01 spec-to-wiki compiler
Make AI knowledge useful to humans.
AI knowledge made accessible to humans
AI-era projects accumulate knowledge fast — agent rules,
spec-driven development files, generated plans — all scattered
across folders and conventions built for tools, not teammates.
That knowledge is hard for people to find and understand.
One command, a real wiki
specwiki scans your project for that knowledge and generates a
navigable wiki from it: categorized pages, an index, and a
browsable HTML view with search — no server, no lock-in, straight
from the files already in your repo.
Everyone on the same page
Reviewers, stakeholders, and new teammates browse one shared wiki
— a clear, generated view of your project's specs and agent
knowledge, not another tour through folders built for machines.
quick start
npx @lucasviola/specwiki generate && npx @lucasviola/specwiki open
npm install -g @lucasviola/specwiki
Install specwiki in this repo: run `npx @lucasviola/specwiki generate && npx @lucasviola/specwiki open`,
then summarize the generated wiki index for me.
View source on GitHub
02 the problem
Every AI tool invents its own convention. Humans get none of it.
AGENTS.md, CLAUDE.md,
.cursor/rules/**, _bmad-output/**,
openspec/**, .kiro/specs/**,
.github/copilot-instructions.md — knowledge your agents
read on every run, but that people never browse. It rots invisibly
in dot-folders; reviewers and new teammates can't find it, and
nobody trusts what they can't see.
specwiki discovers all of it with one command and synthesizes a
categorized, searchable wiki. Out of the box it finds:
Spec sources specwiki discovers by default, and the file patterns
for each
| Source |
Patterns |
| Root agent files |
AGENTS.md, SPEC.md,
CLAUDE.md, GEMINI.md
|
| Cursor |
.cursor/rules/**,
.cursor/skills/**/SKILL.md
|
| Spec frameworks |
specs/**, openspec/**,
.kiro/specs/**
|
| Docs & plans |
docs/specs/**, docs/plans/**,
requirements/**
|
| GitHub Copilot |
.github/copilot-instructions.md |
| Monorepo agents |
**/AGENTS.md (nested packages) |
| BMAD output |
_bmad-output/**/*.md |
| Agent skills |
.agents/skills/**/SKILL.md |
| README files |
**/README.md (standalone wiki pages) |
| All markdown |
**/*.{md,mdc} (any .md /
.mdc in the project)
|
03 how it works
Discover, generate, open. That's the whole pipeline.
i
Discover
A fast-glob scan of known agent and spec conventions plus all
project markdown — ignoring node_modules,
dist, wiki, and friends.
ii
Generate
Categorized markdown pages, an index, and a browsable HTML view
with client-side search, highlighted code blocks, and
Wikimedia-inspired design tokens.
iii
Open
One command launches the wiki in your default browser.
Self-contained output — bundled CSS/JS, system fonts,
file://-safe. No server, no CDN, no lock-in.
iv
Agent-friendly out, too
Machine-readable --json output and an opt-in
--emit-llms-txt manifest, so your tooling can consume
the wiki as easily as your team does.
04 spec → wiki
Three root files in. A browsable wiki out.
examples/agent-harness-parcel
is a minimal agent harness — the three root files many AI tools read
first. Point specwiki at it (from the specwiki repo root):
before — scattered specs
agent-harness-parcel/
├── README.md
├── AGENTS.md
└── CLAUDE.md
one command
npx @lucasviola/specwiki generate --project examples/agent-harness-parcel
npx @lucasviola/specwiki open --project examples/agent-harness-parcel
after — a generated wiki
Main Page
Parcel Path (mock)
3 spec files indexed.
Project Root
-
Agent Instructions —
AGENTS.md
-
Claude Instructions —
CLAUDE.md
-
Readme —
README.md
Two more demos — a BMAD research project and long-form article notes —
live in the
examples folder on GitHub.
05 the fine print
Plain files. No magic.
what generate writes
wiki/
├── index.md
├── {slug}.md
├── llms.txt
└── html/
-
Node.js 20+ — a CLI on the runtime you already
have
- TypeScript — strict, tested, deterministic
-
Zero-config by default — sensible patterns out of
the box
- MIT licensed — read it, fork it, ship it
Get specwiki on GitHub