[MAIN] Add agent operations board

[BarretoDiego]

Problem

The thread tree answers where work lives, but it does not provide a global operational view across environments. Users need one place to see which agents require attention, are working, need review, completed, or failed without opening every thread.

Summary

• Add a dedicated /board destination for web and desktop with entries in the sidebar and command palette.
• Add a native mobile Agent Operations destination with state tabs and environment/project scoping.
• Project all cards from the aggregate thread shell stream through a shared, provider-agnostic view model.
• Reuse existing thread commands for Stop, Archive, Settle, and Unsettle without optimistic state movement.

Behavior

• Groups threads into Needs You, Working, Review, Settled, Issue, and optional Idle history.
• Supports URL-backed multi-select filters for environment, project, provider, account, and model.
• Preserves cached cards during partial disconnection without turning connectivity failures into agent failures.
• Hides mutating actions while an environment is disconnected.
• Reprojects snoozed thread inclusion at the exact next wake boundary without polling or per-card timers.
• Keeps archived and deleted work out of the Board and preserves scoped environment/thread identities.

Architecture and performance

• Uses the aggregate OrchestrationThreadShell stream; cards never subscribe to thread detail.
• Indexes environments, projects, and provider instances with Map and classifies each shell once.
• Uses one coarse elapsed-time ticker for visible active cards and one one-shot snooze wake timer.
• Adds no persistence, migrations, IPC, origin overrides, relay aggregate, or per-card network queries.
• Keeps operational state behind one temporary compatibility adapter that can become an import bridge when Unified Agent State Model lands.

Surfaces and compatibility

• Web: full responsive board, filters, loading/empty/disconnected states, navigation, and actions.
• Desktop: uses the web route inside the existing Electron renderer; no new IPC.
• Mobile: native route, state tabs, scoped list, connectivity handling, and lifecycle actions.
• Direct, remote, relay-managed, and tunneled environments continue using their existing shell streams.
• Optional shell fields preserve compatibility with older servers.

Verification

• vp test run packages/client-runtime/src/agentBoard.test.ts packages/client-runtime/src/agentBoardStateAdapter.test.ts apps/web/src/agentBoardSearch.test.ts apps/web/src/components/agentBoard/AgentBoardPage.logic.test.ts — 20 tests passed.
• vp run --filter @t3tools/client-runtime typecheck — passed.
• vp run --filter @t3tools/web typecheck — passed.
• vp run --filter @t3tools/mobile typecheck — passed.
• Targeted lint, formatting, and git diff --check — passed.

Manual validation and evidence

UI evidence is intentionally pending maintainer validation:

• Attach a before image showing the thread-only navigation.
• Attach an after image of the wide web/desktop Board.
• Verify narrow responsive scrolling and keyboard focus.
• Verify mobile tabs, environment/project scope, and Clear scope.
• Verify a snoozed thread returns without refresh.
• Verify cached cards remain visible and expose no mutating actions while disconnected.

Known limitation

The Board uses a single isolated shell classifier until feat/agent-state-model provides the canonical API. No provider contracts, ingestion, or persistence are duplicated here; replacing the adapter is the intended integration seam.

Model: gpt-5.6-sol (high reasoning)
Harness: Codex in T3 Code

[github-actions]

Thread transfer impact

⚠️ The latest CI run did not produce a thread transfer result for dba69ad.

This comment will update automatically after the next completed run.