Skip to content

armgit / Pulse

Public Imported from GitHub
0

Real-time monitoring dashboard for Proxmox VE, PBS, Docker, Kubernetes, TrueNAS and vSphere. Self-hosted, with alerts for failed backups, full disks, unhealthy containers and offline nodes.

Consolidate current Discovery contract ownership

rcourtmand5a6c887f694
diff --git a/docs/release-control/v6/internal/subsystems/agent-lifecycle.md b/docs/release-control/v6/internal/subsystems/agent-lifecycle.md
index 9e5a5982af..bc0f87fb3c 100644
--- a/docs/release-control/v6/internal/subsystems/agent-lifecycle.md
+++ b/docs/release-control/v6/internal/subsystems/agent-lifecycle.md
@@ -4118,24 +4118,9 @@ Agent` secondary handoff against the live setup wizard instead of relying
### AI Discovery removed (2026-10-11)
-AI Discovery, the model-backed service-identification layer, is gone. The
-`internal/servicediscovery` store and scanner, `internal/ai/infradiscovery`,
-the `pulse_discovery` Assistant/Patrol tool, the `/api/discovery*` routes, the
-mock Discovery fixtures, the resource/guest/node/Docker drawer Discovery tabs,
-identified-service cards, readiness badges, provenance markers, suggested
-web-interface URLs, the Workloads AI Context column, the chat "Discovery is
-off" hint and the AI Service Context settings page were all removed. No
-telemetry counter ever measured its use (the `discovery_enabled` telemetry
-field folded it together with network subnet discovery), the maintainer judged it a flop, and
-it carried a separate scheduler, store and model route for context that
-`pulse_query`, `pulse_read` and the unified-resource context sections already
-supply.
-Network subnet discovery (`internal/discovery`, `/api/discover`, the
-Infrastructure discovery settings) and the `discoveryTarget` agent coordinates
-are separate features and stay. Stored `discovery_enabled`,
-`discovery_interval_hours` and `discovery_model` AI settings are ignored on
-load. The old `/settings/pulse-intelligence/discovery` route redirects to
-Assistant settings.
+Agent resource context retains the backend-authored `discoveryTarget` coordinates.
+The [AI runtime retirement boundary](ai-runtime.md#ai-discovery-removed-2026-10-11)
+owns the removed service-identification layer and its compatibility behavior.
### Registered Windows native-exit completion proof (8 October 2026)
diff --git a/docs/release-control/v6/internal/subsystems/ai-runtime.md b/docs/release-control/v6/internal/subsystems/ai-runtime.md
index c790736f2d..cbe2db5a78 100644
--- a/docs/release-control/v6/internal/subsystems/ai-runtime.md
+++ b/docs/release-control/v6/internal/subsystems/ai-runtime.md
@@ -5193,8 +5193,7 @@ companion tests (`internal/ai/findings_lifecycle_test.go`) pin the behavior.
The interaction-quality scenario corpus
(`internal/ai/chat/interaction_scenario_corpus_test.go`) is the canonical
-regression home for chat-feel promises, mirroring the Discovery corpus
-(`internal/servicediscovery/scenario_corpus_test.go`). Each scenario drives a
+regression home for chat-feel promises. Each scenario drives a
full `ExecuteStream` turn against a scripted provider and pins the
browser-facing event stream — ordered event types, forbidden event types,
answer-text teeth, and payload teeth — for the user-visible promises shipped
@@ -6191,43 +6190,8 @@ runtime must carry that approval identifier into the final
`agentexec.ExecuteCommandPayload` so the host agent can re-check the shared
command policy locally and fail closed on blocked or still-unapproved commands
instead of treating control-plane approval as an implicit bypass.
-Discovery deep scans are the one runtime that does not flow through the
-approval boundary. `internal/ai/discovery_adapter.go` is the only call site
-allowed to mark an `agentexec.ExecuteCommandPayload` as `Trusted`. The
-catalog of probes lives in `internal/servicediscovery/commands.go`, is
-read-only by construction (`cat`, `ps`, `ss`, `find` under known config
-roots) and is wrapped in `docker exec`, `pct exec`, or `qm guest exec`
-without ever interpolating caller-supplied strings. Both the server-side
-`agentexec` authorize path and the agent-side `hostagent` authorize path
-must honor that `Trusted` flag by bypassing the approval requirement,
-while still enforcing `PolicyBlock`. AI tool calls, Patrol fixes, and
-Assistant remediation must continue to flow through the governed approval
-record path and must never set `Trusted` on their payloads.
