feat: separate OpenTUI dashboard app/package

[christso]

Objective

Create a separate OpenTUI-based terminal dashboard app for AgentV that provides a keyboard-first local UI for browsing run history, live eval progress, and result summaries without expanding the core evaluation runtime.

Architecture Boundary

external-first

This should live as a separate app/package in the monorepo rather than inside the main CLI package runtime. The dashboard may consume existing AgentV artifacts and shared libraries, and the CLI may later provide a thin launcher entrypoint, but the TUI itself should remain isolated from core eval execution logic.

Design Latitude

Engineer may choose the exact package/app location and launch model, for example apps/tui with an optional agentv dashboard launcher in apps/cli.

Prefer reusing existing AgentV result/history abstractions and output artifacts over inventing a new plugin system. If a shared dashboard data/query layer is needed for both web and TUI surfaces, keep that shared layer UI-agnostic.

Acceptance Signals

• A dedicated OpenTUI app/package exists in the monorepo and can run independently of the main CLI package internals.
• The TUI can read existing AgentV run artifacts or the same history storage used by dashboard/reporting features.
• The TUI provides at least these core views:
  • run list / history
  • active run or recent run summary
  • per-run detail view with pass/fail and score breakdown
• The UI is keyboard-first and usable entirely in the terminal.
• The implementation does not require introducing a general-purpose tokentop-style plugin runtime.
• Shared data-loading logic, if introduced, is reusable by other dashboard surfaces.
• Packaging/runtime assumptions specific to OpenTUI/Bun remain isolated from the main agentv CLI package.

Non-Goals

• Replacing or merging with the web dashboard work in feat: self-hosted dashboard — historical trends, dataset management, YAML editor #563.
• Designing a general third-party plugin architecture for the dashboard.
• Moving core evaluation logic into the TUI package.
• Requiring the main CLI package to depend directly on OpenTUI internals.
• Full parity with the web dashboard on the first pass.

Related

• feat: self-hosted dashboard — historical trends, dataset management, YAML editor #563 self-hosted dashboard — web/local browser dashboard
• Interactive eval TUI should list available models by provider #520 interactive eval TUI should list available models by provider
• Research: tokentop OpenTUI and plugin/TUI architecture comparison in agentevals-research