Local Md Mermaid Pdf

Local MD & Mermaid to PDF

Overview

Converts one or more Markdown files with Mermaid diagrams into PDF(s) using local mmdc and md-to-pdf. Each source file gets its own PDF (-export.pdf beside the source). All intermediate artifacts live in local-md-mermaid-pdf-sandbox; only the final PDF(s) are written outside the sandbox. The sandbox is removed after each successful run.

Multiple files: process sequentially (one complete run at a time). Produce one PDF per source file. Do not merge PDFs unless the user explicitly asks.

When to Use

Use when

• The user asks to export Markdown to PDF
• The source file contains Mermaid diagram blocks
• The user wants styled PDF output with page numbers from local tools

Do not use when

• The file has no Mermaid diagrams and a simple MD→PDF path is enough (still works, but heavier than necessary)
• The user wants cloud/API-based conversion instead of local binaries
• The user asks to modify the source Markdown (this skill only exports)

Inputs

• Source Markdown file path (with optional Mermaid `mermaid blocks)
• style.css from this skill folder (unless the user requests custom styling)
• Local binaries mmdc and md-to-pdf, or npx --yes fallback
• Google Chrome or Chromium for Puppeteer (executablePath when available)

Outputs

• PDF at /-export.pdf — only deliverable outside the sandbox
• Multiple sources: one PDF per file (same -export.pdf rule beside each source). List every absolute PDF path in the final reply
• Merge: only when the user explicitly requests a single combined PDF (e.g. "merge", "combine"). Otherwise never merge
• Sandbox local-md-mermaid-pdf-sandbox removed after each successful run (all intermediate artifacts deleted with it)

Constraints

• All artifacts go in the sandbox — every file created during the run (input.md, input.tmp.md, input.for-pdf.md, puppeteer-config.json, Puppeteer cache, Mermaid render outputs) must live inside local-md-mermaid-pdf-sandbox. Only the final PDF is written outside the sandbox (via dest beside the source file)
• Use style.css from this skill folder unless the user asks for custom styling (read-only reference; do not copy into the project unless the user requests custom styling)
• Page numbers via md-to-pdf front matter: Page of
• Use a sandbox-local Puppeteer cache (PUPPETEER_CACHE_DIR inside the sandbox); never depend on ~/.cache/puppeteer
• Prefer local binaries; use npx --yes only when a binary is missing
• If Chrome or Chromium exists, pass executablePath to Puppeteer
• On Puppeteer/cache failures, retry only with the documented Chromium fallback; do not invent launch flags
• Do not leave the sandbox directory after a successful run
• Multiple source files — sequential only: run the full workflow (steps 1–5) for file A, wait until it finishes (PDF written, sandbox deleted), then start file B. Never launch parallel conversions, shared sandboxes, or concurrent mmdc / md-to-pdf / Puppeteer runs
• Multiple source files — no merge by default: each source becomes its own -export.pdf. Do not concatenate Markdown sources, merge PDFs, or produce one combined output unless the user explicitly asks

Steps

1. Create sandbox

• Create local-md-mermaid-pdf-sandbox next to the source file
• All workflow artifacts stay inside this directory — do not write intermediate files next to the source or elsewhere
• Copy the source Markdown to local-md-mermaid-pdf-sandbox/input.md
• Set PUPPETEER_CACHE_DIR to a path inside the sandbox (e.g. local-md-mermaid-pdf-sandbox/.puppeteer-cache)

2. Render Mermaid

• Before running mmdc, if Google Chrome or Chromium is available, write local-md-mermaid-pdf-sandbox/puppeteer-config.json with system executablePath and args: ["--no-sandbox"]. This is required when PUPPETEER_CACHE_DIR points to the empty sandbox cache — without it, mmdc cannot find Chrome and diagrams will not render
• Run mmdc -p puppeteer-config.json -i input.md -o input.tmp.md when the config exists; otherwise mmdc -i input.md -o input.tmp.md
• If mmdc emits images or other sidecar files, they must remain inside the sandbox
• If mmdc still fails on Puppeteer/cache, retry only with a corrected executablePath in puppeteer-config.json — do not skip to PDF conversion
• Verify input.tmp.md exists and no longer contains raw `mermaid blocks (expect `` image references and sidecar SVG/PNG files)