-Discovery command-backed scans are additionally gated by the operator's
-Discovery setting, not by the mere presence of a service-discovery store,
-connected agent, or command-capable agent token. `discovery_enabled=false`
-must fail closed for background sweeps, `/api/discovery/run`, forced
-single-resource discovery, and `pulse_discovery` refreshes before any
-`DeepScanner` command dispatch. `discovery_enabled=true` with
-`discovery_interval_hours: 0` is the only manual-command-scan mode: recurring
-scans stay stopped, but explicit admin-triggered refreshes may use the
-hardcoded trusted catalog.
-Each per-organization AI service now passes its organization-pinned
-`AgentServer` interface directly into `internal/ai/discovery_adapter.go`.
-Deep discovery does not recover the concrete global `agentexec.Server` or
-silently drop command execution when the runtime supplies an
-`agentexec.OrganizationServer`; connected-agent lookup and command dispatch
-therefore remain on the same tenant-scoped view. The interface-backed service
-regression in `internal/ai/service_test.go` requires a real deep-discovery run
-to dispatch the trusted probe catalog and retain its command evidence.
-The value boundary for keeping Discovery is observed workload context:
-Assistant and Patrol may consume normalized service name, version, endpoint,
-port, config path, data path, log path, bind-mount, confidence, and user-note
-fields through `pulse_discovery` or scoped prefetch. Raw command output remains
-debug/admin material and must not become the primary Assistant context. When a
-Discovery record includes a suggested web URL, the tool response and prefetch
-summary may include that URL as observed context, but it must be treated as a
-candidate rather than an operator-approved management URL.
+AI tool calls, Patrol fixes and Assistant remediation must not mark
+command payloads as `Trusted` to bypass the governed approval path.
The same action-audit boundary now also requires persisted action records to
carry a normalized plan and preflight: action id, request id, capability,
approval policy, dry-run availability, safety checks, verification steps, and
diff --git a/docs/release-control/v6/internal/subsystems/api-contracts.md b/docs/release-control/v6/internal/subsystems/api-contracts.md
index fbf465d0a4..b8f7c07e23 100644
--- a/docs/release-control/v6/internal/subsystems/api-contracts.md
+++ b/docs/release-control/v6/internal/subsystems/api-contracts.md
@@ -4961,24 +4961,9 @@ auto-register mutation boundary.
### AI Discovery removed (2026-10-11)
-AI Discovery, the model-backed service-identification layer, is gone. The
-`internal/servicediscovery` store and scanner, `internal/ai/infradiscovery`,
-the `pulse_discovery` Assistant/Patrol tool, the `/api/discovery*` routes, the
-mock Discovery fixtures, the resource/guest/node/Docker drawer Discovery tabs,
-identified-service cards, readiness badges, provenance markers, suggested
-web-interface URLs, the Workloads AI Context column, the chat "Discovery is
-off" hint and the AI Service Context settings page were all removed. No
-telemetry counter ever measured its use (the `discovery_enabled` telemetry
-field folded it together with network subnet discovery), the maintainer judged it a flop, and
-it carried a separate scheduler, store and model route for context that
-`pulse_query`, `pulse_read` and the unified-resource context sections already
-supply.
-Network subnet discovery (`internal/discovery`, `/api/discover`, the
-Infrastructure discovery settings) and the `discoveryTarget` agent coordinates
-are separate features and stay. Stored `discovery_enabled`,
-`discovery_interval_hours` and `discovery_model` AI settings are ignored on
-load. The old `/settings/pulse-intelligence/discovery` route redirects to
-Assistant settings.
+The [AI runtime retirement boundary](ai-runtime.md#ai-discovery-removed-2026-10-11)
+owns the removed service-identification API and ignored legacy AI settings.
+Network subnet discovery keeps its separate `/api/discover` contract.
### Masked webhook updates preserve edits without disclosing saved fields
diff --git a/docs/release-control/v6/internal/subsystems/monitoring.md b/docs/release-control/v6/internal/subsystems/monitoring.md
index 93b7c13e31..63a65fd906 100644
--- a/docs/release-control/v6/internal/subsystems/monitoring.md
+++ b/docs/release-control/v6/internal/subsystems/monitoring.md
@@ -3672,26 +3672,10 @@ truthfulness, not native thaw, containing-release or workload acceptance.
