu, Windows and macOS plus the previous release's suite (compat-previous); one release gate (npx tsx scripts/gate.ts --publish --require-all) with a veraPDF PDF/A step, a PDF/X step and the compatibility replay.

See SECURITY.md for the full security policy and vulnerability disclosure procedure.

Getting Help

Have a question?

Found a bug?

Want to contribute?

  • 🤝 See CONTRIBUTING.md
  • 📝 All PRs add value — tests, docs, translations, samples

Related Projects

Citation

If you use pdfnative-cli in research or academic pipelines, please cite it:

@software{pdfnative_cli_2026,
  title  = {pdfnative-cli: Official CLI for the pdfnative PDF generation library},
  author = {Nizoka},
  year   = {2026},
  url    = {https://github.com/Nizoka/pdfnative-cli},
  license = {MIT}
}

See CITATION.cff for the full metadata (auto-detected by GitHub and Zenodo).

License

MIT — see LICENSE.

About

High-performance, zero-dependency CLI for the pdfnative engine. Automate JSON-to-PDF/A rendering, digital signatures (RSA/ECDSA), and compliance inspection within CI/CD pipelines via shell-pipe workflows.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Used by

Contributors

Languages