[vishr]

Summary

Fix reproduced regressions found while auditing Driga's CLI, TUI, desktop, MCP, chat and service interfaces.

 service status
-  read stderr before stdout; wait indefinitely
+  drain both streams; cap retained output; time out and reap owned query
 TUI navigation
-  retain hidden stop confirmation
+  cancel before keys, clicks or wheel navigation
 desktop close
-  hide Chat while Full Auto may remain enabled
+  wait for known mode and acknowledge Confirm before close
 desktop navigation
-  overwrite history or re-push a missing dashboard fallback
+  preserve deliberate navigation; replace missing entries

Also restore table k navigation, retain dashboards on transient errors, and show resume commands when clipboard access fails. Update provider privacy wording and the header metric limit.

Evidence

• Before: regression fixtures reproduced blocked launchctl output, hidden TUI stop confirmations, broken Back/Forward, unsafe Chat close, lost dashboard selection and silent clipboard failure.
  After: regression tests exercise repaired shipping components and fixture subprocesses. Independent Standards and Spec reviews exposed pending-mode and wheel-cancellation gaps; Astra independently reproduced the deleted-dashboard history trap. Fixes and final results are recorded in the audit.

Merge Danger

Door: two-way

Blast Radius: interfaces

No schema, wire, configuration or assistant-history changes. Service status executes a bounded read-only launchctl query; owned fixture/process cleanup never targets user workloads. TUI stop execution retains existing confirmation and identity checks. Chat now requires a successful Confirm reset before hiding a known conversation.

Risk and rollout

Reviewable bug fixes; no migration. Installed release remains unchanged. Native windows/notifications, real vendor authentication/model turns, live fleets and real destructive reclamation remain outside fixture coverage.

Contract changes

None.

Verified

• Isolated go test ./... -count=1, go test -race ./..., final affected TUI/service tests and TUI race tests.
• Native/Windows vet, Linux CLI build, tidy, generated config and generated-reference checks.
• UI unit suite (219 tests), final App audit tests (7), typecheck, both production bundles and affected browser plays.
• Full Chromium catalogue: 1,266 tests/172 files across light/dark at desktop/laptop widths; final history plays separately pinned across all four cells after the last change.
• Final Go web/desktop tests after bundle rebuild; staged Gitleaks scan.

See the audit for browser-matrix results and explicit live-platform limits. No real assistant CLIs or production data were used in tests.