A stored rule result keeps the rule's template; only the web and the MCP
server resolved it. The notification builders sent the explanation as
stored, so a channel card showed {{reason}}, {{appName}},
{{failedSubcategory}} and {{errorCode}} although the session had recorded
all four values.
- RuleTemplateInterpolator: the backend copy of the resolution, applied in
BuildRuleFiredAlert and AddRuleResultSections before the text is cut.
In a notification a list item whose value was not recorded is dropped
and a token in running text becomes an en dash.
- The cut lands on a line, sentence or word boundary instead of a
character count; the limits are unchanged.
- tests/fixtures/rule-template-interpolation/cases.json: one case file
run by the backend, the web and the MCP server.
- The engine's evidence auto-field list is one constant. The web, the MCP
interpolator and the validate_rule token lint gain timeoutHours, which
the engine already recorded.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* feat: tenant Hello wait timeout up to 1 hour
The "Hello Wait Timeout" setting (30..300 s in the UI) had no effect in the
V2 agent: its only consumer, HelloTracker.StartHelloWaitTimer, has no
production caller, while the decision engine armed a fixed 300 s
hello_safety deadline after the ESP exit and the tracker ran a fixed 300 s
completion timer after the wizard start.
The setting is now the real Hello budget, clamped to 300..3600 s
(HelloWaitTimeout in AutopilotMonitor.Shared). Every value that was valid
before (<= 300, including the old 30 s default) keeps today's 5 minutes.
- Engine: the agent stamps helloWaitTimeoutSeconds on EnrollmentFactsObserved
(only when > 300), recorded set-once in EnrollmentScenarioObservations.
All six hello_safety arm sites share one builder with the tenant window;
a fact that arrives after hello_safety was armed moves it out. The WDP
backstop is never shorter than the tenant window; the signal census
surfaces hello_wait_extended.
- Agent: the Hello tracker's wizard completion timer uses the same budget;
the stall probe skips session_stalled while hello_safety is pending.
- Backend: ValidateModel rejects values outside 300..3600; stored rows are
read as their effective value, so legacy 30 s rows never block unrelated
saves or backup reverts. Default 30 -> 300; shared manifest and wire types
regenerated.
- Web: input 300..3600, bounds pinned against the C# constants.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S17pipjWqjZ8ULAu1JrFyG
* fix: keep the Hello wait's 30 s minimum and default, raise only the maximum
Product decision: the setting keeps its UI contract (30 s minimum and
default, same description); only the maximum moves to 3600 s. Configured
values up to 300 s keep the built-in 5-minute Hello window, larger values
extend it.
- HelloWaitTimeout: MinSeconds/DefaultSeconds 30; BuiltInSeconds 300 is the
floor of EffectiveSeconds.
- Backend validates only a changed value (30..3600), like the retention cap;
the read-time normalization is gone again.
- The agent stamps the tenant window only above BuiltInSeconds.
- Web keeps the original text and default; only the maximum and its hint
change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S17pipjWqjZ8ULAu1JrFyG
* fix: Hello wait default 300 s, stale-fire guard, deferred stall report
Product decision: default 300 s (what V2 has always waited), minimum 30 s
still accepted, maximum 3600 s. Values up to 300 s keep the built-in
5-minute window, which the settings hint now says.
Review follow-ups:
- Engine: a hello_safety fire whose incarnation was replaced (armed due
later than the fire's due time) is dead-ended, same guard as RealmJoin —
a late tenant window could otherwise be undercut by the queued fire of
the built-in deadline. The tenant fact is recorded only when it extends
the built-in window.
- Stall probe: session_stalled is held back while a tenant-extended Hello
wait (hello_safety or the WDP backstop) is pending and emitted once it is
over and the session is still idle; real activity drops it. Default
tenants never hit the guard.
- Comments: WDP backstop wording, 5-min reference, AgentConfiguration.
- Web pins the default against the shared manifest; tracker constructor
defaults use HelloWaitTimeout.DefaultSeconds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S17pipjWqjZ8ULAu1JrFyG
---------
Co-authored-by: Claude <noreply@anthropic.com>
Every visit by a user without a tenant role joined the tenant broadcast
group before the dashboard redirected to /progress, and the settings
area loaded the bootstrap session list for every role. The backend
refused both by rule: 134 AddToGroup 403s, 19 serial-less session
joins and 27 bootstrap-list 403s in 30 days, each one also an
api_request_failed event.
