Summary

Single-file FE-only addition to `DashboardPage.tsx`. v1 row 1 untouched (Health hero + 6 cards: Operational / Cost / Sea / STOP-KORA / Last boot / DR). Adds row 2 with 4 cards for surfaces shipped after DASHBOARD v1 (#63 4d1bc11): Capabilities / Charter / Recent events / Runbooks.

No new backend endpoints. No new api.ts types. Reuses existing `DashboardCard` component, `isStubbed` helper, and per-card error fallback pattern.

Layout

Mobile collapses both rows to single column.

New card body renderers

• `CapabilitiesCardBody` — `total_tools` + `total_caps` headline + success/warning badge based on `unmapped_count`. KR-P2-CAP-PANEL still surfaces all 17 cap_* groups as unmapped until KR-P2-N closes the C2 mirror gap; v1 dashboard had no visibility into that.
• `CharterCardBody` — `revision_id` (mid-truncated, hover-full title), `rules_hash` truncated, `loaded_at` relative. "Rules pending" badge when `rules_available` is false (Charter v1 fallback mode).
• `RecentEventsCardBody` — last 5 chain events; relative timestamp in a fixed-width column + truncated `event_type` ("kora." prefix stripped for density).
• `RunbooksCardBody` — available vs pending counts + most-recently-modified available runbook (best proxy for "what's currently maintained").

Refactors

• `isStubbed` loosened to `LoadStatus` — `CapabilitiesResponse` and `RunbooksManifest` don't carry a `stub` field (always-live by design); the v1 `LoadStatus<{stub?:boolean}>` generic would have rejected them at the type layer. Property-check approach keeps runtime behaviour identical for sources that DO have the flag.
• `ALL_SOURCES` array consolidates the dashboard's 11 fetched sources for `anyStubbed` + footer aggregate. Replaces the v1 inlined 7-source `||` chain.

Footer aggregate

• live = `ready && !stub` — counts sources that returned successfully with real data (or sources without a stub flag by design)
• stubbed = `ready && stub` — counts sources returning the fallback shape
• loading/error excluded from both — only resolved states tallied

Footer count is 11 sources total (the dashboard's actual fetches). Spec said "13 sources" but that includes DIAG-BUNDLE button + DASHBOARD itself as "surfaces", not endpoint-backed counters. Footer reports the truthful number to avoid drift.

Diag bundle button

Unchanged. Spec mentioned an optional `(13 sources)` subtitle — skipped to avoid drift (DIAG-BUNDLE actually aggregates 11 sources per PR #89; the literal "13" would be wrong).

Test plan

• No new tests (pure FE composition over already-tested endpoint shapes)
• `npx tsc -b` on `web/` — clean (one TS narrowing issue caught + fixed in the `isStubbed` signature)
• `npx vite build` on `web/` — clean
• 150/151 admin-panel tests pass. 1 failure pre-existing:
  • `tests/kora_cli/test_web_server_runbooks.py::test_get_runbook_content_404_for_unauthored_manifest_entry` — pinned `dr_runbook` as unavailable in PR feat(KR-P2-RUNBOOKS-PANEL): operator runbooks viewer (12th admin panel) #84, but PR docs(KR-P2-RUNBOOKS-AUTHOR): DR + token-rotation runbooks (R4.1 §12 closure) #86 (KR-P2-RUNBOOKS-AUTHOR) since vendored it. Exact auto-improvement behaviour my RUNBOOKS-PANEL PR body predicted — just no one updated the now-stale test. Out of scope here; trivial fix would be swapping the assertion to `kora_dna` (still unvendored) or rewriting the test to mock a missing manifest entry abstractly.

Manual smoke targets

• Navigate to `/`, verify row 1 is byte-identical to v1
• Verify row 2 renders 4 cards in a 4-col grid on desktop / single column on mobile
• Click each row-2 card → confirm navigation to `/capabilities`, `/charter`, `/chain-events`, `/runbooks` respectively
• Verify per-card error fallback (red border + Retry) works by stopping the backend mid-load
• Verify footer shows correct live/stubbed counts (today: probably ~3 live + ~8 stubbed depending on which flips have landed)

🤖 Generated with Claude Code