Leadtype
Changelog

Leadtype 0.3

Leadtype 0.3 adds browser-side WebMCP docs tools, plus config-driven RSS and Atom feed generation for changelogs, release notes, blogs, news, and any generated docs content selected by canonical URL prefix. It also hardens docs search for large corpora with prototype-safe indexing and linear query scaling, validated by a new benchmark against real TanStack-scale docs.

Browser WebMCP

  • Added leadtype/webmcp for registering generated docs as browser-side WebMCP tools with document.modelContext or navigator.modelContext.
  • Added default search-docs and get-page tools over the generated search index and Markdown mirror, with bounded search limits, page-path validation, and default known-page checks.
  • Added React, Vue, and Svelte lifecycle helpers so host apps can register and clean up docs tools from framework-owned client entry points.
  • Added leadtype init --webmcp scaffolding for framework examples.
  • Documented the distinction between browser WebMCP and the server-side leadtype/mcp endpoint.

Feeds

  • Added a feeds config field for docs.config.ts and leadtype.config.ts.
  • Added leadtype/feed with generateFeedArtifacts(), renderRssFeed(), and renderAtomFeed() for custom pipelines.
  • Added site-mode generation for configured feeds, with generated paths reported in JSON output under files.feeds.<id>.
  • Added RSS 2.0 and Atom output, including XML escaping, control-character stripping, stable Atom entry IDs, absolute canonical links, feed-level author metadata, and newest-first entry sorting.
  • Added support for mounted changelog URLs such as /changelog, so feed entries do not have to live under /docs.

Date handling

  • Feed entries use stable frontmatter dates. Changelogs and release notes should set an explicit date because the publication date is part of the release record.
  • Pages where the feed should represent source edits can use Git-derived lastModified; leadtype generate writes it automatically when git metadata is available.
  • Generation fails when a selected feed page has no date or lastModified frontmatter, so feed order does not churn based on CI-generated file mtimes.
  • The default docs lint schema now accepts optional date frontmatter, including YAML dates parsed as Date values.

CLI defaults and deprecations

  • leadtype generate now enriches generated markdown with Git-derived lastModified and lastAuthor by default. Enrichment is best-effort: missing .git metadata, shallow history, untracked files, or a missing git binary skip those fields without failing generation.
  • Git-derived lastAuthor now skips bot and automation commits, falling back to the latest human author for the page while keeping lastModified tied to the latest commit. Projects can add repo-specific automation names with git.ignoredAuthors in docs.config.ts.
  • Package-mode MCP artifacts are inferred from agents.mcp.enabled in docs.config.ts, so leadtype generate --bundle is enough for packages that configure MCP support.
  • Deprecated shortcut flags that duplicate config or defaults: leadtype generate --mcp, leadtype generate --enrich-git, and leadtype init --webmcp. They still work for compatibility and print warnings; new docs and scripts should use docs.config.ts or explicit app code instead.

Search reliability and scaling

  • Fixed a crash when indexed content or a search query contains a term that collides with an Object.prototype member, such as constructor. The term postings record is now built with a null prototype and query-time lookups, including synonym expansion, use Object.hasOwn guards. Previously such a term crashed both createDocsSearchIndex at build time and searchDocs in the browser.
  • Made searchDocs latency scale linearly with corpus size. Chunk lookups now use a cached map instead of a per-result linear scan, and excerpts are built only for results that survive the limit. Ranking is unchanged. On a 20,000-chunk corpus this cuts p95 query latency from roughly 665 ms to 186 ms; at the scale of the combined TanStack Router, Start, and Query docs (about 5,000 chunks), median queries run in about 13 ms.
  • Added a search scalability benchmark (pipeline:search-bench in the TanStack example) that measures index build time, artifact sizes, parse cost, memory, and query latency across real and synthetically grown corpora.

Validation and safety

  • Agent Skills discovery manifests now emit the v0.2.0 format: the $schema URL plus per-entry type: "skill-md", a relative url, and a sha256: hex digest, so spec-conformant clients (such as Vercel's skills tool) resolve every entry. The pre-0.2.0 leadtype path/integrity fields are still emitted alongside for older consumers, and skill descriptions over the spec's 1024-character cap now fail generation instead of being silently dropped by clients.
  • Organization JSON-LD now supports email, sameAs, contactPoint, and address fields from docs.config.ts, so hosted docs can publish direct email contact details, brand-disambiguation links, and Schema.org ContactPoint/PostalAddress data. These objects are emitted verbatim into JSON-LD, so config validation rejects unknown contactPoint/address keys (catching typos like telphone) and empty address objects.
  • Feed generation requires --base-url or a deployment URL environment variable so RSS and Atom item links are absolute.
  • Search generation now treats shared/ and _shared/ route segments as internal by default, so template content does not appear as duplicate public search results or answer citations unless a page explicitly sets search: true.
  • Feed output paths must start with /, end with .xml, resolve inside the output directory, and be unique across all configured feeds.
  • Pages with draft: true are excluded from feeds when that field reaches generated markdown.
  • Locale-prefixed generated pages are excluded from root feeds so localized mirrors do not duplicate entries.

Crawler artifacts

  • Site-mode generation now writes robots.txt, sitemap.xml, and sitemap.md at the output root only. Nested /docs sitemap and robots copies are no longer emitted because crawlers treat the origin-root robots.txt as the authoritative policy file.

Agent artifacts without a docs tree

  • Added generateAgentArtifacts() to leadtype/llm for sites without .mdx docs — CMS-backed blogs, marketing sites, and data-driven pages. It takes an in-memory page list and emits llms.txt (plus the /.well-known copy), per-page Markdown mirrors at ${urlPath}.md with canonical_url and last_updated frontmatter, sitemap.xml, sitemap.md, robots.txt with Content-Signals, and a root-level agent-readability manifest.
  • Page inputs fall back from explicit fields to frontmatter inside content to the URL slug; duplicate, relative, and query-bearing routes fail loudly before anything is written.
  • emitRootCrawlerFiles: false supports microfrontend fragments whose host app owns the origin-level crawler files and merges fragment manifests.
  • Root URL mounts (urlPrefix: "/") now resolve correctly instead of emitting //page paths.

Docs reorganization

  • The docs are now organized by capability: Docs Pipeline, AEO & Agent Readability, Writing for Agents, Search & AI Answers, Package Docs, and Integrations. Because navigation drives llms.txt and sitemap.md, the new structure is also the agent-facing index.
  • Added an AEO overview page mapping every agent artifact to the Vercel agent-readability spec and the ora scoring rubric, with an audit workflow.
  • Page URLs moved (/docs/build/* and /docs/sources/* → /docs/pipeline|aeo|integrations/*, /docs/authoring/* → /docs/writing/*); leadtype.dev serves 301 redirects from every old path and its .md mirror.

Dogfooding

  • The Leadtype docs now publish /changelog/rss.xml and /changelog/atom.xml from the real changelog.
  • The Next.js, Astro, Nuxt, SvelteKit, and TanStack examples now generate the changelog feeds. Next.js, Nuxt, SvelteKit, and TanStack smoke-test the feed endpoints.
  • The TanStack custom pipeline calls generateFeedArtifacts() directly so the lower-level API stays covered outside the CLI.
  • Added a task-focused guide for feed setup, feed date policy, output validation, and verification steps.