Changelog

Added

• Page-spec v2 vocabulary on pageTypeSpecSchema: audience, voice, expectedCta, requiredProofTypes, formFriction, defaultSections, forbidsArchetypes, validationRubric, guidance. All optional or defaulted — v1-shape literals still parse.
• Section archetype catalog with auto-discovery: Hero, ValueProps, SocialProof, FeatureDeepDive, Pricing, Faq, FinalCta, plus ComparisonTable, AwardBadgeRow, IconFeatureGrid, InlineLeadForm, IntegrationGrid synthesized from the swipe-file corpus. Filesystem-driven — drop a .ts under archetypes/ and restart.
• Best-practice page specs (product, content, service, demo) authored from 89 per-page analyses across 5 categories of a 110-page landing-page swipe file.
• Page-spec auto-discovery: drop a spec file under page-specs/specs/ and it's picked up at boot via readdirSync + dynamic import + top-level await.
• SpecRepository interface + InMemorySpecRepository as migration insurance for an eventual code → DB swap.
• Admin browse endpoints under /admin/api/{page-specs,section-archetypes,block-specs} with cross-reference counts and rendered-prompt previews (Zone 2 for blocks, Zone 3 for specs), gated by requireSession.
• Admin SPA catalog pages (Page Specs / Section Archetypes / Block Specs + detail views) with cross-linking, debounced search, and source filters. Sidebar nav extended.
• New CSS patterns for detail pages: .card, .kv, .breadcrumb, .prose-block, .block-id-list, .chip, .badge-custom/.badge-native, .empty-state, .id-cell, .num-cell.
• Page-type research pipeline tooling under page-type-data/: extract-from-coda.js (DevTools snippet), fetch-images.mjs (parallel image fetcher with native-resolution URL cleanup), analyze-screenshot.md (per-screenshot analysis prompt), synthesize-best-practice.md (per-category synthesis prompt), README documenting the 5-step pipeline.
• Prompt-rendering snapshot tests (prompts.snapshot.test.ts) covering all shipped specs.
• pnpm dump-prompt CLI: dump the full assembled system prompt for a (spec, stage) pair to stdout, with optional fixture file for Zone 4. Useful for offline iteration without burning API credits.
• Inline archetype definitions in Zone 3 of the system prompt for LLM grounding.
• Universal page rules + canonical propose_plan example baked into BASE_PROMPT.
• purchase and start-trial CtaAction enum values.

Changed

• Existing comparison spec upgraded based on best-practice synthesis (13-sample corpus): ComparisonTable added as required default section, Faq moved from required to forbidsArchetypes, Pricing added to forbidsArchetypes, SocialProof split into trust-logo-belt + switcher-testimonial slots, FeatureDeepDive ×1 required + 1 optional added, requiredProofTypes expanded with logos and metric, validation rubric expanded from 5 to 9 testable assertions.
• book-demo, comparison, resources specs re-authored with full v2 vocabulary content.
• data-index.js renamed to data-index.json to reflect its actual content (pure JSON) and avoid ESLint parsing it as code.

Fixed

• Entrypoint detection (isEntrypoint) now uses pathToFileURL so the boot block runs correctly on paths containing spaces (e.g. /Users/.../Local Sites/...); previously the check silently failed off-Railway.
• pnpm dump-prompt script imports updated for the post-restructure monorepo layout (../plugins/page-ai/src/... for runtime modules, @airfleet/nexus-ai-server-wp-contracts for types).

Fixed

• Build: Root pnpm run build now runs build:packages before apps/server, compiling platform-sdk, platform, wp-contracts, and page-ai to dist/ and pointing each package exports default entry at ./dist/index.js. Production previously only emitted apps/server/dist; Node still resolved workspace imports to TypeScript sources and failed at runtime (ERR_MODULE_NOT_FOUND for *.js siblings), breaking Railway health checks after the pnpm workspace migration.
• Deploy: predb:migrate invokes build:packages so the migration script always resolves @airfleet/nexus-ai-server-platform from emitted dist output; pnpm test does the same so Vitest matches production resolution.

Fixed

• Deploy: Railway startCommand aligned with the pnpm workspace so production runs apps/server via pnpm start (avoids broken health checks after the composition root moved).
• CI: GitHub Actions fixes for pull-request / release validation paths.

Changed

• Docs: Deployment and release workflow documentation updates.

Added

• Initial Airfleet Nexus AI Server workspace with pnpm, TypeScript, ESLint, Vitest, and shared build, test, typecheck, migration, and spec-generation scripts.
• Express server composition root with health checks, admin UI serving, plugin route mounting, session handling, access logging, and centralized error handling.
• Page AI plugin with API-key authenticated WordPress endpoints for page spec lookup, session lifecycle management, staged page-build orchestration, block-tree hydration, and implementation revisions.
• Admin capabilities for GitHub OAuth login, API key creation/listing/revocation, log search, session lists, and session journey inspection.
• Shared platform packages for configuration, PostgreSQL access, logging, event bus, middleware, plugin contracts, request context, and WordPress-facing Zod schemas/types.
• AI generation foundation using Anthropic Claude through the Vercel AI SDK, including stage-scoped tool registration, turn budgeting, and research/plan/approve/implement workflow support.
• PostgreSQL-backed observability with request logs and domain/admin event subscribers.
• Admin SPA package and plugin-aware route/navigation structure for server-hosted administration.
• Local development and release infrastructure, including Docker Compose for PostgreSQL and GitHub Actions workflows for release validation and GitHub releases.
• Initial automated test coverage for WordPress block-tree contracts, tree operations, stage transitions, tool registry behavior, turn budgeting, prompt handling, and log subscribers.