Ivan Kuznetsov releases llm-wiki, a one-command tool to build and search AI-maintained project wikis for Claude and Codex

What happened: Ivan Kuznetsov packaged llm-wiki as an installable plugin for Claude Code and Codex (GPT-5.5). It automates the creation of project-local wiki folders with architecture docs, decision records, and open questions — then makes those wikis searchable before coding begins.

How it works: The tool runs three workflows: bootstrap-wiki (creates the knowledge base), wiki-researcher (searches your project and cross-project master wikis before planning), and wiki-plan (grounds implementation plans in past decisions). It uses QMD semantic search when available, falling back to ripgrep text search if not. Instead of starting from scratch each time, you get a 'Past Knowledge' section that recalls what your team already decided.

So what: Engineers stop rediscovering the same architectural patterns, dependency choices, and gotchas across projects. When you run /llm-wiki:wiki-researcher before refactoring an auth flow or adding billing features, the AI reads what went wrong last time and what gaps remain — cutting the planning-to-implementation cycle for developers working across multiple codebases.