### AI Discovery removed (2026-10-11)
-AI Discovery, the model-backed service-identification layer, is gone. The
-`internal/servicediscovery` store and scanner, `internal/ai/infradiscovery`,
-the `pulse_discovery` Assistant/Patrol tool, the `/api/discovery*` routes, the
-mock Discovery fixtures, the resource/guest/node/Docker drawer Discovery tabs,
-identified-service cards, readiness badges, provenance markers, suggested
-web-interface URLs, the Workloads AI Context column, the chat "Discovery is
-off" hint and the AI Service Context settings page were all removed. No
-telemetry counter ever measured its use (the `discovery_enabled` telemetry
-field folded it together with network subnet discovery), the maintainer judged it a flop, and
-it carried a separate scheduler, store and model route for context that
-`pulse_query`, `pulse_read` and the unified-resource context sections already
-supply.
-Network subnet discovery (`internal/discovery`, `/api/discover`, the
-Infrastructure discovery settings) and the `discoveryTarget` agent coordinates
-are separate features and stay. Stored `discovery_enabled`,
-`discovery_interval_hours` and `discovery_model` AI settings are ignored on
-load. The old `/settings/pulse-intelligence/discovery` route redirects to
-Assistant settings.
-
-The service-discovery availability-suggestion helpers under `internal/servicediscovery` were deleted with the package; availability probes are configured directly.
+The [AI runtime retirement boundary](ai-runtime.md#ai-discovery-removed-2026-10-11)
+owns the removed service-identification scheduler and suggestion helpers.
+Network subnet discovery remains monitoring-owned. Availability probes activate
+through the canonical saved-target configuration path.
### Interface-associated guest address selection
@@ -4308,16 +4292,8 @@ evidence envelopes, telemetry, logs, or Patrol context. Execution-defined
contract edits use the existing server-authored configuration revision so
history visibly breaks at the new proof definition instead of blending unlike
checks.
-Discovery-led assurance proposals remain monitoring-owned read evidence until
-an operator explicitly creates a canonical availability target. The
-`internal/servicediscovery` suggestion contract derives only bounded
-HTTP/HTTPS/TCP candidates, normalizes each candidate into a stable evidence
-fingerprint, and persists dismissal against that exact fingerprint. Equivalent
-rediscovery retains a dismissal; a material endpoint, service identity, or
-inference-reason change clears it. Disposition writes reject stale
-fingerprints, and neither discovery refresh nor dismissal may create, enable,
-schedule, or execute a network check. The explicit availability-target create
-path remains the sole activation boundary.
+The explicit availability-target create path is the activation boundary for
+saved availability probes.
HTTPS checks also author one canonical certificate observation from the same
probe execution. `internal/availabilityprobe` captures the presented leaf and
`pkg/tlsutil/certificate.go` derives subject, issuer, SANs, SHA-256 fingerprint,
diff --git a/docs/release-control/v6/internal/subsystems/patrol-intelligence.md b/docs/release-control/v6/internal/subsystems/patrol-intelligence.md
index 77d8a0dc09..e5f74462aa 100644
--- a/docs/release-control/v6/internal/subsystems/patrol-intelligence.md
+++ b/docs/release-control/v6/internal/subsystems/patrol-intelligence.md
@@ -1328,24 +1328,12 @@ contract.
### AI Discovery removed (2026-10-11)
-AI Discovery, the model-backed service-identification layer, is gone. The
-`internal/servicediscovery` store and scanner, `internal/ai/infradiscovery`,
-the `pulse_discovery` Assistant/Patrol tool, the `/api/discovery*` routes, the
-mock Discovery fixtures, the resource/guest/node/Docker drawer Discovery tabs,
-identified-service cards, readiness badges, provenance markers, suggested
-web-interface URLs, the Workloads AI Context column, the chat "Discovery is
-off" hint and the AI Service Context settings page were all removed. No
-telemetry counter ever measured its use (the `discovery_enabled` telemetry
-field folded it together with network subnet discovery), the maintainer judged it a flop, and
-it carried a separate scheduler, store and model route for context that
-`pulse_query`, `pulse_read` and the unified-resource context sections already
-supply.
-Network subnet discovery (`internal/discovery`, `/api/discover`, the
-Infrastructure discovery settings) and the `discoveryTarget` agent coordinates
-are separate features and stay. Stored `discovery_enabled`,
-`discovery_interval_hours` and `discovery_model` AI settings are ignored on
-load. The old `/settings/pulse-intelligence/discovery` route redirects to
-Assistant settings.
