Super Document Skills

Turn rough notes into visual reports, presentations, and decision briefs.

Six free, open-source skills for AI agents. Bring your notes and tell your agent who the document is for. Get a local HTML document you can open, edit, and share. No account or publishing connection required.

See the examples · Install · Try a complete sample

New: decision decks you'd actually send

Not happy with the decks Claude or Codex make? I gave both agents the same 12 product decisions. A blind reviewer read every slide and said whether they'd send each deck as it is.

	Claude Opus 5.5	Codex (GPT-6.1 Sol)
On its own: would send	3 of 12	3 of 12
With this skill: would send	9 of 12	8 of 12

Across all 24 decks, the skill took invented statements from 16 decks to 5, broken layouts from 11 to 0, and template-looking decks from 8 to 0. It isn't a longer prompt. It's six small pieces, each fixing one thing agents get wrong, and the agent stays free everywhere else:

• A router picks the method for the job: an executive decision gets the decision recipe, a talk or lesson gets the general one.
• Argument rules: the answer and the ask on slide 1, every option (including waiting) with the same numbers, only facts from your notes.
• A starter layout in CSS: a fixed frame with the title at the top and a balanced body, like a well-built web page. Lopsided slides went from 54 of 166 to 0.
• Style as a few settings, so the agent still draws its own charts for your decision.
• A checker that screenshots every slide in your Chrome so the agent fixes its own overflow, stray labels and empty space.
• Your instructions win: ask for your brand, your template or your own layout and the agent follows you.

# Claude Code
npx skills add arunai30/super-document-skills --agent claude-code --skill create-html-presentation --copy
# Codex
npx skills add arunai30/super-document-skills --agent codex --skill create-html-presentation --copy

Then give your agent your notes and the decision you need. How it was tested, with every number.

New: talks and lessons that say only what you gave them

Teaching a class, giving a meetup talk, opening a conference? I gave Claude and Codex the same 12 talk and lesson requests, three of them requests the skill was never tuned on. A blind reviewer read every slide and said whether they'd send each deck as it is.

	Claude Opus 5.5	Codex (GPT-6.1 Sol)
On its own: would send	2 of 12	4 of 12
With this skill: would send	12 of 12	12 of 12

What the talk recipe fixes:

• No facts you didn't give. On its own, a moon-phases lesson gained "0–3 eclipses a year" and a "blood moon". With the skill, every fact, number and story comes from your notes; on held-out talks, facts presented as real but not in the notes went from 39 to 0 (Claude) and 9 to 1 (Codex).
• Slides that support you, not your script. About 30 words a slide; speaker notes go in the reply. No timers, "click to reveal" or placeholders left on a slide.
• Titles that make the point. Each title is a full sentence, so the titles alone tell the talk; under it, a labelled diagram, a worked example or one number.
• Lessons that keep every step. The example worked in stages, the common mistake shown as a mistake, and the practice answer after the question.
• No AI look. No emoji titles, rows of identical cards or gradients.
• The same deck for another audience. Ask for an engineering or sales version and the skill keeps the facts, changes the emphasis and the ask, and adds no plans of its own.

Install it with the same commands as the decision deck above; the skill picks the talk recipe for anything that isn't a product decision, update or roadmap. How it was tested, with every number.

New: start from the doc you already wrote

Already wrote the memo? Give your agent the doc and ask for the deck. The skill keeps your conclusion, your reasoning and your caveats, and adds nothing the doc doesn't say. Before this change, every test deck made from a finished memo added decision criteria and conditions the memo never stated; after it, a blind reviewer picked the new deck 8 times out of 8 and would send all 8 as they are. How it was tested.

New: roadmap decks your leadership can agree to

A roadmap deck has one job: get leadership to agree what gets built, in what order, and what that pushes out. I gave Claude and Codex the same 12 roadmap requests, three of them requests the skill was never tuned on. A blind reviewer read every slide and said whether they'd send each deck as it is.

	Claude Opus 5.5	Codex (GPT-6.1 Sol)
On its own: would send	0 of 12	2 of 12
With this skill: would send	12 of 12	11 of 12

What the roadmap recipe fixes:

• No plans you didn't make. On their own, every deck added owners, deadlines, checkpoints or "decision needed today". With the skill, dates, estimates and next steps come only from your notes.
• No recommendation you didn't give. When the reader has to choose, the options are set out alike.
• Evidence at its real size. No "biggest churn driver" for a reason 31% of exit surveys gave.
• Honest timelines. Quarters or now / next / later, whichever your leader wants, with confidence falling with distance and real commitments marked.
• Tradeoffs in the open. What each order delays and a short "not now" list, on their own slides.
• Customer roadmaps that are safe to send. No customer names, revenue, estimates or slips; uncertain items not promised; quarters labelled; no added marketing claims.

