human-htmllisted
Install: claude install-skill rhnfzl/human-html
# human-html
## What this is, in plain terms
Picture someone opening one of your documents for the first time: a person who will act on it and did not write it. They have ten minutes. They want to know what the work is, whether the plan is sound, where the risk sits, and what they need to do next.
If the document is a long Markdown file, they skim. Skim turns into rubber-stamp. They miss the assumption that was buried in paragraph nine. They approve something they didn't really read.
If the document is a single HTML page with a plain-language summary at the top, a diagram in every before/after section, color-coded risks, a checklist, and the key snippets inline, they read. They ask sharper questions. They catch the thing that was wrong. They redirect work before it lands in the wrong place.
This skill enforces that switch as a workspace contract. When an agent (you, Claude, Codex, or anyone else) produces a *human review surface* (a plan, a code review, an architecture explainer, an understanding doc, a research synthesis, a decision aid, a prototype, a status report, or an incident postmortem) the artifact lands as HTML under `docs/human-html/` of the active workspace. Not as Markdown. Markdown is still the right format for scratch notes, ticket drafts, durable references, and meeting transcripts; it is the agent's memory layer. HTML is the human's review layer. The split is the point.
The skill enforces two layers of contract: a **file contract** (naming, metadata, al