- lib/signalrGroupAccess mirrors the backend's same-tenant join rules
(broadcast and member-notify groups need an own role or platform
scope, the admin-notify group the Admin or a GA, a session group a
serial for roleless callers, global-admins platform scope, an
unparseable name is a 400). Cross-tenant joins stay with the
backend's platform/delegated admission. The gate keys on
hasOwnTenantOrPlatformRole, not hasTenantReadScope: a delegated admin
without an own role is refused the home tenant's groups too.
- SignalRContext.joinGroup applies it before sending: a refused join
reaches onDenied with the server's status and is never held for the
reconnect rejoin. joinGroup depends on the caller's role fields as
primitives, so its identity changes only when auth/me resolves, the
same commit that delivers the tenant id consumers join with. That
also ends the "tenant-" join the dashboard sent before the tenant id
was known (the 400s).
- TenantConfigContext loads bootstrap sessions only for
canManageBootstrapSessions (mirror of BootstrapManagerOrGA: GA, Admin,
Operator with the bootstrap grant).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Under D-267 a managed tenant saw every operator-provisioned access as
"Platform support". A customer in two operator groups run by different teams
got two identical entries under "Who can read this tenant" and could not tell
them apart.
A Global Admin can now give an operator group a customer label (at most 80
characters) with the existing PATCH global/tenant-groups/{groupId}, which takes
only the fields to change: name and/or customerLabel, where an empty label
clears it. The customer sees the label as the entry name; a group without one
and direct grants keep "Platform support". Owned (self-service) groups refuse a
label with 409, because their customers see the managing tenant's name.
RenameTenantGroupAsync becomes UpdateTenantGroupAsync, one read-modify-write on
the meta row.
DelegationSelfService.CustomerNameOf is now the one source of the name a
customer sees for a group. The managers view and the Group detail of the audit
rows written under every member tenant (assign, unassign, group delete) both use
it. Those audit rows carried the internal group name to the customer's audit
page before this change.
Admin, Tenant Groups: operator groups show "Customers see: ..." and get a
"Customer label" button next to Rename. The customer page is unchanged.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
WhiteGlove seal: IME reports "EspPhase: AccountSetup" from the device session
as soon as the device apps are done when the user ESP is enabled. The sealing
classifier counted that as user activity (-40) and left pre-provisioning Part 1
unsealed, so no whiteglove.complete marker was written and the Part-2 boot hit
the 48-hour emergency break. The excluder now also needs the ESP registry to
show AccountSetup progress: the adapter stamps accountSetupProgress on
WhiteGloveShellCoreSuccess, only after the tracker has actually read the
AccountSetup subcategories (unread is never "untouched"). The new excluder is a
subset of the old one, so a verdict can only rise; signals without the key
score as before. WhiteGloveShellCoreSuccess v1 gets its replay fixture.
The Part-1 exit writes whiteglove.complete as its first step instead of after
analyzers, diagnostics upload, grace and drain, where a reseal could kill the
process before it.
Registration failures: a run that exits because register-session failed
records outcome, last error, network link at start and end, the wall-clock
attempt window and the config-fetch outcome in registration-failure.json. The
emergency break sends the record as AgentErrorReport.PriorRegistrationFailure
(null for older agents); the backend bounds it and adds it to the ops details,
the App Insights dimensions and the timeline data. Retry behaviour is unchanged.
The raw registry trigger whiteglove_signal_* is renamed save_result_step_*: the
step exists in every 25H2 enrollment and the old name read as a
pre-provisioning signal.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The agent measured the session age at its 48 h break but never sent it.
AgentErrorReport now carries SessionAgeHours (optional, null for older
agents), and the backend puts it into the ops details, the timeline data and
the App Insights dimension. It is information only: the backend cannot derive
it (no row, or a row registered seconds before the break after a hibernate
resume), and it never feeds the severity.
The backend now reads the Sessions row once for the owner binding and the
break verdict. A report for a session without a row (in practice one that
never registered) is Info and writes no timeline event and no EventType index
row: nothing can close or open that session, and the orphan sweep deleted the
event a day later with a Warning of its own. A row under cascade delete gets
nothing written past the lock. Terminal stays Info with lateCleanup; an open
status or a failed read stays Warning with the historical write. Break
materialization now requires GUID ids.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Each Sessions row is bound to the device identity that registered it. The
binding ran in shadow; it now refuses every foreign-identity outcome:
register-session answers 403 with session_owner_mismatch before anything is
written, agent telemetry 403 in the error envelope (SessionOwnerMismatch),
the hosted diagnostics upload URL 403. The error channel stays 200 and only
skips its write into the named session. An evaluation defect fails open.
