feat(status + dashboard): "ready/iterating" + 4 UX polish items

[fstamatelopoulos]

Summary

Two-themed PR bundling related dashboard UX work.

Part 1: "ready/iterating" status (the original PR scope)

Multi-loop projects (PA + SA driving the workspace through several sequential milestone loops, like the gmbot run) had a misleading "COMPLETED" badge between loops — the project wasn't done, it was iterating.

Fix: one new explicit transition + one label rename.

• Transition: {completed, failed, stopped} + cfcf review (standalone) → idle. Triggered at the top of startReview(). Narrow trigger by design — only SA (scope work). Not reflect (retrospection), document (finalizing), or spec (PA, independent surface with its own chip from PR feat(ui): interactive-agent surfacing — History section + workspace card 'PA active' chip #51).
• Paused stays paused — preserves resume mechanics + refine_plan action.
• In-loop architect (runReviewSync) doesn't flip — it transitions through running already.
• failed and stopped stay distinct internally for audit-trail preservation.
• Label rename: idle → \"ready/iterating\" in both StatusBadge.tsx (web) and cfcf status CLI. Internal value stays idle — zero state-machine churn.

Part 2: 4 UX polish items (added during dogfood review)

1. History tab section headers: (N) → (active: A | total: T). Surfaces "is something running RIGHT NOW" at a glance.

2. Status tab PhaseIndicator subtitle: \"Iteration 24\" → \"Iteration 24 (max: 28)\". The ceiling sits next to the live timer.

3. Status tab Loop State block trimmed: iteration count + max moved to PhaseIndicator (above). Pause-every lives in Config tab already. Block replaced with conditional "Stall warning" — only renders when consecutiveStalled > 0.

4. Workspace card three changes:

• Live elapsed timer for running loop. New workspace.loopStartedAt field (server-enriched from loopState.startedAt when activeAgent === \"loop\"), rendered inside the loop chip via the existing useElapsed hook.
• Chips moved BELOW title+badge row. With loop + PA chips both possible, single-row header was tight; two-row layout gives each chip room. Conditional rendering — no chips, no row.
• Agents row adds Reflect. PA NOT added (global config — identical on every card). Architect + Documenter omitted to keep row scannable; can add later if dogfood demands.

Test plan

• bun run test — 1079 tests pass (+8 new for Part 1; Part 2 is pure presentation, no new tests)
• bun run typecheck — clean
• Part 1: 8 new tests in architect-runner.test.ts covering TERMINAL_LOOP_STATUSES exhaustiveness; flips from completed/failed/stopped; does NOT flip paused (load-bearing — preserves resume mechanics); does NOT flip running/idle; defensive undefined-status handling
• Manual smoke (post-merge):
  • On gmbot (currently "completed"), run cfcf review → dashboard badge flips to "ready/iterating" + cfcf status shows the new label
  • Start a loop, watch the workspace card's chip show \"● loop running · 12s\" ticking live; PA chip appears in a second row when PA is active
  • History tab section headers show \"Interactive sessions (active: 1 | total: 3)\" correctly
  • Status tab top-bar shows \"Iteration N (max: M) · 15m 00s\"; "Loop State" block is gone unless consecutiveStalled > 0

Versioning

Pairs naturally with PR #51 (also in [Unreleased]). Recommend bundling both into v0.24.5 when tagging. The CHANGELOG already has both entries side by side.

🤖 Generated with Claude Code