TheoRata/Report-Skill

4 stars · Last commit 2026-06-11

A Claude Code skill that turns long-form agent output into a designed HTML report — TOC, callouts, framed figures, footnotes, light/dark themes, and one-click export back to Markdown.

README preview

# Report Skill

> A [Claude Code](https://claude.com/claude-code) skill that turns long-form agent output — research, deep-dives, postmortems, analyses, write-ups — into a **designed HTML document** instead of another flat `.md` file.

The agent writes plain Markdown. `render.mjs` does everything else: serif typography, light/dark theme, sticky table of contents, syntax-highlighted code, callouts, framed figures with click-to-zoom, hover-preview footnotes, in-browser comments, and one-click export back to source Markdown.

Token spend is identical to writing a plain `.md`. The HTML is generated mechanically.

![Designed reports, in-document review, light & dark themes](docs/screenshots/00-demo-strip.png)

**See an example:** **[report-skill.vercel.app](https://report-skill.vercel.app/)** — a fully rendered report, theme-switchable, openable right now.

The renderer is plain Node and runs against any Markdown — whatever agent (or human) wrote it. The Claude Code plugin is the optional auto-trigger surface; rendering itself is editor- and agent-agnostic. And because review threads are persisted inside the embedded source, an agent can read its own comment feedback and revise the report against it.

---

## The moment this matters

You asked Claude to *"research the failure modes of LLM-generated frontend code"*, or *"postmortem the outage from last Thursday"*, or *"document our auth migration so the team can review it"*. Three minutes later you have **3,000 words of genuinely good output** sitting in a `.md` file. And then what?

View full repository on GitHub →