Build and Maintain Structured Research Wikis

Wiki Builder creates and maintains configurable research wikis with source provenance and local Markdown output. It is an official skill from DAIR.AI's academy plugins under the MIT license. Each wiki is a standalone folder with its own sources, compiled pages, derived artifacts, prompts and local configuration. By default wikis live under ~/dair-wikis/ (the WIKI_ROOT environment variable or the --root flag of the init script changes that), in a lowercase kebab-case folder such as agent-memory, ai-evals, open-source-models or company-research.

The skill is intentionally general. Every wiki has a wiki.config.md that defines its purpose, audience, page types, style rules and update workflow, and that file beats any generic default. A new wiki starts with this layout: wiki.config.md, raw/ for source material, wiki/ with an index.md, derived/, prompts/ (compile-index, compile-source-page, compile-concept-page, query-and-file and lint-wiki prompts), logs/maintenance-log.md and sources.md. More folders are added only when the config needs them, for example wiki/papers, wiki/concepts, wiki/people, wiki/products, wiki/organizations, wiki/timelines, wiki/questions, wiki/maps and assets.

The operating workflow has five steps:

1. Resolve the task: decide whether the user wants to start, ingest, compile, query, restructure, lint or export, and inspect an existing wiki's config first.
2. Use the local config: read wiki.config.md, sources.md when provenance matters, and any custom prompts.
3. Preserve provenance: never turn loose claims into wiki facts without a source. Each sources.md entry records at least a title, a path or URL, the date added and a short note on what it contributes.
4. Compile pages: prefer durable pages over one-off summaries, each with a concise overview, source-grounded key points, links to related pages, open questions or uncertainty, and update notes.
5. Maintain the wiki: when many pages change, update wiki/index.md, relevant maps and the maintenance log, and update wiki.config.md first if the purpose or structure changes.

Seven default flavors are supported, each with suggested page structures: research, paper (individual papers, concepts, comparisons and research gaps, with problem, method, results, limitations and implementation notes), domain (landscape, timelines, glossary), product (features, use cases, competitors, separating documented from observed behavior), person (work, themes, timeline, with no unsupported biographical claims), organization (projects, people, timeline, strategy, separating facts from interpretation) and project (decisions, specs, notes, making the current state obvious to the next agent). Use research when unsure.

Use it when the user asks to start a wiki or knowledge base for research notes, papers, products, people, organizations, domains, projects or events; ingest source material into an existing wiki; generate wiki pages, source pages, concept pages, maps, timelines, briefs or indexes; query a wiki and file the answer back into it; refactor a wiki's structure or requirements; or maintain provenance records, notes on each item and update logs.

Do not use it to hard-code every wiki into one fixed structure. It requires whatever upstream tool or local setup the workflow names, and does not authorize destructive, production, paid or external-message actions without explicit approval. Validate generated artifacts against the user's real sources before treating them as final.

Scaffolding uses the bundled script, resolved through the plugin install location:

bash "${CLAUDE_PLUGIN_ROOT}/skills/wiki-builder/scripts/init_wiki.sh" <slug> --title "Readable Title" --flavor research

After scaffolding, edit wiki.config.md to match the real goal, put copied or downloaded sources in raw/, record provenance in sources.md, generate pages under wiki/ and log major maintenance actions. Outputs are a navigable set of linked Markdown pages, an index, maps and a maintenance log. The quality bar is to make the first page useful immediately, use explicit filenames and stable slugs, separate raw source material from compiled interpretation, link related pages, mark speculation and unknowns, avoid repeating the same source summary in many places, and keep pages navigable for future agents and humans.

It runs in Claude Code, Codex CLI and Cursor, and relies on a shell script plus plain local Markdown files, so no external service is required.

It suits researchers, analysts and agent-assisted teams who want a persistent, source-grounded knowledge base that future sessions and readers can navigate and extend.