/pdf — Export Any Doc to PDF

Convert a framework-generated document (markdown, HTML, BPMN) to PDF for sharing with non-technical stakeholders, board members, customers, or auditors.

Sits alongside /c4 (Mermaid markdown), /dfd (Mermaid markdown), /tech-vision (markdown), /write-spec (PRD markdown), /journey (self-contained HTML), /process (BPMN XML), and the audit family (/threat-model, /launch-check, etc., all of which write dated markdown audits). Those skills emit a single source-of-truth artefact in its native format; /pdf is the dedicated bridge to PDF for the moments when prose isn't enough.

The destination question

A PDF can land in one of two places, and the answer depends on whether the doc should travel with the code if the project spun out tomorrow. This mirrors the existing rule in docs/multi-project.md:

If YES (travels with the code)	If NO (ApexYard's view)
Project's own repo: workspace/<name>/docs/	Ops fork: projects/<name>/pdfs/
Examples: API spec, deployment runbook, internal sequence	Examples: handover assessment, stakeholder update, launch-check verdict

The skill asks, doesn't guess. The 4-option prompt below covers every common case.

Usage

/pdf projects/curios-dog/architecture/vision.md
/pdf workspace/curios-dog/docs/architecture/context.md
/pdf projects/curios-dog/audits/security/2026-05-19.md
/pdf projects/curios-dog/journeys/checkout-v2.html
/pdf projects/curios-dog/processes/onboarding.bpmn
/pdf <input> --no-prompt                  # use default_destination from config
/pdf <input> --converter=pandoc           # force a specific converter
/pdf <input> --destination=workspace      # skip the prompt, write to workspace/<name>/docs/
/pdf <input> --destination=projects       # skip the prompt, write to projects/<name>/pdfs/
/pdf <input> --destination=keep           # skip the prompt, keep next to source
/pdf <input> --destination=/absolute/path/out.pdf  # explicit path
/pdf <input> --project=curios-dog         # override auto-detected project name

Path resolution

Read workspace_dir and projects_dir from .claude/hooks/_lib-portfolio-paths.sh so split-portfolio adopters resolve to the sibling private repo transparently:

source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-read-config.sh"
source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-portfolio-paths.sh"
projects_dir=$(portfolio_projects_dir)
workspace_dir=$(portfolio_workspace_dir)

Defaults to single-fork (./projects, ./workspace). Don't hardcode literal projects/ or workspace/ paths in the bash blocks below — let the helper resolve whichever mode the adopter is in.

Process

1. Resolve the input file

INPUT="$1"
if [ ! -f "$INPUT" ]; then
  echo "/pdf: input file not found: $INPUT" >&2
  exit 2
fi
# Absolute path for downstream resolution
ABS_INPUT=$(cd "$(dirname "$INPUT")" && pwd)/$(basename "$INPUT")

2. Sniff the input format

By extension:

Extension	Format
.md, .markdown	Markdown
.html, .htm	HTML
.bpmn, .bpmn20.xml	BPMN

Anything else → exit 2 with a clear "unsupported input format" message + the supported list.

3. Auto-detect <name> from the input path

The destination prompt needs a project name to fill in. Inference order:

1. If --project=<name> was passed → use it.
2. If ABS_INPUT is under <projects_dir>/<name>/... → name is the path segment after projects_dir.
3. If ABS_INPUT is under <workspace_dir>/<name>/... → name is the path segment after workspace_dir.
4. If neither matched (e.g. cwd is ops-fork root and input is docs/foo.md) → name is unresolved. The prompt will show "(no project — supply via --project)" in slots 1 + 2, and slots 3 + 4 remain valid.

4. Show the destination prompt

Always show this prompt unless --no-prompt or --destination=... was passed.

Where should the PDF land?

  (1) workspace/<name>/docs/<stem>.pdf  ← travels with the code
  (2) projects/<name>/pdfs/<stem>.pdf   ← ApexYard's view
  (3) <custom path>                     ← anywhere
  (k) keep next to source               ← <input-dir>/<stem>.pdf

  Hint: pick (1) if a downstream reader of the project repo would want
  this PDF (API spec, deployment runbook). Pick (2) if it's framework
  context (handover, stakeholder update, audit). Pick (k) when in doubt.

>

When <name> couldn't be resolved, slots (1) and (2) print as (no project — supply via --project) and accepting them prompts for a name.

5. Compute the output path

STEM=$(basename "$INPUT")
STEM="${STEM%.*}"   # strip the extension

Filename rule:

• Default → <stem>.pdf
• Audit-class outputs: if ABS_INPUT matches <projects_dir>/<name>/audits/<dim>/<YYYY-MM-DD>.md (the dated-subdir convention from AgDR-0019), keep the date in the filename — the stem already contains it, so <stem>.pdf is correct as-is. No special case needed.

By destination:

Destination	Output path
1 / workspace	<workspace_dir>/<name>/docs/<stem>.pdf
2 / projects	<projects_dir>/<name>/pdfs/<stem>.pdf
3 / <path>	the operator-supplied path (relative to cwd, or absolute)
k / keep	<dir-of-input>/<stem>.pdf

If the parent dir doesn't exist, mkdir -p it.

If the output file already exists, ask the operator: overwrite (o), pick a new path (n), or quit (q). No silent overwrite.