Install it with the same commands as the decision deck above; the skill picks the roadmap recipe when you ask for a roadmap. How it was tested, with every number.

New: product updates that don't make things up

Your VP reads your update to see where things stand and what you need. I gave Claude and Codex the same 12 product updates, three of them requests the skill was never tuned on. A blind reviewer read every slide and said whether they'd send each deck as it is.

	Claude Opus 5.5	Codex (GPT-6.1 Sol)
On its own: would send	0 of 12	2 of 12
With this skill: would send	12 of 12	11 of 12

On their own, both agents explained numbers the notes couldn't explain and added owners, deadlines and recommendations nobody made. With the skill, the reviewer flagged invented statements in 0 of 24 decks, against 16 of 24 without it. The recipe puts the state and the ask on slide 1, shows every result against its target, says "we don't know yet" when that's true, and spends the slides on what missed or moved.

Install it with the same commands as the decision deck above; the skill picks the update recipe when you ask for a regular update. How it was tested, with every number.

Six skills for the documents you make

A complete fictional client brief: recommendation, tradeoffs, and the decision needed. Open the live example.

Start with the job you need done

I need to…	Start here	What the skill helps you do
Turn project notes into a weekly update	Project-management artifacts	Surface progress, blockers, risks, and the decision needed
Make a complex idea understandable	Visual explainer	Carry one concrete example through a meaningful visual
Help a client choose between options	Client decision brief	Compare costs and tradeoffs, then make a conditional recommendation
Turn research or results into a decision	Visual report	Connect evidence, uncertainty, and a useful next step
Give a talk or lesson, update leadership, agree a roadmap, or get a product decision approved	HTML presentation	Build a story across a paced, browser-readable deck; tested recipes turn your notes into a talk or lesson, a decision deck, a product update against goals, or a roadmap leadership can agree to
Show how a request moves through a system	Workflow diagram	Lay out a short main path with supporting and feedback paths

See what your notes can become

These examples use fictional inputs. They demonstrate possible results, not a required visual style. Follow each example to its input, exact request, and output.

From this…	To this…	Explore
Cohort counts and a stakeholder claim	A growth brief that checks the claim and compares the options	Report walkthrough
Six rules about seat reservations	A visual explanation of why two people can see the last seat	Explainer walkthrough
Pilot results and incomplete feedback	A five-slide case for a bounded next step	Presentation walkthrough

HTML presentations: a complete evidence slide, rather than just the title.

Workflow diagrams: named steps and labeled connections in a blueprint style.

Explore all six visual examples, including report charts, a worked timeline, and a weekly update. Styles are examples, not fixed templates.

Install

For Codex, run this from your project to install the visual explainer used in the complete starter example:

npx skills add arunai30/super-document-skills --agent codex --skill create-visual-explainer --copy

Or install all six:

npx skills add arunai30/super-document-skills --agent codex --skill '*' --copy

Review an existing copy before replacing it. Confirm the skill appears in your client's skill selector or discovery list before using it.

Using another agent? These are portable Agent Skills. Use your client's supported import flow to import a complete folder from skills, including its assets, references, and license. Client support varies; see manual installation and compatibility notes. A clone or ZIP download alone does not activate the skills.

Try it with everything supplied

Start with Why can two customers see the last seat? The sample includes the complete prompt and source material, a reference output, and a short checklist for reviewing your result.

For your own work, give the installed skill four things:

Use create-visual-report to turn the attached pilot notes into a decision brief for our operations lead. Help them decide whether to extend the pilot. Compare the options, show the evidence and unknowns, and use a warm editorial style. Create a local HTML file; do not publish it.

Your material · Your reader · The decision or takeaway · Your visual direction.

The default output is local HTML. Several skills can honor a requested different format when suitable tools are available; the presentation skill specifically creates HTML decks. Open the result in a browser and review its facts and appearance before sharing. See format and review guidance.

Why use a skill?

Each skill gives your agent a repeatable method for choosing the story, preserving the evidence, designing useful visuals, and checking the result. You can change the visual direction without having to rewrite that method for every document. The source notes and outputs make that method inspectable.

Want a shareable link?

After reviewing your document, you can optionally publish it with share/artifacts through an already-connected compatible MCP client. Publishing is a separate action you request; new publications default to owner-only access. You keep the local file whether or not you publish. Connection instructions · Publishing details

Documentation and contributing

• Usage, compatibility, and validation
• Diagram vision and supported scope
• Client decision brief guide
• Report a problem or request a document type

When reporting a result, include the skill, agent client, intended reader, and what was unclear. Use redacted or fictional source material when sharing an example. Maintainers can run node --test tests/*.test.mjs with Node.js 24 or newer.

License

MIT. Each skill includes its own license copy.