[T3.2] Desktop entry-preflight vs workbench-runtime split

[DeliciousBuding]

Task: T3.2 - Desktop entry-preflight vs workbench-runtime split

Phase: 3 - Desktop/Web Boundary And Backend Truth
Priority: P0 | Size: M | Lane: B
S.U.P.E.R Drivers: U,E
Test Expectation: Desktop Playwright request-phase assertions
Memory/Governance Impact: None

Description

Keep Desktop Local Edge health probe as entry-preflight only and prevent Demo workbench backend access.

Acceptance Criteria

• Entry health probe can be recorded as entry-preflight
• After entering Demo workbench, no Hub/Edge runtime requests occur
• Failures are reported by phase and boundary
• Passes S.U.P.E.R Quick Check for: U,E
• Satisfies test expectation: Desktop Playwright request-phase assertions
• Updates resolved memory or instruction surfaces if durable project knowledge or agent instructions changed

Affected Files

• app/desktop/src/
• app/desktop/src/__e2e__/

Dependencies

• Depends on: See docs/plan/dependency-graph.md

---

Managed by Spec-Driven Develop workflow

[DeliciousBuding]

📊 Execution Telemetry

Metric	Estimated	Actual
Effort	M	M
SUPER Score	—	9/10
Unplanned Deps	0	1

Deltas: effort 0, SUPER +1, deps +1
Task drift contribution: 1
Cumulative drift_score: 1 (thresholds: annotate=1, replan=2, rescope=2)

Evidence: Desktop Vite Playwright/UI + fixture-unit/build. real_tested=false; no packaged Desktop, real login, real CLI/model/API, production deploy, installer/signing, or release upload tested.

[DeliciousBuding]

Closed after #408 merged into dev/delicious233. Evidence boundary remains Desktop Vite Playwright/UI + fixture-unit/build with real_tested=false; no packaged Desktop, real login, CLI/model/API, production deploy, installer/signing, or release upload tested.