Your Claude plan, a brief, screenshots, and your brand

The AI pass can run in Claude Code on a Pro or Max plan, follows a wikipilot.md brief, puts real screenshots in the guides, and dresses the wiki in the product's own colours.

  • --engine claude-code runs the deep investigation through an installed, signed-in Claude Code, so it counts against a Pro or Max plan instead of API credits. It may only read the repo, write inside wiki/, and run git log plus wikipilot's own commands.
  • The default model is now Claude Opus 5.5 at high effort. --effort low|medium|high|xhigh|max trades depth for spend.
  • A wikipilot.md brief at the repo root sets the audience, the sections (it can replace the preset's list entirely), the site name, and the sites the agent may screenshot. Its claims are checked against the code before they reach a page.
  • Screenshots: the agent captures the pages it documents with the Playwright already in your repo, looks at each image, and embeds the useful ones. New wikipilot screenshot and wikipilot image commands do the same by hand.
  • Brand theme: colours and fonts are read from your Tailwind config or CSS and saved under theme, adjusted until links pass 4.5:1 contrast in both modes. Drop a logo.* or favicon.* into the wiki folder to use your own.
  • Mermaid diagrams are drawn in the wiki's brand colours, and every page opens with its title and description.
  • wikipilot build lists links that point at pages that don't exist.
  • Sync dates: each page shows the date of the commit it was last verified against, next to the commit itself.
  • Keys are read from .env.local as well as .env, and init exits with code 1 when the AI pass fails, so scripts can tell.

A new look for generated wikis

  • Redesigned wiki theme with a colour per section and fully clickable cards.

More than one AI provider

  • --provider runs the investigation on OpenAI, Gemini, or any OpenAI-compatible endpoint, as well as Claude.

Onboarding section

  • Technical wikis gain an onboarding section that takes a new developer from a clean checkout to a first change.
  • The investigation now has to cover the product story, user flows, and why each technology is there.

Cheaper investigations

  • The investigation conversation is prompt-cached, and every run prints what it cost.

The AI deep investigation

  • init can hand the draft to an AI agent that reads the code and rewrites the pages from what it finds.

The full wiki by default

  • init drafts every section unless you ask for a technical wiki or a user guide.

First release on npm

  • init, build, and serve, with the freshness contract in every page's frontmatter.