mirror of
https://github.com/okieselbach/AutopilotMonitor.git
synced 2026-10-24 12:42:38 +00:00
Oliver Kieselbach
50295e8304
Session 2bc884b6: ESP failed AccountSetup/Apps with 0x80073cf9 while all 8 tracked Win32/WinGet apps succeeded — the failing app was a MSIX/Store app that installs outside the IME Win32 pipeline and is invisible to ImeLogTracker. The session ended as a blackbox failure. When an Apps-subcategory failure arms the 30s settle window, the tracker now queues a one-shot scan of Microsoft-Windows-AppXDeploymentServer/Operational (Critical+Error, session window, 200-record/10s cap), scores candidate packages (exact-HRESULT match > AppX-family error > time proximity, recency bonus), and emits esp_appx_failure_analysis with flat RuleEngine fields plus a capped candidates[] detail list — before enrollment_failed lands in the timeline. Negative results are emitted too (no_appx_candidates points back at Win32). Fire-once per session; scan runs lock-free on the threadpool; settle expiry never waits; failures degrade to collector_degraded. - New: AppxDeploymentFailureScanner (one-shot reader, injectable seam), AppxFailureAnalyzer (pure scoring/verdicts, incl. negative-decimal HRESULT + PackageFullName extraction from message/properties) - ProvisioningStatusTracker: optional scanner/dispatcher ctor seams, MaybeStartAppxFailureScan at settle-arm time - Rule ANALYZE-APP-014 (MSIX/Store app failure during ESP) keyed on verdict=appx_candidate_identified; dist regenerated - MCP resource-catalog + web HRESULT badge + gather-rules eventTypes - Tests: 16 analyzer + 7 trigger; settle-window fixture now injects a fake scanner + sync dispatcher (no real event log in tests) Inert until agent redeploy (rule needs backend redeploy). Post-deploy checks in tasks/todo.md. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>