3. Build PDF input

Write local-md-mermaid-pdf-sandbox/input.for-pdf.md with YAML front matter:

• dest: /-export.pdf — the only output path outside the sandbox
• stylesheet: absolute path to this skill’s style.css
• pdf_options.displayHeaderFooter: true
• pdf_options.headerTemplate: ''
• pdf_options.footerTemplate: centered page numbers only
• Body: contents of input.tmp.md

4. Convert to PDF

• Build --launch-options JSON (executablePath when Chrome/Chromium is available; args: ["--no-sandbox"])
• Run md-to-pdf --basedir local-md-mermaid-pdf-sandbox --launch-options '' input.for-pdf.md from inside the sandbox (or npx --yes md-to-pdf@5.2.5 ...)

5. Clean up and report

• Delete local-md-mermaid-pdf-sandbox
• Reply with the final PDF path (or all PDF paths when multiple sources were converted)

Multiple files

When the user provides more than one Markdown file:

1. Confirm the file list (paths and order if order matters to the user)
2. Queue: convert file 1 through step 5 completely before starting file 2
3. One PDF per file: each output is /-export.pdf beside its source
4. No merge unless the user explicitly requested a single combined PDF — if they did, use a separate merge step after all individual PDFs exist (outside this skill's default path)
5. Final reply lists every generated PDF path

Rationalization Traps

| Rationalization | Reality | | --- | --- | | Skip Mermaid render for MD without diagrams | mmdc is still required when diagrams exist; inspect the file first | | Run mmdc without puppeteer-config.json when sandbox cache is empty | mmdc will fail to find Chrome; write config with system executablePath before the first run | | Skip to PDF when mmdc fails | Never convert without input.tmp.md; raw `mermaid blocks do not render in PDF | | Reuse ~/.cache/puppeteer | Breaks isolation and causes cross-project cache conflicts | | Invent Puppeteer flags on failure | Only the documented Chromium executablePath retry is allowed | | Keep sandbox for debugging | Sandbox must be removed after success; use --keep in scripts/e2e.sh for local dev only | | Overwrite the source PDF name | Use the stable -export.pdf suffix to avoid clobbering prior runs | | Write intermediates beside the source file | Only the final PDF leaves the sandbox; everything else stays in local-md-mermaid-pdf-sandbox | | Run multiple conversions in parallel to save time | Sequential queue only — parallel runs conflict on Puppeteer, cache, and sandbox paths | | Merge multiple files into one PDF by default | One PDF per source; merge only when the user explicitly asks | | Reuse one sandbox for several source files at once | Fresh sandbox per file; complete and delete before the next file |

Red Flags

• Intermediate files (input.md, input.tmp.md, input.for-pdf.md, puppeteer-config.json, cache) exist outside local-md-mermaid-pdf-sandbox
• PDF missing or smaller than ~1 KB after conversion
• input.tmp.md was not produced by mmdc
• input.tmp.md still contains `mermaid blocks (diagrams were not rendered)
• Sandbox directory still exists after a successful run
• stylesheet points outside this skill folder without user request
• npx used when global mmdc / md-to-pdf binaries are already available
• Multiple files converted in parallel or from a shared sandbox
• A single merged PDF produced without an explicit user request to merge/combine

Verification

• [ ] All intermediate artifacts are inside local-md-mermaid-pdf-sandbox (no stray files beside the source)
• [ ] local-md-mermaid-pdf-sandbox/input.tmp.md exists after mmdc and has no raw `mermaid blocks
• [ ] local-md-mermaid-pdf-sandbox/input.for-pdf.md has dest, stylesheet, and footer page-number template
• [ ] -export.pdf exists beside the source file
• [ ] PDF size is greater than 1 KB
• [ ] local-md-mermaid-pdf-sandbox was deleted
• [ ] User received the absolute PDF path
• [ ] Multiple files: each source has its own -export.pdf; conversions ran sequentially (not in parallel)
• [ ] Multiple files: no merged/combined PDF unless the user explicitly requested it

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: wanderlima
• Source: wanderlima/agent-skills
• License: MIT
• Homepage: https://github.com/wanderlima/agent-skills

Install and usage instructions live in the source repository linked above.