6. Run the converter

Delegate to convert.sh (the skill's converter-dispatch helper). It takes --from, --to, optional --converter=<name>, optional --pdf-engine=<eng>, and outputs to --out.

SKILL_DIR="$(dirname "$(realpath "$0")")"
"$SKILL_DIR/convert.sh" \
  --from="$ABS_INPUT" \
  --to="$OUT" \
  ${CONVERTER:+--converter="$CONVERTER"} \
  ${PDF_ENGINE:+--pdf-engine="$PDF_ENGINE"}
RC=$?

convert.sh exit codes:

• 0 — converted cleanly
• 1 — conversion failed (offending converter output streamed to stderr)
• 2 — bad input / unsupported format
• 3 — no converter available; advisory printed to stderr naming each install option. The skill propagates this exit code.

7. Report

On success:

✓ PDF written: <OUT>
  Source:     <ABS_INPUT>
  Format:     <markdown|html|bpmn> → PDF
  Converter:  <pandoc|md-to-pdf|wkhtmltopdf|bpmn-to-image+pandoc>
  Size:       <size>

On exit 3:

✗ No PDF converter is installed.

Markdown inputs can use:
  • pandoc           — brew install pandoc (or apt-get install pandoc)
                       For best output also install xelatex (mactex / texlive-xetex)
  • md-to-pdf (npm)  — npm install -g md-to-pdf  (or run via npx, no install)

HTML inputs can use:
  • wkhtmltopdf      — brew install --cask wkhtmltopdf
  • pandoc           — same as above (uses its HTML reader)

BPMN inputs need a two-step pipeline:
  • bpmn-to-image (npm) → SVG → pandoc → PDF

Install at least one of the above and re-run /pdf.

The skill does not silently fall back to leaving you without a PDF — it explicitly exits 3 so the operator knows the gap and can fix it.

Config

.claude/project-config.defaults.json ships a pdf block:

"pdf": {
  "preferred_converter": "pandoc",
  "pdf_engine": "xelatex",
  "default_destination": "ask"
}

• preferred_converter — when both pandoc and a fallback are installed, prefer this one. null means "first one found in the dispatch order".
• pdf_engine — passed to pandoc as --pdf-engine=<engine>. Common values: xelatex (best Unicode), pdflatex (smaller install), wkhtmltopdf (no LaTeX needed but degraded typography).
• default_destination — used when --no-prompt is passed. Must be one of workspace, projects, keep, or ask. ask with --no-prompt is an error (the skill exits 2 — operator must change config or drop --no-prompt).

Adopters override in .claude/project-config.json:

{
  "pdf": {
    "preferred_converter": "md-to-pdf",
    "default_destination": "keep"
  }
}

Rules

1. Always ask about destination unless --no-prompt or --destination=... was passed. The "would it follow the code?" question is genuinely contextual; guessing wrong creates landed-in-the-wrong-repo cleanup work.
2. Never silently overwrite an existing PDF. Always prompt or require --destination=<explicit-path> which the operator owns.
3. Graceful degrade only on missing dep, not on conversion failure. Missing converter → exit 3 + advisory. Converter installed but threw an error → exit 1 + propagate the converter's stderr. The two are different failure modes with different fixes.
4. Source format detection by extension only. We don't sniff content (would add complexity for marginal value — operators know what they're converting).
5. No special-case templates. v1 uses the converter's defaults. Custom LaTeX templates, branded header/footer, etc. are out of scope (separate ticket).
6. No batch mode. One input → one PDF. If the operator needs ten, they loop in shell. Keeps the skill's destination-prompt logic single-purpose.

When to use this

Trigger	Use /pdf?
Board needs a one-pager PDF of a PRD	Yes
Sharing an audit verdict with a customer	Yes
Customer-facing API documentation	Yes (write to workspace/<name>/docs/)
Internal stakeholder update	Yes (write to projects/<name>/pdfs/)
Quick-share a diagram from /c4 for a meeting	Yes (use keep to drop next to the source)
Replace the source markdown with the PDF	No — the source stays canonical; PDF is a render
Batch-converting 50 audit reports	No — loop /pdf in shell, the skill stays single-input

Out of scope (v1)

• --pdf flag on each doc-emitting skill that converts at write-time. Cleaner one-command UX but couples every doc skill to the converter dep. Standalone /pdf is v1; per-skill integration can land in v1.5 if operator usage demands it.
• Custom LaTeX templates / branding / headers / footers. Use system pandoc defaults for v1.
• Batch export across multiple inputs. Loop in shell.
• Auto-watch + regenerate on source change. Out of scope.
• PDF accessibility audit (tagged structure, alt-text propagation). The accessibility-audit skill covers the source markdown; PDF accessibility is a downstream concern.

See also

• AgDR-0034 — converter dispatch + destination prompt rationale + standalone-skill vs flag-on-each-skill decision
• docs/multi-project.md § "Architecture diagrams" — the "would it follow the code?" rule extended to PDF outputs
• .claude/skills/process/lint.sh — graceful-degrade-on-missing-dep pattern that this skill mirrors for converter detection
• .claude/skills/_lib-mermaid-lint.sh — same pattern, npx-fallback case

---

Part of ApexYard — multi-project SDLC framework for Claude Code · MIT.