The agent already rotates its session id on the 403 at registration; it now
does the same for the coded 409 (session id claimed by another tenant) instead
of retrying six times and exiting, and fails without retry on any other coded
409 (BackendRefusalException, never an auth failure). A snapshot restored
after a rotation adopts the current session id.
A guard test pins the four call sites and fails when a site observes the
binding without acting on the refusal. Privacy page: session binding under
how data is protected.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The audit log, ops events, rule submissions and session reports each
carried a copy of the same paging state. Next bumped the page number and
cursor stack before the request and kept them when it failed, so a failed
request showed "Page 3" over the rows of page 1.
useCursorPager (pure transitions in lib/cursorPager) owns rows, cursors
and the stack; the page number is derived from the stack. A move commits
only after its page loaded, and only the latest move commits. The fetch is
read through useLatest, which retires the per-page workarounds for the
token getter's changing identity.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The cross-partition pages (audit log over all tenants, ops events over
all categories) kept one continuation entry per partition that still had
rows. On the tenant fan-out the nextLink outgrew the host's URL limit:
the follow-up request was rejected with HTTP 414 before the function ran,
so Next never loaded a second page for a Global Admin.
The continuation is now the (RowKey, Partition) of the last returned row.
Every page reads each partition through BoundedFanOut (pageSize + 1 rows
after the watermark) and merges by (RowKey, Partition), which is the
display order in both tables. Partitions after the watermark's partition
continue with `RowKey ge`, the others with `RowKey gt`, so equal ops
RowKeys across categories survive a page cut. The merge no longer sorts
by the resolved OccurredUtc while advancing by RowKey, which could skip
rows where the two disagree.
The fan-out was unbounded (Task.WhenAll over every tenant); it now runs
with the shared cross-tenant concurrency ceiling like the session list.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
App install rows now carry their install channel. The ingest folds the RealmJoin
watcher's package started/completed events into AppInstallSummaries rows (channel
"realmjoin", keyed by scope + package id; duration = completion minus the observed
start). IME rows keep their key and have no channel column, so they read as before.
Every per-app aggregate groups on (channel, name): apps list, app analytics and
sessions (new ?source=, default ime), metrics/app, SLA top failing apps and the
version-regression radar (tracker key, episode key, bell link, ops event). The app
install SLA rate includes RealmJoin packages. The ESP-blocking stamp stays IME-only.
RealmJoin packages do not feed the software inventory or CVE correlation.
Web: a RealmJoin pill and an All / Intune / RealmJoin filter on Software > Installs,
the pill on app detail, Fleet Health and SLA, source-aware app links and PDF meta
line, and a hint in the empty Inventory tab. The pill moves out of InstallProgress
into a shared component. MCP: get_app_install_metrics names the source field.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The platform forwards whatever certificate the TLS client presented, so an
unauthenticated caller can hand in a leaf signed by a CA it made up. The
chain engine then fetched the leaf's AIA caIssuers URL before rejecting it:
an attacker-directed GET from the backend, and a worker blocked for the
10 s retrieval timeout. Every legitimate issuer is embedded and Intune
certificates carry no AIA, so downloads are now disabled and the unused
retrieval timeout is gone.
The regression test signs a leaf with a CA that is neither embedded nor
sent and asserts that the listener behind its AIA URL sees no connection;
a self-signed leaf never reaches that path.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A tenant configuration row is never written blind any more (D-290). Three writers saved
from the 5-minute instance cache and four wrote a fresh read without an ETag: auto-re-enable
could lift a suspension a Global Admin had just renewed, the PUT could write the cached
Disabled=false over the offboarding tombstone during the drain barrier, and a concurrent
first login could overwrite the write-once OnboardedBy.
- TenantConfigurationService.CreateOrUpdateAsync next to UpdateAsync (shared core): fresh
read with ETag, tombstone refused, mutation decides, conditional replace, conditional
insert for a missing row, re-read and re-decide on a lost race.
- SaveConfigurationAsync and IConfigRepository.SaveTenantConfigurationAsync deleted; new
TryCreateTenantConfigurationAsync. The default row of GetConfigurationAsync is a
conditional insert.
