Nuzo 0.9.0

Added

• The CLI now supports local store recovery commands:
  nuzo memory integrity, nuzo memory backup --path <file>, and
  nuzo memory restore <file> --yes, including WAL-safe SQLite online backup
  validation before restore.
• nuzo memory doctor --json now exposes the same content-free store,
  integrity, Git tracking, warning, and status diagnostics in a stable
  machine-readable shape.
• The Nuzo CLI now includes explicit host bootstrap commands:
  nuzo setup, nuzo host install codex, nuzo host install claude-code, and
  nuzo host install --all, with dry-run, JSON, and non-interactive
  confirmation support.
• nuzo update and targeted nuzo host update commands now update only
  already-installed Codex and Claude Code plugins, preserve the Claude Code
  install scope, and avoid repeating first-time setup.
• The shared runtime resolver now discovers ancestor project configuration,
  resolves canonical project roots, enforces trusted host authorization, and
  reports non-sensitive configuration provenance.
• Repository marketplaces now expose the tracked Nuzo plugin as
  nuzo@nuzo-memory for Codex and Claude Code without a source checkout or a
  redundant global runtime install.
• CI now validates native marketplace discovery and installation with pinned
  Codex and Claude Code CLIs on Node.js 24.
• MCP and host hook doctor diagnostics now include the same content-free
  SQLite integrity summary as the CLI.
• Deterministic public-documentation contracts now guard release versions,
  version-gated commands, MCP tool names and count, supported Node.js lines,
  npm package guidance, and user-versus-maintainer navigation in CI and Pages.
• Supply-chain validation now requires GitHub Actions workflow uses: entries
  to be pinned to reviewed commit SHAs and release npm installation to use an
  exact reviewed version.
• Documentation CI now validates Markdown links, anchors, bounded external
  links, shell snippet syntax, setup/update command drift, and core public
  terminology across user-facing entry points.
• CI now validates staged npm and generated plugin artifacts across the
  supported Linux x64, macOS x64, and Windows x64 runtime matrix on Node.js 22
  LTS and 24 LTS.

Changed

• nuzo memory doctor now reports SQLite store integrity status and surfaces
  integrity failures as warnings.
• Host setup output now explains why each command runs and shows explicit
  non-interactive examples for Codex, Claude Code, and both hosts.
• Published MCP and lifecycle-hook paths now default to the active project and
  user scopes in restricted mode, while the local CLI remains an explicit
  administrator workflow.
• Public GitHub, npm, getting-started, CLI, and host documentation now define
  setup as a one-time operation and managed update as the routine upgrade path
  planned for 0.9.0.
• Tracked host plugin sources now resolve the exact matching
  @nuzo/memory runtime and remain version-aligned through release tooling.
• README, homepage, Getting Started, clean install, CLI, and host plugin pages
  now lead with tested user installation and cross-session verification paths.
• Current 0.8.1 onboarding now excludes host-bootstrap and recovery commands
  until those capabilities are published, while the 0.9.0 release gate
  requires the new setup path to return to GitHub, Pages, and npm documentation.
• @nuzo/memory-cli and @nuzo/mcp-server now identify 0.9.0 as their final
  planned release before npm deprecation in favor of @nuzo/memory.
• npm staging, publish-target checks, and publish automation now derive from a
  shared package policy: 0.9.0 still includes the two legacy transition
  packages, while later versions publish only @nuzo/memory-core and
  @nuzo/memory and reject retired legacy staging.
• Benchmark fixtures now distinguish portable JSON exports from operational
  SQLite backups.
• MCP memory.list, memory.history, and memory.export responses are now
  bounded with explicit limit, next_cursor, and truncated pagination
  fields. The local CLI keeps full export as an explicit administrator action.

Fixed

• Semantic model inspection now verifies pinned file digests and rejects
  missing, truncated, altered, unreadable, or symlinked model files before
  reporting ready or skipping provisioning repair.