ai-delivery plugin

Self-contained host plugin for @aviaratech/ai-delivery. Its built launcher starts the package's existing mcp:serve command in the adjacent runtime directory, which includes the built CLI and runtime dependencies. It can start from a native plugin root without a parent npm installation, source build or postinstall. The root contains portable plugin.json / mcp.json and the preserved Claude-compatible .claude-plugin/plugin.json / .mcp.json. Package, CLI, MCP server and native manifest versions agree; application and bundled dependency notices accompany the launcher. MCP schemas, GitHub operations, repository routing and delivery logic retain their package owners in the package.

The bundled skills cover issue intake, host-owned worktree development, and pull request handoff. Use ai-delivery plugin install --host codex --scope user --version <published-version> or select --host claude-code with a native user, project, or local scope. The same command group provides doctor, explicit-version update, rollback to the recorded previous version, and remove, with --dry-run and --json on each. The package guide describes managed version retention, read-only diagnostics, and data preservation. Restart the host after changing its plugin selection.

Configure operator-owned JSON at ~/.config/aviaratech-ai/ai-delivery.json and set its author/reviewer credential environment references outside source control. Every remote tool accepts an explicit repo and works without a checkout or repository policy files; one launched MCP server can switch repositories. The launcher needs no AI_DELIVERY_IDENTITY: command roles default from user settings. issue_pr_create with dryRun: true reports configured actors and rule visibility. Hosts own worktrees, contributor checks, commits and pushes. Explicit runtime:stage / runtime:admit remain separate controller operations; installing the plugin does not activate or alter a private runtime or adopt historical custody.