- Converted: first-login seed (writes only while the stored row has no domain; only the
winning login fires the signup side effects), auto-re-enable (decided on the fresh row),
auth client-id tracking, app-homing flip, conferred-Pro grace stamp, offboarding tombstone.
- PUT/POST config/{tenantId} removed (function, catalog entries, response DTO). The Global
Admin tenant editor sends only its changed fields to the field PATCH; the Collect Logs
quick-config PATCHes its two diagnostics fields with ?intent=collect-logs, which keeps
the CollectLogsQuickConfigEnabled ops event. Shared diff helper lib/tenantConfigSave.ts.
- Tests: create-or-update and default-row races, concurrent first login, renewed
suspension, tombstone paths; repository and service doubles in ConfigRepoMockExtensions.
Regenerated shared manifest and wire types (web, MCP).
Deploy order: backend, then web right after (the old web's editor save and quick-config
use the removed PUT).
Bumps internal (D-290, plan, docs).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Offboarding deletes the tenant configuration, and the returning tenant's first login recreated
it with TrialConsumed=false, so the one-time 30-day Pro trial could be chained by offboarding
and re-onboarding.
- The offboard click reads the configuration fail-loud and ORs TrialConsumed into the
OffboardingByTenantPointer (sticky across rounds, same ETag CAS as OffboardCount).
- The first-login write (HandleNewTenantDomainAsync, the one write that sets OnboardedBy)
carries it onto the new row; an unreadable pointer means no write, retried next login.
Re-onboarding itself is unchanged, auto-approve included.
- TenantConfigurationService.UpdateAsync: fresh projected row with ETag, offboarding
tombstone refused (403 TenantSuspended), mutation may decline, conditional replace with
the ordinary pre-write backup, re-read and re-decide on a lost race (409 after 3).
SetTenantPlanTier and StartTrial use it, so a warm instance's cached row is never saved
(it could lift the tombstone during the drain barrier) and two concurrent trial starts
cannot both pass.
- Tests: TenantConfigUpdateTests (service + both endpoints), pointer stickiness and the
offboard endpoint in TenantOffboardFunctionTests, first-login carry-over in
AuthFunctionSideEffectTests; shared EndpointHarness.
Bumps internal (D-288, D-289, plan, docs).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
SsrfGuard.ValidateDestinationAsync resolved the webhook host and checked
it, then HttpClient resolved it again for the POST, so a DNS-rebinding
host could pass the check and be connected to an internal address.
The webhook client's SocketsHttpHandler now connects through
SsrfGuard.ConnectToPublicAddress: it resolves once per connection,
refuses the connection if any address is blocked, enforces HTTPS and
connects the socket only to the checked addresses; TLS still runs against
the request host. The pre-flight check is removed, the registration moves
to AddWebhookNotificationHttpClient so a test can prove the wiring, the
notification retry policy no longer retries a refused destination, and
the test endpoint reports the refusal with the same message as before.
IsBlockedAddress follows the same model as the MCP server: all IANA
special-purpose space, IPv6 only 2000::/3, zone ids and IPv4-mapped
addresses refused (IPNetwork.Contains judges a mapped address by its
IPv4 form, so those are refused first).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The SSRF gate resolved the client_id host, checked the answer and then
handed the hostname to fetch(), which resolved it again at connect time.
A DNS-rebinding host could pass the check with a public answer and be
connected to an internal one.
The metadata fetch now runs through a dedicated undici dispatcher whose
connect.lookup is the gate: one resolution per connection, every address
must be public, the socket connects only to those addresses. The global
fetch is untouched.
The address classifier now refuses every IANA special-purpose block
(IPv4), allows only global unicast 2000::/3 minus its special-purpose
blocks (IPv6), and refuses zone ids and IPv4-mapped addresses. The tests
enumerate the unedited IANA registries (tests/fixtures/
iana-special-purpose-addresses) and prove on loopback sockets that a
refused host opens no socket and that an allowed one is connected to
exactly the address the gate returned.
Adds undici ^8.11.2 (engines >=22.19.0).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
When ipinfo.io (plus retry) and ifconfig.co both fail, the agent now asks
geo.prod.do.dsp.mp.microsoft.com/geo. It returns only CountryCode and
ExternalIpAddress, but it belongs to the DO/Windows Update endpoint set Autopilot
needs anyway, so it stays reachable where public geo services are blocked.
