Architecture audit: 6 milestone blockers for Living with AI Demo (04-17) and NyxID M0 (04-18)

[loning]

Summary

First architecture audit (2026-04-08) found overall health 5.0/10. CI guards pass (9/10), but hot-path deep dive revealed 6 milestone blockers.

Milestone Blockers

#	Location	Severity	Description
1	agents/NyxIdChatActorStore.cs:20	Critical	ConcurrentDictionary singleton as fact source. Process restart loses all conversation state. Cannot scale multi-node.
2	agents/StreamingProxyActorStore.cs:11-12	Critical	2x ConcurrentDictionary singletons (rooms + participants). Same as #1.
3	agents/ (all 3 projects)	Critical	Zero test coverage. NyxID M0 critical path has no automated validation.
4	ScopeServiceEndpointsTests.cs:1087	High	InvokeStreamEndpoint 3 tests fail (500 error). Affects API call path.
5	tools/Aevatar.Tools.Cli/Frontend/	High	TypeScript build fails. Missing @types/node, @tanstack/react-virtual, vitest.
6	apps/aevatar-console-web/	High	Build fails. max CLI not installed.

Guard Blind Spot

architecture_guards.sh does not scan agents/ directory. The ConcurrentDictionary violations (#1, #2) are invisible to CI.

Full Scorecard

See docs/audit-scorecard/2026-04-08-architecture-audit.md for complete dimension scores, backlog items, and guard supplement recommendations.

Action Required

1. Fix Refactor/project namespace #1-MAF-Inspired Framework Improvements #3 before NyxID M0 (04-18): migrate ConcurrentDictionary stores to Actor state or persistent storage
2. Fix Aevatar context database #4-Refactor/core cqrs parallel subsystems #6 before Living with AI Demo (04-17): resolve test failures and frontend build issues
3. Extend architecture_guards.sh scan roots to include agents/