+Patrol reads canonical resource evidence through the shared AI runtime.
+The [retirement boundary](ai-runtime.md#ai-discovery-removed-2026-10-11) applies
+to the removed service-identification layer, independently of the retained
+Watch only, Ask first, Safe auto-fix and Autopilot modes.
+
+### Patrol operator workflow
The Patrol 2.0 frontend now uses a one-task-at-a-time, decision-first
information hierarchy. The compact, borderless command row owns enabled
diff --git a/docs/release-control/v6/internal/subsystems/performance-and-scalability.md b/docs/release-control/v6/internal/subsystems/performance-and-scalability.md
index 5926197054..c70867819a 100644
--- a/docs/release-control/v6/internal/subsystems/performance-and-scalability.md
+++ b/docs/release-control/v6/internal/subsystems/performance-and-scalability.md
@@ -2745,15 +2745,10 @@ app-container routing and Docker-only actions. `frontend-modern/src/hooks/useWor
`frontend-modern/src/components/Workloads/workloadTopology.ts`, and
`frontend-modern/src/components/Workloads/useGuestRowState.ts` must preserve a
canonical app-container navigation path while keeping Docker runtime action
-identifiers explicit. Discovery affordances on the Workloads drawer must follow
-the canonical `discoveryTarget` contract, not generic app-container host
-fallbacks. Guest-drawer manual discovery run controls must route through the
-shared `DiscoveryTab` manual-run action and the canonical target identifiers
-instead of adding Workloads-local discovery trigger state. TrueNAS
-app-containers may reuse runtime metadata such as image and
-runtime strings, but they must not inherit Docker-only update affordances or
-agent-only Discovery tabs on the Workloads row path unless the unified
-resource contract explicitly supplies discovery ownership.
+identifiers explicit. Agent resource context follows backend-authored
+`discoveryTarget` coordinates. TrueNAS app-containers may reuse image and
+runtime metadata, while Docker-only update actions remain scoped to
+Docker-managed workloads.
That same row-and-selection boundary also requires canonical app-container
identity to stay intact on the workload surface. Workload rows, deep links,
drawer selection, anomaly correlation, and route-state filters must key on the
@@ -3437,35 +3432,6 @@ other than `will_fix_later`. The newer `AutoResolved` attribution flag
on the same UnifiedFinding shape stays a fixed-size `bool` in the same
struct, so the operator-vs-Pulse closure attribution adds no per-row
allocation pressure on the same hot path.
-That same Workloads drawer boundary now also owns passive discovery
-surfacing in the overview. `useGuestDrawerState.ts` may issue a single
-`GET /api/discovery/{type}/{target}/{id}` per drawer open through
-`getDiscovery` to populate an "Identified Service" card in
-`GuestDrawerOverview.tsx`, but must never trigger a scan or modify
-discovery state — manual scans, progress UI, and approval prompts stay
-owned by `DiscoveryTab` and `useDiscoveryTabState`. Render is gated by
-`getDiscoveryIdentifiedSummary` so empty or low-signal records collapse
-the card entirely instead of mounting an empty container. The drawer
-fetch must reuse the existing `discoveryTarget` derivation (resource
-type, agent id, resource id) rather than re-implementing target
-resolution, so workloads without canonical discovery ownership (e.g.
-TrueNAS app-containers without explicit `discoveryTarget`) skip the
-fetch cleanly.
-Observed service version, source labels, observed age, and suggested web
-URL candidates may be rendered from that same cached discovery summary,
-but they must not add a second discovery request, route-level Suspense
-dependency, row-time endpoint probe, or auto-save mutation. Suggested
-URLs remain copy/open/adopt affordances only; persisting them as a
-workload link stays a user action through the existing Web Interface URL
-field. For Docker-managed app-containers, that field must receive the stable
-host-plus-container-name metadata ID rather than the runtime container ID, while
-the row and drawer still use the canonical workload ID for all non-metadata
-state.
-The compact provenance marker for those values is presentation-only. It
-may identify already-loaded Discovery fields, but must not introduce
-additional discovery fetches, provider lookups, endpoint probes, or
-per-row layout measurement on the Workloads hot path.
