meet_vyas
Joined 1 year ago
Karma
1
Submissions
6
Member since
Jun 24
Recent submissions
Comment · 1 month ago
Yeah, my bad naming.
Inkling sounded nice and kind of resonated.
But yes, it is not related to the inkling that you pasted.
Comment · 1 month ago
There are examples in the README file in the repo.
Pasting them here as well:
--> Live tutorial — the interactive walkthrough that brought you here - https://meetvys.github.io/inkling/
--> Vision doc example — a project's vision rendered as an inkling artifact - https://meetvys.github.io/inkling/examples/vision/
--> inkling explainer — the skill explained as one of its own outputs - https://meetvys.github.io/inkling/examples/doc-explained/
Comment · 1 month ago
Actually I have added one comment to this post, however, it is not visible to anyone but me.
Comment · 1 month ago
Some documents aren't mostly words. An explainer, an HLD, a system design doc — they want visual elements: diagrams, charts, flow charts, things you can click into. HTML, CSS and JS are perfect for that, and always have been. The doc becomes much more than words.
Hand-writing it was never practical. Nobody codes a stylesheet and an interactive SVG for an internal design doc — the hours don't justify it.
Agents are very good at HTML, CSS and JS. That cost is gone. So why stay limited to a markdown file?
Inkling is a Claude Code skill built on that. You talk in plain English — "create a doc for the Q1 onboarding revamp, chart this CSV as a funnel" — and the agent designs and edits a real folder on your disk: index.html, style.css, runtime.js, data/. Turn by turn, as you think out loud. No slash commands, no templates.
Live tutorial, built with itself: https://meetvys.github.io/inkling/
How it works:
- The doc folder is the source of truth. There is no markdown file behind it.
- Every doc opens in a browser straight from disk. No server, no build step, no dependencies.
- The agent designs each doc instead of filling a template. It records what structure and behavior it invented on each turn, and fails itself if a render only reuses what already existed.
- `freeze` collapses a doc into one shareable .html file.
It is V0: each doc carries its own copy of the CSS and JS, so nothing compounds across docs yet, and it needs Claude Code installed.
But the thing I actually want opinions on is the premise, not the tool. HLDs, system design docs and explainers live in markdown files and Google Docs today — where a diagram is a pasted screenshot that goes stale, and a table is numbers someone retyped by hand. I think that is a mismatch between what those documents are and the format we keep putting them in.
So: does this class of document want to be an interactive HTML artifact? Or is markdown genuinely good enough, and I am solving a problem people do not have?