Skip to content
Development
Skill

/html-doc

Generates a polished, static HTML reading view from an existing Markdown PRD or technical design. Use when a user asks to render, present, visualize, or make a PRD or design document easier for humans to read in a browser.

From plugin
blueprint
43111 skills

Claude Code

$ npx -y skills add owainlewis/blueprint --skill html-doc --agent claude-code

SKILL.md

html-doc.SKILL.md
name: html-doc
description: Generates a polished, static HTML reading view from an existing Markdown PRD or technical design. Use when a user asks to render, present, visualize, or make a PRD or design document easier for humans to read in a browser.
user-invocable: true
argument-hint: "<Markdown PRD or design file>"
disable-model-invocation: true

HTML document

Turn one complete Markdown PRD or technical design into one verified static HTML page. Keep Markdown canonical. Change presentation, not meaning.

Boundaries

  • Accept an existing Markdown file or exact inline Markdown.
  • Support only `prd` and `design` document kinds.
  • For `REQUIREMENTS.md`, pass `--kind prd`. For `ARCHITECTURE.md` or `spec.md`, pass `--kind design`.
  • Classify `prd.md` as `prd` and `design.md` as `design`. For other names, use the request or ask if the kind is unclear.
  • Do not write requirements, settle design choices, summarize missing content, or edit the source.
  • If the source is only a rough brief, use the appropriate authoring skill first.
  • Do not hand-edit generated HTML. Change the template or renderer, then regenerate.
  • Do not publish or host the result unless the user separately asks.

Requirements

Generation requires Python 3.11 or newer, Pandoc 3 or newer, Node.js 22.12 or newer, and npm. The skill directory contains a locked Mermaid dependency tree.

Before the first Mermaid render, run:

npm ci --prefix <skill-directory>

Use `scripts/html_doc.py` from this skill directory for all generation and lifecycle operations.

Workflow

1. Read and classify the source

Read the complete source. Confirm it is no larger than 2 MiB. Determine the document kind from the request or the basename mappings above.

Treat content problems separately from presentation problems. If the Markdown contradicts itself, omits a required decision, or contains an invalid diagram, report that source problem. Do not silently repair its meaning.

2. Create a candidate

For a file source:

python3 <skill-directory>/scripts/html_doc.py render <source.md> --kind <prd-or-design> --out <output.html>

`--kind` may be omitted only for exact `prd.md` and `design.md` basenames. `--out` may be omitted to create a sibling HTML file.

For exact inline Markdown, pipe it through standard input and provide both kind and output:

python3 <skill-directory>/scripts/html_doc.py render --inline --kind <prd-or-design> --out <output.html>

The command returns JSON containing `candidate`, `output`, and `token`. It does not replace the final output. Keep those exact values for verification and finalization.

If an output lock already exists, do not delete it automatically. It may represent another pending render. Report its path. Use `discard --force-stale` only when the state is known to be abandoned.

3. Verify the candidate in a real browser

Open the candidate as a local file. Do not treat source inspection as browser proof.

Check all of these:

  • desktop at 1440 CSS pixels;
  • tablet at 768 CSS pixels;
  • mobile at 390 CSS pixels;
  • A4 and US Letter print preview;
  • table-of-contents links and section targets;
  • keyboard order from skip link through navigation and document links;
  • visible focus indicators;
  • one `main` landmark and a sensible heading outline;
  • unique DOM IDs;
  • horizontal overflow, clipped text, code, tables, and diagrams;
  • useful diagram alternatives and accessible disclosure controls;
  • console errors and failed resource requests;
  • operation with browser network access disabled.

Compare the canonical content inside `main` with the Markdown abstract syntax tree. Navigation, source metadata, and derived accessibility text are not canonical content. Each top-level source block must map to one ordered `data-source-block` node with its SHA-256 identity and nested content intact. A Mermaid block maps to one figure whose exact source appears once in its disclosure.

If a presentation defect comes from the shared renderer or styles, fix that source and regenerate. If it is a content defect, leave the source and candidate unchanged and report it.

4. Finalize or discard

Only after every check passes, atomically install the candidate:

python3 <skill-directory>/scripts/html_doc.py finalize --out <output.html> --token <token>

If any check fails, remove only this pending candidate:

python3 <skill-directory>/scripts/html_doc.py discard --out <output.html> --token <token>

Finalization performs one same-directory atomic replacement. It never deletes the previous output first. A failed render, check, or replacement must leave the previous verified output byte-for-byte unchanged.

Output standard

The final page must:

  • preserve source wording and order;
  • use canonical `prd` or `design` machine metadata;
  • record the source path and SHA-256 hash;
  • use semantic HTML, inline CSS, and no runtime JavaScript;
  • contain no external runtime assets;
  • use a restrained blue-grey visual system;
  • remain readable without color, a mouse, a wide screen, or network access;
  • include responsive and print styles;
  • embed validated Mermaid SVG as a base64 image and retain the exact source in a disclosure.

Return

Stop when the verified candidate is finalized. Report the output path, source hash, and browser checks performed.

Read more
@owainlewis
Ships withblueprint

Senior engineering habits for coding agents. Small skills for Claude Code and Codex that turn an idea into a reviewed, tested pull request.

Stars on GitHub

Momentum

stars in total
431
Star history of owainlewis/blueprint on GitHub, a chart by star-history.com

Other skills on blueprint.

architecture-review

architecture-review

Reviews a technical proposal before any code is written: an architecture, spec, RFC, ADR, or design issue. Finds choices that…

architecture

architecture

Writes root ARCHITECTURE.md from a fixed template: the intended system, a data flow diagram, the data model, and the rules every…

improve

improve

Makes existing code easier to understand without changing behavior. Use to simplify structure, remove duplication or dead code,…

plan

plan

Splits a spec into small tasks, each a working slice an agent can finish in one pull request. Use when a spec is too big for one…

requirements

requirements

Writes the product requirements document (PRD) in root REQUIREMENTS.md as user stories with checkable outcomes. Use for a new…

More skills from other plugins.

brainstorming
Auto-invoked

brainstorming

You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or…

@obra@obraInstall Skill
ask-matt

ask-matt

Ask which skill or flow fits your situation. A router over the skills in this repo.

skills11d309k
@mattpocock@mattpocockInstall Skill
accessibility
Auto-invoked

accessibility

Design, implement, and audit accessible UI to WCAG 2.2 Level AA across Web, iOS, and Android — semantic ARIA roles and labels,…

ecc13d293k
@affaan-m@affaan-mInstall Skill
compiler-commit

compiler-commit

Use when you want to verify compiler changes and commit with the correct convention. Runs tests, lint, and format, then commits…

react2d252k
@facebook@facebookInstall Skill

Questions about html-doc.

Which coding agents can use html-doc?
html-doc installs in Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode and Windsurf with the skills CLI, using each agent's flag shown on this page. Flowy's auto-invocation, which fires a skill when your prompt matches, runs in Claude Code only.