Provider framework and unified provider UX #134

Goal

Shared product/UI infrastructure so every provider family looks and behaves consistently:

• provider discovery
• installed / not installed state
• healthy / unavailable state
• Native vs external
• selection and persisted choice
• configuration entry points
• optional install flow
• previews where relevant
• consistent /settings presentation
• graceful fallback
• provider-specific capability reporting

Why

Prompt providers (NMSh Native, Starship, Powerlevel10k: #58, #67) already do this ad hoc. Welcome, Suggestions, Completions, History and Pickers are all gaining external options. Without shared infrastructure, each family would reinvent status, fallback and settings UI differently.

Product principle

Excellent Native implementation + optional external providers. Native stays the default in most categories. External tools enhance NMSh and never become required dependencies.

Architectural constraints

• No universal Provider interface with dozens of optional methods. Each family has its own TypeScript interface for its runtime job: prompt render, welcome render, suggestion query, completion candidates, history search, picker selection.
• The shared layer covers metadata, status and UI only: id, family, display name, Native/external, detection (executable, version), health, family-specific capability flags, install reference (Optional shell-tool discovery and first-run setup #9), configuration link (Tool configuration center: read/edit supported tool config inside /settings #83), preview hook.
• Detection is cheap, cached and async. It never blocks startup or keystrokes.
• A missing binary, timeout, crash or bad output falls back to Native (or None) with a factual notice. It never breaks the editor or shell.
• External invocation uses argv (no shell interpolation), timeouts and bounded output.

Foundation checkpoint (this issue)

• Shared provider descriptor/status model and registry.
• Shared provider list/gallery component on the panel shell (Shared panel shell and temporary top boundary #90): status badges, preview area, persisted choice, fallback notice. Extracted from /prompt provider selection.
• /settings (Central /settings root browser #86) presents each family's providers the same way.
• Prompt providers migrate onto the shared descriptor with no visible change.

Provider families

• Welcome — Welcome providers: Vespyr, Fastfetch, Neofetch and None #135
• Suggestions — Suggestion providers and predictive ghost text #137
• Completions — Rich completion and CLI IntelliSense system #140
• History / search — Rich history search and metadata #143
• Pickers — Picker abstraction and external fuzzy finders #145
• Prompt — Native, Starship, Powerlevel10k already exist. Oh My Posh remains a research candidate only, not a committed provider.

Non-goals

• A plugin marketplace or third-party plugin API (Optional shell-tool discovery and first-run setup #9 explicitly excludes a marketplace).
• Automatic installation of any provider.

Acceptance criteria

• Prompt provider UI runs on the shared components without visual regression.
• A missing or broken external provider shows its status and falls back cleanly.
• Family runtime interfaces stay separate; shared code covers metadata, status and UI only.

Dependencies / related

• Next after Transcript presenter: centralize command/output row presentation #122 on the main implementation path.
• Optional shell-tool discovery and first-run setup #9 discovery/install, Tool configuration center: read/edit supported tool config inside /settings #83 tool configuration, Performance strategy: TypeScript first, native helpers only when measured #155 performance strategy.

Research references

• Starship: https://github.com/starship/starship (existing provider)
• Oh My Posh: https://github.com/JanDeDobbeleer/oh-my-posh (research candidate only)