The session then gets a country-only device_location (source "do-geo", no
timezone, so timezone auto-set stays pending) instead of none.
- GeoLocationService: the three providers share one TryProvider skeleton;
response parsers are internal and unit-tested; GeoLocationResult.Describe()
skips missing parts ("DE" instead of ", , DE") in the event message and log.
- Failure trace carries doGeoError + lastResortProvider.
- Backend: geographic location keys skip empty region/city, so country-only
sessions group as "DE" rather than ", DE"; drilldown still round-trips.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The ClientAppNotRegistered refusal (D-279, now enforced) sends Tenant Admins
to this form; its hint required access_as_application, which an app acting
on behalf of users does not need.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Each save handler still builds the edited configuration but now sends changedAdminConfigFields(loaded, edited) through lib/adminConfigSave.ts, and nothing at all when nothing changed. This covers the Global Settings, the diagnostics paths, the ops alerts and the vulnerability section. The MCP access policy patches its single field without reading first. A page loaded before another writer (the agent release pipeline, a second operator, a sync job) can no longer revert that writer's fields. Deployed after the backend that serves PATCH global/config.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
On 2026-09-25 a Global Settings save from a page loaded before an agent release wrote the whole configuration back and reverted the release's LatestAgentV2* hashes. After that, every 2.0.1465 agent reported IntegrityCheckFailed. The flaw was structural: every writer (portal PUT, plan tiers, three vulnerability-sync stamp paths) replaced the whole row from the copy it held, usually the 5-minute cache.
- IConfigRepository.UpdateAdminConfigurationAsync(mutate, updatedBy) is the one write path. It reads the row fresh with its ETag and applies the mutation to a copy materialized from that read. It computes the changed columns from the converter projections and writes only those onto the freshly read raw entity; a null value drops the column. The replace is conditional, and a conflict triggers a re-read and retry. A no-op writes nothing. Pipeline-owned and unknown columns survive. The full-model SaveAdminConfigurationAsync / SaveConfigurationAsync are removed, and the first row is created with AddEntity.
- PATCH global/config with { fields } (PatchAdminConfigurationRequest) replaces PUT/POST. Its field gate is derived from the persisted columns and rejects server stamps, LatestAgentV2* / LatestBootstrapV2ScriptVersion, PlanTierDefinitionsJson, the sync stamps and redacted placeholders. The route catalog lists PATCH.
- Plan tiers, TriggerMsrcSync, TriggerVulnerabilityDataSync, VulnerabilityDataSyncFunction and VulnerabilityCacheRefreshService.StampAsync each hand in a mutation.
- Tests: AdminConfigurationChangedFieldsTests (stale-page toggle keeps the release hashes, concurrent pipeline write re-read and kept, no-op, rejection, null drop, first row, gate); existing tests adapted; the request guard allows the admin field map.
- Regenerated shared manifest and wire types (web utils + MCP). The web still sends PUT until the web commit that follows the backend deploy.
- The internal pointer is bumped to D-285.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A portal-registered client (amc_) now receives its Entra refresh token only as a compact JWE (RFC 7516, dir + A256GCM, kid amc-rt-v1). The key is derived from OAuthStateSigningKey. At refresh, the registration comes from the envelope, never from the client_id parameter. Before, a copy of the token redeemed without client_id went to the organizations authority and skipped the registration check.
A raw refresh token presented with an amc_ client id is accepted and returned sealed, so these clients migrate on their next refresh. All other clients are unchanged. jose becomes a production dependency; it was a dev-only transitive dependency before. The internal pointer is bumped to D-283/D-284.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The option is a regular tenant-admin toggle in its own "Without pre-registration" block, with the
add-on permission state read from the existing graph-permissions status endpoint. The dashboard
banner and the admin Ready badge key on "any validation method enabled" (feature flag
deviceValidationEnabled, manifest-driven validate* fields) instead of Autopilot alone, so a
device-preparation tenant that validates by device association is no longer told that agent
ingestion is blocked. Stale "shadow mode" wording for Device Association is gone, the
Not-registered report names every method, sessions show "Intune Enrollment", and the privacy
page and FAQ list the new method. Grant script help describes the IntuneDeviceBinding feature
as Intune Enrollment Validation.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The certificate's Intune managedDevice id is looked up in the tenant's Intune. With the tenant
switch ValidateIntuneDeviceBinding on, that lookup is the last accepting link of the validator
chain: an enrolled device that is not being retired, wiped or deleted is admitted as
ValidatorType.IntuneEnrollment. A 404 for a certificate younger than 60 minutes is
NotFoundRecentEnrollment (503, never cached); a missing grant is 503 with 120 s.
