Skip to content
Back to skills

Wiki Viewer

ASecurity

Render a self-contained HTML viewer for a pro-workflow wiki. Pages, sources, claims, seed queue, page-link graph and full-text search all in one file. No external dependencies, no JS framework, S3-uploadable. Use when the user wants to browse a wiki visually, share its current state with someone, audit research progress, or hand off a knowledge base. Inspired by Thariq Shihipar's "Unreasonable Effectiveness of HTML" — favors information density and shareability over markdown-only outputs.

  • 2,892 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added May 28, 2026
ai-agentsbashsqlnodeaws

Works with

  • claude code

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned May 28, 2026

npx -y skills add rohitg00/pro-workflow --skill wiki-viewer --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Wiki Viewer?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Wiki Viewer
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/rohitg00-wiki-viewer/badge)](https://www.skillsdirectory.com/skills/rohitg00-wiki-viewer)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: wiki-viewer
description: Render a self-contained HTML viewer for a pro-workflow wiki. Pages, sources, claims, seed queue, page-link graph and full-text search all in one file. No external dependencies, no JS framework, S3-uploadable. Use when the user wants to browse a wiki visually, share its current state with someone, audit research progress, or hand off a knowledge base. Inspired by Thariq Shihipar's "Unreasonable Effectiveness of HTML" — favors information density and shareability over markdown-only outputs.
---

# Wiki Viewer

Single-file HTML view of a pro-workflow wiki. Reads `~/.pro-workflow/data.db`, dumps the wiki into one self-contained HTML document with in-browser search, link graph and a seed-queue panel.

## Why HTML, not markdown

- **Information density** — pages, sources, claims, seeds, link graph in one viewport
- **Visual clarity** — long wikis are unreadable as raw markdown; HTML scales
- **Shareability** — upload to S3, send the URL; recipient does not need pro-workflow installed
- **Two-way interaction** — "copy as seed" buttons turn open questions into seed queue prompts you can paste back into Claude Code
- **Auditable** — quick visual proof that the auto-research loop produced something useful

## When to use

- After a `/wiki research` run, to see what it built
- Before sharing a wiki with a teammate or with leadership
- Code review: render a `codebase`-flavored wiki for an unfamiliar module
- Incident review: render an `incident` wiki for a post-mortem readout
- Periodic audits: stale-claim detection, orphan-page review

## Commands

```
node $SKILL_ROOT/scripts/render.js <slug> [--out <path>] [--theme dark|light]
```

Defaults:

- output: `<wiki-root>/derived/viewer.html`
- theme: `dark`

## What ships in the file

| Panel | Contents |
|-------|----------|
| Header | wiki slug, flavor, scope, root path, last-update timestamp, page count, source count, kill-switch status |
| Sidebar | page list grouped by `page_type`, in-page filter input |
| Main | selected-page detail: title, summary, full markdown content (rendered), inline citations resolve to source rows |
| Sources | table of every `wiki_sources` row + manual `sources.md` rows |
| Seeds | seed-queue table grouped by status; "copy as research prompt" button per pending seed |
| Link graph | SVG force-layout of cross-page citations + back-links |
| Search | in-browser substring + token search over title/summary/content |
| Footer | meta: schema versions, embedding model if present, generator version |

## Self-contained

No CDN, no external fonts, no `<script src=>`. Inline CSS, inline SVG, inline JS only. Result is a single `.html` file that opens locally or from any static host.

## Compose with the rest

```bash
# Generate after auto-research run completes
/wiki research agent-memory --max-pages 5
node skills/wiki-viewer/scripts/render.js agent-memory
open ~/.pro-workflow/wikis/agent-memory/derived/viewer.html

# Hand off to a teammate
aws s3 cp ~/.pro-workflow/wikis/agent-memory/derived/viewer.html s3://my-bucket/agent-memory.html --acl public-read
```

## Design principles

1. **Type-first** — every panel reads as text; visualizations are auxiliary, not load-bearing.
2. **Zero decoration** — no gradients, no glow, no atmospheric backgrounds.
3. **Color is meaning** — Anthropic coral marks the active page and CTAs only.
4. **Print-friendly** — `@media print` collapses sidebars so the markdown content prints clean.

## Limits (initial release)

- Markdown rendering is a small in-file parser (headings, lists, code blocks, links, blockquotes, tables, footnote citations). No HTML-in-markdown.
- Link graph is precomputed and serialized as SVG; no live re-layout.
- Search is substring + tokenized AND match. BM25 stays in SQLite; the viewer is a snapshot.
- Re-render after each batch of changes. The file is not live.

## Future hooks (not in initial release)

- `--include-council` to bundle every linked council transcript inline
- `--include-survey` to bundle generated surveys
- "Copy as council prompt" buttons next to claims tagged `contested`
- Diff view: `--against <previous.html>` to highlight new claims since a prior render

Files in this skill

  • SKILL.md4.1 KB
  • scripts/render.js25.3 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.