The supervisor's backend health monitor in scripts/hot-dev.sh only
checked whether a `./pulse` process existed via pgrep. If the Go binary
hung, panicked into recovery without exiting, or failed to bind :7655
while the process stayed alive, the monitor saw a healthy process count
and never restarted, leaving the dev frontend talking to a dead listener.
Hit that twice in one session.
Add an HTTP probe of /api/health on the dev port. After two consecutive
5s polls where the process exists but /api/health is unreachable, kill
the process and restart the same way the dead/duplicate branches do.
Cut the poll interval from 10s to 5s so the worst-case unresponsive
window is ~10s instead of ~30s. Bring deployment-installability into
line by naming alive-but-unresponsive recovery as part of the dev-runtime
orchestration boundary, and cover the new branch with static-source
assertions in scripts/tests/test-hot-dev-bg.sh.
The connections-table flip in 00c6dc2dd removed the workspace's top-of-
page subtab strip but left the install panel and platforms workspace
rendering bare with no path back to the inventory table. Operators who
clicked Add a system, picked a platform, and landed on
/settings/infrastructure/install or /settings/infrastructure/platforms/*
had to leave Settings entirely and return to the tab to get back to the
table.
Render a persistent breadcrumb header above every non-inventory view
inside InfrastructureWorkspace: a "← Connections and Inventory" button
that navigates back to the bare /settings/infrastructure route, plus a
heading naming the current subview. Inventory itself is unchanged.
Bring agent-lifecycle rule 7 into line by requiring this back-to-
inventory affordance on every detail route under /settings/infrastructure
so deeper surfaces are never reachable without a visible return path.
Cover the new header on install and platforms subviews — and its absence
on the inventory landing — in InfrastructureWorkspace.test.tsx.
settingsTabPath('infrastructure-operations') and the initial-tab resolver
still returned /settings/infrastructure/install from the pre-flip
workspace, so clicking the Settings nav dropped users on the install
detail page instead of the unified Connections table shipped in
00c6dc2dd. Point the tab path and canonical resolver at the bare
/settings/infrastructure, extend deriveTabFromPath to recognise that
prefix as the infrastructure-operations tab, and bring
frontend-primitives rule 13 into line with the already-flipped
agent-lifecycle rule 7 so both contracts name the bare
/settings/infrastructure as the canonical landing. Update the three
guardrail suites (settingsRouting, settingsNavigation.integration,
settingsArchitecture) that encode the tab->path map so they reflect the
new landing.
Replace the three-subtab install/connect/inventory narrative with a
single unified connections view at /settings/infrastructure. The base
route now renders one alpha-sorted table whose rows cover every
monitored system — Proxmox VE, PBS, PMG, TrueNAS, VMware, and agent
hosts — under the same name/kind/method/status/last-reported shape.
Adding a new system goes through a single "Add a system" picker whose
tiles route operators straight into the right flow:
/settings/infrastructure/install for the agent choice and
/settings/infrastructure/platforms/<kind> for Proxmox, TrueNAS, and
VMware. The install, platforms, and operations routes remain reachable
as detail surfaces, and read-only sessions continue to redirect the
install view back to the inventory base and suppress the add-system
entry point.
Update the agent-lifecycle subsystem contract to reflect that the
workspace shell no longer mandates a "bare routes default to install"
first-host narrative or a three-subtab layout. Refresh the guardrail
and integration tests that pinned the old tab shape and the legacy
operations landing route.
Rename the third subtab to "Inventory" and hide the first-system
orientation card once any platform connection or agent resource exists.
In the installer, drop the two cross-pointer callouts, replace the
locked-state placeholder with a disabled command preview, collapse the
uninstall block into a details disclosure, and keep the Platform
connections CTA as an inline link next to the target-profile note.
Remove the dead CalloutCard branch from ProxmoxSettingsPanel and update
the architecture guardrails accordingly. Pin the "Inventory" label and
the orientation-card gating into the agent-lifecycle contract so the
workspace-shell rule stays aligned with the surfaced narrative.
Audit of customer-facing Patrol surfaces against the patrol-intelligence
contract (obligation #5) confirms the migration boundary:
- Customer chrome (page titles, nav, route labels, summary copy, actions,
empty states) leads with Patrol/Pulse Patrol/Pulse Assistant
- Internal identifiers (file names, stores, types, transport fields,
log tags) stay under the shared AI-runtime carve-out
- `/settings/system-ai` is the provider/configuration carve-out where
AI terminology is reserved; its chrome is already product-first
("Assistant & Patrol")
- The AgentProfilesPanel "Ideas" tooltip was the last customer-facing
surface outside those carve-outs and was neutralized in the prior
commit
With the boundary documented, the only release-ready blocker is RA8
(rc-to-ga-promotion-readiness), which is held intentionally for the
governed RC-to-GA rehearsal.
Cut three surfaces that nagged Community users without explicit feature
engagement: the Dashboard RelayOnboardingCard paywall, the app-wide
ActiveUseTrialNudge (already dead code with no render sites), and the
SetupCompletionPanel "Monitor from Anywhere" Relay trial block.
Feature-gated discovery that fires only when a user clicks a locked
feature (alert investigate, history ranges, Patrol AI modes, Settings
panels) is intentionally left alone — those are user-initiated, not
blanket funnels.
Governance cascade: cloud-paid.md extends the Settings no-funnel rule
to Dashboard and setup completion; registry.json, status.json readiness
assertions RA5/RA18, frontend-primitives/storage-recovery/
performance-and-scalability contracts, the high-risk verification
matrix, the relay registration proof script, the subsystem lookup
tests, and the obsolete integration spec are all scrubbed of
references to the removed files. Blocked-record regenerated against
current VERSION=6.0.0-rc.2 so the promotion-policy test no longer
diffs on a stale rc.1 artifact.
Settings -> Plan was funneling Community users through marketing:
the monitored-system upgrade arrival banner (which triggered
license-server portal handoffs that emsofo hit as "Pulse Account
unavailable"), the trial-start CTA, the inactive-Pro upsell, and
the capacity-section Upgrade button all rendered unsolicited for
users without paid features.
Plan page now shows factual license state only for Community
users. Trial starting is still reachable from other entry points
(dashboard onboarding, patrol). Contract updated to forbid
regression. Dashboard-level nags remain; inventory pending.
White-label branding was advertised as a Cloud Enterprise feature in
docs/CLOUD.md and docs/PULSE_PRO.md while the underlying capability is
still marked "not implemented" in features.go. Pulled from the customer-
facing matrix and prose so we are not selling a vapor feature.
TERMS.md described Pulse Pro as "infrastructure monitoring and AI-powered
automation" — narrowed to "infrastructure monitoring tool with AI-assisted
operations" so the legal document matches the product's actual posture.
Also trimmed the Pro row subtitle in the self-hosted pricing table to
drop "automation" since the AI autonomy surface is framed as assistance
with approval, not automated operations.
The `Current State` section of `docs/release-control/internal/CONTROL_PLANE.md`
still described `v6-rc-cut` as the active engineering target and listed
`v6-rc-stabilization` as planned. The active target has been
`v6-rc-stabilization` since the first governed RC shipped, and
`documentation_currentness_test` now fails on that drift. Update items 2, 4,
and 5 to match the current governed state.
Share OpenAI ChatStream finalization across DONE and clean EOF so buffered final chunks still produce tool calls and done events for OpenAI-compatible providers.\n\nFixes #1411
Fix snapshot, backup, powered-off, and config-reevaluation guest threshold resolution by routing them through shared guest alert context instead of override-only fallback paths.\n\nFixes #1418