-
The canonical Workloads table mobile floor is 36rem. The mobile table branch
must preserve that intrinsic width inside its existing scroll shell instead of
combining it with `min-w-full`, which allows the operational columns to
diff --git a/docs/release-control/v6/internal/subsystems/security-privacy.md b/docs/release-control/v6/internal/subsystems/security-privacy.md
index 77bcd0064c..df247b26bd 100644
--- a/docs/release-control/v6/internal/subsystems/security-privacy.md
+++ b/docs/release-control/v6/internal/subsystems/security-privacy.md
@@ -2267,17 +2267,8 @@ legacy `.json` knowledge files and plaintext `.enc` knowledge files may only
serve as migration input, and the knowledge store must rewrite canonical
encrypted-at-rest storage immediately on load instead of leaving guest
knowledge plaintext on disk until a future note update.
-That same trust boundary also applies at store construction time: the AI
-knowledge store and the service discovery store may not fail open into
-plaintext-at-rest mode when crypto initialization fails. If encryption cannot
-be established for those stores, construction must fail closed instead of
-quietly persisting runtime state unencrypted.
-That same rule also applies to persisted service-discovery records after store
-construction: `internal/servicediscovery/store.go` may only accept plaintext
-`.enc` discovery files as migration input. Once a discovery record can be
-read, canonical persistence must rewrite encrypted-at-rest storage immediately
-on load/list/id-scan instead of leaving plaintext discovery metadata or user
-secrets on the steady-state runtime path.
+The AI knowledge store must fail closed at construction when encryption cannot
+be established. It must not persist runtime state in plaintext.
That same trust boundary also covers audit-signing key persistence:
`pkg/audit/signer.go` may keep the 32-byte HMAC signing key in runtime memory,
but `.audit-signing.key` may only accept plaintext key material as migration
diff --git a/docs/release-control/v6/internal/subsystems/unified-resources.md b/docs/release-control/v6/internal/subsystems/unified-resources.md
index 905360e196..384e1e387f 100644
--- a/docs/release-control/v6/internal/subsystems/unified-resources.md
+++ b/docs/release-control/v6/internal/subsystems/unified-resources.md
@@ -1736,17 +1736,6 @@ coverage: `TestRegistrySeededFromMonitorJudgesFreshnessByItsThresholds` and
`TestContract_ResourceListJudgesFreshnessLikeTheMonitorBroadcast` in
`internal/api/contract_test.go`.
-Service-discovery readiness is a unified-resource payload contract, not a
-drawer-local decoration. Resource list/detail payloads that expose a
-`discoveryTarget` must also carry the backend-authored `discoveryReadiness`
-projection when the discovery owner is available. That projection is
-metadata-only (`fresh`, `stale`, `missing`, `running`, `failed`, `unavailable`,
-or `unsupported`, with provenance, generated/observed freshness, target
-coordinates, optional service/category, and bounded fact count) and must not
-include raw command output, provider commands, environment variables, config
-paths, or secret-bearing metadata. Workload and resource drawers, Assistant
-handoffs, and optional table columns consume this field instead of inferring
-freshness from independent discovery reads.
Across platform/runtime pages, workflow tabs are evidence-gated from the
canonical model that owns their rows. `Overview` is the stable landing tab;
supporting tabs appear only when their native inventory or signal exists, and
@@ -3129,26 +3118,9 @@ whole-fleet CPU relief, publication or release availability.
### AI Discovery removed (2026-10-11)
-AI Discovery, the model-backed service-identification layer, is gone. The
-`internal/servicediscovery` store and scanner, `internal/ai/infradiscovery`,
-the `pulse_discovery` Assistant/Patrol tool, the `/api/discovery*` routes, the
-mock Discovery fixtures, the resource/guest/node/Docker drawer Discovery tabs,
-identified-service cards, readiness badges, provenance markers, suggested
-web-interface URLs, the Workloads AI Context column, the chat "Discovery is
-off" hint and the AI Service Context settings page were all removed. No
-telemetry counter ever measured its use (the `discovery_enabled` telemetry
-field folded it together with network subnet discovery), the maintainer judged it a flop, and
-it carried a separate scheduler, store and model route for context that
-`pulse_query`, `pulse_read` and the unified-resource context sections already
-supply.
-Network subnet discovery (`internal/discovery`, `/api/discover`, the
-Infrastructure discovery settings) and the `discoveryTarget` agent coordinates
-are separate features and stay. Stored `discovery_enabled`,
-`discovery_interval_hours` and `discovery_model` AI settings are ignored on
-load. The old `/settings/pulse-intelligence/discovery` route redirects to
-Assistant settings.