For devices another validator admitted, the lookup only observes: in the background, never
admitting or rejecting, and only when the app token carries
DeviceManagementManagedDevices.Read.All, so a tenant without the add-on costs no Graph call.
Outcomes ride on the request row (CertDeviceBinding, role, serial match, owner and enrollment
type).
TenantConfiguration.HasAnyDeviceValidation() is the single gate precondition; the feature flags
expose it as deviceValidationEnabled (replaces validateAutopilotDevice).
The agent parses ValidatorType strictly: release the agent before this backend.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The registrations (name, callback URL, client id) live in their own table,
not in TenantConfiguration, so the report only showed the limit. Fetch
GET tenants/{tenantId}/mcp-client-registrations fail-soft alongside the
other side views and render them in the Tenant Status section, plus the
platform switch. The limit row is relabelled to say it is a limit.
Claude-Session: https://claude.ai/code/session_01DbRs7NCppizQ7wWFjN7n3M
Co-authored-by: Claude <noreply@anthropic.com>
The query recognizes only npm limiter packages, never oauthRateLimit /
oauthTokenRateLimit, and flags a handler as soon as it calls a function
named like *auth*, *login* or *verify*. All six alerts it ever raised were
on throttled OAuth routes (the last two came from tenantAuthority()).
What it stood in for is now a test: every route of the OAuth router must
run a per-IP limiter first unless it is one of the three named discovery
documents, /oauth/token must carry the strict budget, and a router.use()
middleware fails the check. Verified by mutation: removing the limiter
from /oauth/authorize fails the test with the route name.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The registration lookup put a client_id-derived value into the backend URL
path unencoded. The amc_ regex already limits it to 32 hex characters, so
nothing changes at runtime, but every other request-derived path segment in
the MCP server goes through encodeURIComponent, and CodeQL counts only that
(or a ?/# prefix) as a request-forgery barrier - a regex capture group is
not one. Closes code-scanning alert js/request-forgery on oauth.ts.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A ?serial= parameter (serial number or exact device name) runs one lookup
as soon as sign-in has resolved the tenant, so organisations can send users
a link to follow their own device. Every hit writes the found session's
serial back into the URL for reload, bookmark and copying. Authorization is
unchanged: sign-in stays mandatory and roleless callers still need the exact
serial server-side. The query survives the login redirect because
ProtectedRoute keeps pathname + search. progress_serial_submitted carries
source form | link.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Settings > Tenant > AI Integration (between SLA Targets and Bootstrap
Sessions) explains the normal connection first - server URL for Claude,
ChatGPT, VS Code and other hosted assistants, nothing to register - and
shows the self-hosted client registration below it (Tenant Admin, platform
switch on). The page is visible to admins, platform scopes and anyone with
MCP access. The self-hosted example text is neutral.
- A tenant may hold one self-hosted client registration. The GA-only
TenantConfiguration.McpClientRegistrationLimit (1-10) raises it on request:
reverted for non-GA callers on PUT, refused on PATCH, edited in the Global
Admin tenant dialog. MCP_SERVER_URL joins the portal URL registry.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A self-hosted AI client redirects to its own domain, which the vendor
redirect allowlist never admits, so it could not use the browser sign-in.
Tenant Admins can now register the client's exact OAuth callback in the
portal and configure the client with client id amc_<registrationId>.
- MCP OAuth proxy: an amc_ id is resolved through the backend; its
registration is the redirect allowlist entry for that exact callback.
Authorize and token run at the registering tenant's Entra authority, and
the token endpoint discards a token of any other tenant (code and
refresh). Lookups are cached for 60 s, a failed lookup fails closed. Every
other client keeps the allowlist check before its resolution.
- Backend: table McpClientRegistrations, TenantAdminOrGA routes
tenants/{tenantId}/mcp-client-registrations (list, create, delete; at
most three, https or loopback http, exact URI, audited) and the anonymous,
per-IP rate-limited lookup auth/mcp/client-registrations/{id} (under
/api/auth for the client-certificate exclusion list).
- AdminConfiguration.McpClientRegistrationEnabled (default off) is the
platform switch: off hides the portal section and makes every lookup 404.
auth/me carries it for the sidebar.
