wiki-spaces
A minimal nestable wiki, which is a folder with index.md and a ## Spaces navigation contract, for any use case. Research, recipes, code notes, writing, team docs, a personal life wiki, your shape, your call.
Markdown flavor is Obsidian, including wikilinks, frontmatter, callouts, embeds, comments, and Bases. This is the single dialect across the spec and the skills. Non-Obsidian renderers like GitHub preview, vanilla VS Code, or plain markdown viewers display the content but will not render Obsidian-specific syntax like provenance comments, embeds, or .base files the way Obsidian does. View your wiki in Obsidian for full fidelity.
Audience
wiki-spaces is built for AI coding harnesses with filesystem access, such as Claude Code, Codex, Cursor, Copilot, Gemini CLI, OpenCode, and Kiro. Browser-only AI assistants like ChatGPT in a tab or the Claude.ai web interface are out of scope. They cannot reach the filesystem to read or write your wiki. If your AI lives in a browser, this isn't the tool.
Detached wikis are first-class. Any folder anywhere can be a wiki, including a company repository root. Wikis relate to each other via mounts, which can be symlinks, git submodules, or clones.
Install
Two tiers, one rule: pick one channel per harness. Plugin-installed skills are namespaced (wiki-spaces:ws-search), so installing both tiers into the same harness produces confusing near-duplicates rather than an error.
Minimal — the three skills
Install the reference skills directly into your AI coding harness (reaches 72+ harnesses via vercel-labs/skills):
npx skills add anfreire/wiki-spaces
This command installs the core skills: ws-search, ws-update, and ws-tend. Setup is driven by the installed skills themselves, which guide you through a short interview to configure your workspace.
Complete — skills plus hooks
One plugin per harness, bundling the same three skills with two deterministic hooks. At session start, an orientation line tells the agent when your project has a space in your wiki, or when the current directory sits inside one — and stays silent otherwise. On Claude Code and OpenCode, an after-turn footnote additionally reminds the agent that ws-update can capture durable work; Codex and Copilot ship orientation only (their stop hooks can't inject context yet).
| Harness | Install |
|---|
| Claude Code | /plugin marketplace add anfreire/wiki-spaces, then /plugin install wiki-spaces@wiki-spaces |
| Codex | codex plugin marketplace add anfreire/wiki-spaces, then codex plugin add wiki-spaces@wiki-spaces |
| GitHub Copilot CLI (≥1.0.22) | copilot plugin install anfreire/wiki-spaces |
| OpenCode (≥1.16) | add "plugin": ["github:anfreire/wiki-spaces"] to your opencode.json |
The hooks are sensors, not policers: they compute deterministically, inject at most a one-liner, and a session with no wiki anywhere costs zero extra tokens.
Each plugin bundles only the three ws-* skills. For full Obsidian syntax depth (callouts, embeds, Bases), add the companion skills below — they are a separate install on every channel.
Companion skills
For Obsidian-specific formatting and database views, install the companion skills:
npx skills add kepano/obsidian-skills
These companion skills add support for Obsidian markdown and Obsidian Bases.
No Tooling Path
You don't need any tools to start. Run this command to create a wiki:
mkdir -p ~/Documents/Wiki && printf '# My Wiki\n\n## Spaces\n\n' > ~/Documents/Wiki/index.md
A folder with index.md and a ## Spaces heading is already a complete wiki. The entire specification is defined in a single page, AGENTS.md. Your installed skills will discover the wiki automatically from your current working directory.
Bundled Script
Each skill bundles a single script, scripts/ws.py, which handles space traversal, size checks, and audit or repair tasks. It uses only the Python standard library, meaning it has zero dependencies beyond a standard python3 installation.
Skills
The core package provides three reference skills for your AI agent:
ws-search: Find content across your spaces.
ws-update: Capture, save, and sync notes with byte-cap size discipline.
ws-tend: Audit the wiki structure, normalize tags, and manage cross-links.
Search at Scale
Using grep or ripgrep works well for personal or team wikis up to a few hundred pages. For larger vaults, we recommend installing qmd. This is the markdown-aware search backend that Andrej Karpathy references in his LLM-wiki gist.
Learn More
Dependencies