kandev/docs/plans/context-window-reset-freshness/task-02-context-reset-e2e.md

id, title, status, wave, depends_on, plan, spec

id	title	status	wave	depends_on	plan	spec
02-context-reset-e2e	Cover reset freshness end to end	done	2	01-invalidate-reset-context-usage	plan.md	../../specs/ui/requirements/context-window-reset-freshness.md

Task 02: Cover reset freshness end to end

Intent

Prove through the real reset UI and mock-agent runtime that the old ring disappears after reset and only a fresh agent usage report makes it return.

Acceptance

• The browser regression starts with a visible seeded context-window ring, completes the existing reset confirmation flow, and observes that the ring is absent while the fresh conversation has no report.
• A later mock-agent command that emits usage_update makes the ring visible with the new conversation's reading.
• The reset control is available again once the reset settles and the session returns to idle; it remains hidden during the busy reset turn.
• The existing reset divider, idle recovery, and post-reset message flow remain covered.

TDD sequence

1. Extend the existing reset-context recovery scenario with a seeded stale ring and fresh-report assertion.
2. Run the focused scenario against a fresh production build after Task 01; it passes with the reset invalidation and fresh usage propagation in place.
3. Refactor only shared seeding/page-object code needed to keep the scenario readable.

Files likely touched

• apps/web/e2e/tests/session/session-recovery.spec.ts
• apps/web/e2e/pages/session-page.ts only if a stable context-ring locator is not already adequate

Dependencies

Task 01.

Parallelism

sequential — this test depends on both backend and frontend invalidation behavior.

Mobile parity

This change normalizes shared session-runtime state and does not alter composition, controls, touch behavior, scrolling, or breakpoints. Both toolbars consume the same TokenUsageDisplay; existing apps/web/e2e/tests/chat/mobile-context-window-source.spec.ts covers the mobile rendering and touch disclosure. No new mobile-specific scenario is required.

Verification

• cd apps/web && pnpm e2e:run --host tests/session/session-recovery.spec.ts -- --grep "reset context hides stale usage"

Inputs

• Task 01 completed behavior.
• Existing reset flow in apps/web/e2e/tests/session/session-recovery.spec.ts.
• Mock-agent /background 1ms usage update path in apps/backend/cmd/mock-agent/handler.go.

Verification result

The focused Playwright regression passed against a fresh backend and Vite build:

cd apps/web && pnpm e2e:run --host tests/session/session-recovery.spec.ts -- --grep "reset context hides stale usage"
1 passed (9.4s)

No browser artifacts or blockers were produced. The regression uses the existing reset dialog/divider flow, verifies the control returns at idle, and uses the mock agent's /background 1ms usage update.