- Portal: Settings > Tenant > Self-hosted AI clients (Tenant Admin / Global
Admin), and the switch in Admin > Settings > Global.
- Tests: MCP tenant-client flow (authority, tid check, revocation, cache,
allowlist), backend round trip, wire parity, validation, cap, switch,
audit; nav visibility. Manifests and wire types regenerated.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The proxy issues no client secrets, but its authorization server metadata
named no token_endpoint_auth_methods_supported (RFC 8414 then implies
client_secret_basic) and /oauth/register echoed a requested secret-based
method. A client following that default registered with client_secret_basic
and then failed the token exchange ("client_secret_basic authentication
requires a client_secret"), reproduced with the SDK calls a self-hosted chat
client makes. The metadata now lists ["none"] and registration always answers
none (RFC 7591 lets the server replace a requested value).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A user token obtained by an application other than the platform's own
registrations (for example a self-hosted AI client using on-behalf-of) was
accepted with the user's full role; only audience, issuer, signature and
lifetime were checked. Such tokens are now classified on every request.
- ClientAppBindingMiddleware (after authentication, before policy enforcement)
records the request dimensions ClientAppBinding (registered | unregistered |
unidentified) and ClientAppId, plus a throttled Warning.
- AdminConfiguration.EnforceClientAppBinding (default off; Admin > Settings >
Global): when on, a foreign client is refused with 403 ClientAppNotRegistered
unless the application is an enabled member of the caller's tenant (the
app:<client-id> row a service principal uses). An admitted caller keeps their
identity but gets the application caps: at most Viewer, no platform role,
DelegatedReader on managed tenants, roleless tiers closed except auth/mcp.
- AdminIdentity.IsCapped (application or admitted foreign client) is what the
global, delegated and member-role resolvers switch on; RequestContext carries it.
- Own clients = the audience trust set plus EntraId:TrustedClientAppIds, a
client-only trust for the local development registration that never widens
the accepted audiences.
- Tests: classification, unforgeable cap marker, middleware in both modes, caps
through the real policy middleware, pipeline order; manifests and wire types
regenerated.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A backend 401 on `initialize` logged only the UPN, and the backend's own trace
("IDX10214 audience validation failed") never names the token; a registration
rejection logged only a count of redirect_uris. Diagnosing a self-hosted client
(LibreChat) that first failed DCR on its own callback host and then presented
tokens minted for another resource took a customer round trip.
- access-guard: `invalid-token-claims` and `backend-rejected-token` carry
`aud=... azp=... tid=...` (unverified claims; control characters stripped, capped)
- oauth/register: the rejection names origin + path of up to three rejected
redirect_uris (never the query); the error_description names the actual
defaults (VS Code, not Google)
- tests for both lines; skill catalog in .claude/commands (mcp-telemetry) updated
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Route payloads (__next.*.txt, index.txt), static JSON and the router's HEAD
self-requests were 20k of 62k dependency rows per week and nobody reads them.
lib/webTelemetryExclusions.ts builds the patterns (own origins minus /api/,
plus relative non-API URLs) for two layers: excludeRequestFromAutoTracking-
Patterns for string fetches, and a dependency initializer that drops the
finished item by its absolute target - the SDK's own check sees an empty URL
for the URL objects Next's router passes to fetch (verified against the ingest
batch: 11 such fetches, 0 shipped; API, SignalR and token rows still shipped).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
P1 - full loads
- staticwebapp.config.json: /_next/static/* immutable, /landing/* max-age=86400
(SWA default was max-age=30 for hashed assets); guard test pins it
- components/NavLink.tsx: prefetch on hover/focus/touch for the portal chrome
(prefetch={false} also disabled the hover prefetch); landing links stay off
- app/layout.tsx: preconnect to the API origin
- lib/dashboardSeed.ts + lib/prefetchSeeds.ts: the dashboard's first list and
stats requests start with auth/me and are handed to the hooks as one-shot
seeds (the own-tenant list URL carries no tenantId; the backend reads the JWT)
- lib/msalConfig.ts: the login request carries the API scope, so one token
redemption serves the first API call; Graph consent via extraScopesToConsent
P2 - chatter per navigation
- lib/signalrGroupRegistry.ts + contexts/SignalRContext.tsx: group membership