-
-`Resource.discoveryReadiness` and the agent resource-context `DiscoveryReadiness` field are gone; the agent context `Runtime` section keeps the `Target type/agent/resource/hostname` facts sourced from `discoveryTarget`.
+The [AI runtime retirement boundary](ai-runtime.md#ai-discovery-removed-2026-10-11)
+owns removed service-identification fields and compatibility behavior. Resource
+context retains `discoveryTarget` coordinates in its Runtime projection.
### Interface-associated display addresses
@@ -3167,14 +3139,6 @@ production host adapter; monitoring's #2757 fixtures bind guest list/broadcast
order and name/address/traffic association. Native readings and publication
remain separate evidence.
-Discovery-led availability proposals present their evidence and canonical
-attachment boundary as short, complete sentences in both the inline card and
-machine-wide review. The colocated component test pins that operator-facing
-copy so the shared copy-style audit cannot regress it back to semicolon-joined
-instructions. The card's editable check name, interval and observation
-location use the shared form label and control chrome (`FormSelect` for the
-two choices), so they match the availability settings form they hand off to.
-
### Provider workflows hydrate only their active route inventory
Docker / Podman, Kubernetes, TrueNAS, and VMware surfaces declare an explicit
@@ -4763,7 +4727,7 @@ agent resources may include `agent.sensors.thermalState` even when
`agent.temperature` and `agent.sensors.temperatureCelsius` are absent, so
platform consumers can show macOS pressure without creating provider-local
thermal payloads or fake Celsius values.
-AI discovery and query surfaces now follow the same rule. Assistant runtime
+Assistant query and resource-context surfaces follow the same rule. Runtime
paths such as `pulse_query` and unified AI context must expose TrueNAS-backed
canonical `agent`, `vm`, `app-container`, `storage`, and `physical-disk`
resources through those shared contracts, filtering dataset-topology storage
@@ -5361,43 +5325,6 @@ already shown elsewhere in the drawer.
The analysis panel now expands directly inside the outer `Access` disclosure
instead of as a second peer support block, so the access surface reads as one
flattened reveal instead of another card group under the overview.
-That access-side analysis surface still follows the same shell/runtime split as
-the rest of the drawer: `DiscoveryTab.tsx` owns presentation and disclosures,
-while `useDiscoveryTabState.ts` owns API fetches, websocket progress, and
-note/discovery mutations.
-Discovery validity in that runtime state must defer to the shared
-`discoveryPresentation.ts` meaningful-context gate. CLI access, confidence, and
-diagnostics may support an audit trail, but they must not turn placeholder
-records such as unknown service/category/version with no ports, paths, facts,
-or suggested URL into a valid identified result. Generic workload types such
-as `service`, `container`, or `lxc` remain placeholders until Discovery also
-provides a more specific service signal.
-Discovery endpoint candidates are support context, not authoritative resource
-links. Resource and workload drawers may pass observed Discovery URLs into the
-shared web-interface field with visible provenance, copy/open affordances, and
-an explicit adopt/save path, but row-name links and persisted metadata must
-continue to use the operator-saved web-interface URL. Version, config-path,
-port, and endpoint facts surfaced outside the Discovery sub-tab must be
-labelled as Discovery-observed so API-owned resource facts and command-derived
-facts remain distinguishable.
-Availability proposals are the bounded exception that can become active
-resource evidence only through explicit review. The resource drawer supplies
-the canonical resource ID; the proposal surface previews inferred endpoint and
-behavior separately from operator-controlled name, cadence, and observation
-location; and the final action posts one enabled target with that exact
-`linkedResourceId`. Equivalent attached or standalone endpoints are detected
-before creation. Machine-scoped review may dismiss or restore several service
-proposals, but an item without the current canonical resource context remains
-review-only rather than attaching by hostname, address, or display name.
-That label must be visible through the shared Discovery provenance marker on
-compact cards, suggested URL panels, and other out-of-tab Discovery values
-rather than buried in helper text or inferred from the tab where the operator
-found the value.
-Command-availability guidance inside that analysis surface must consume the
-shared `frontend-modern/src/utils/discoveryPresentation.ts` settings targets
-and scan-error copy instead of hard-coding legacy settings labels such as
-Unified Agents, API Tokens, or `/settings/integrations/api`; the canonical API
-access route is `/settings/security/api`.
The overview keeps access, investigation detail, service detail, and host
detail as collapsed sibling disclosures under the primary card pair, so the
drawer keeps