is reference-counted with a 2 s grace leave; reconnect rejoins run in
parallel; on/off/invoke and the context value are memoised; the dashboard's
unconditional leave no longer drops the notification bell's global-admins
membership; consumers pair every join with a cleanup leave
- app/sessions/hooks: both group joins in parallel, the join catch-up is
dropped while the initial walk is in flight, identical session deltas skip
the state update
- lib/cachedAuthFetch.ts: TTL cache for feature flags, tenant list and latest
versions with invalidation on every write path; FeedbackBubble defers its
status call
P3 - pure derivations only (D-274): memoised filter arrays, one-pass phase
summary, status counts, useDeferredValue on both search inputs
P4 - landing
- lib/authHint.ts + AuthGate + globals.css: the loading overlay is applied by
an inline head script only when the site root is opened by a browser with an
auth hint; the prerendered HTML carries no overlay; CSS lifts it after 8 s
- hero as WebP srcset (81 KB instead of a 180 KB PNG), dead landing-v1 CSS
removed, terminal demo pauses off-screen, duplicate twitter image dropped
Telemetry (D-276)
- lib/appInsightsRuntime.ts: the SDK is its own chunk, deferred until idle on
the public host, remote config fetch off (blkCdnCfg), initial trackPageView
on both hosts, Web Vitals as web_vital_* custom metrics
Verified: tsc, eslint --max-warnings 0, vitest 127 files / 1442 tests, next
build, Playwright click-through of every portal route, settings section,
session views and public pages on the production bundle against the live API.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A platform script that ran 10 s was shown as "1h 00m" (session 4377911b): its
start line was self-anchored (+120 min) while its result line, re-read after an
overwrite rewind and therefore never fresh, fell back to the agent's cached zone
(+60 min), was rejected as ahead of the clock, and the duration was taken from
the raw pair. The same shape fired script_timeout_suspected daily in one tenant
on 2-9 s scripts read as 17 h. Fleet 2026-09-20..23: about 1 % of script
durations, 28 of 28 inflated ones carrying derivedTimestamp.
Agent: ScriptExecutionState binds the provenance of its start, result and exit
lines (CmTraceLineProvenance, persisted); EmitScriptCompleted judges the pair -
verified (both measured, or both assumed with the same offset from the same
file), unverified, or suppressed with durationSuppressedReason
(rejected-endpoint | offset-grid, CmTraceOffsetCalibrator.IsOnOffsetGrid) - and
always emits durationProvenance plus startSourceOffset*. A held completion's
sourceOffset* tags now describe its own result line; script_timeout_suspected
only judges a measurable duration; the emitted-run marker is capped at the
clock so a future-shifted result no longer swallows the next run's start.
Web: the reducer applies the same rule to stored events from older agents,
pairing a final with the latest script_started of its policy, and the panel
shows "n/a" with the explanation where the run time would be.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
ImeRegistryAppStateObserver reads a fourth IME registry surface,
Win32Apps\OperationalState\<userGuid>\<appGuid>: the install deadline a
downloaded app waits for (ExecutionDeadlineTime, IME 1.106) and the
app-in-use deferral counters (DeferralsUsed, DeferUntilTime,
DeferralMaxDeferrals, DeferralAutoDeferred, IME 1.105). Both travel on the
existing registry_app_state event as data (installDeadlineUtc, deferralsUsed,
deferralMaxDeferrals, deferUntilUtc, deferralAutoDeferred) with the change
triggers installDeadline / deferralsUsed / deferUntil, and the message says
the state in words ("downloaded, install deadline ... UTC", "deferred by user
1/3 until ... UTC", "auto-deferred ..."). No new event type.
ClassifyStatusServiceStatus maps IME's legacy AppInstallStatus explicitly:
1000/1002/1004 installed, 1001/1003 installing, 2000 band notApplicable,
3000 failed, anything else unjudged. The 1000 band was treated as installed,
but IME maps every InProgress enforcement state to 1001, so a running app
counted as terminal success on the registry rail (observability only).
Tests: explicit status mapping, both registry timestamp shapes, the new
change triggers, the wait-state text, and the tick flow for a deadline and a
deferral event. Mutation proof: classifier and trigger broken on purpose
turn exactly the five expected tests red.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The offboard endpoint tombstones the tenant (Disabled=true, "Offboarding in
progress") before it answers 202, so the departing admin submits the farewell
feedback from an already-suspended tenant. Since the suspension gate moved onto
the shared authorization path (43ddc601, 2026-08-29) every feedback POST was
403 TenantSuspended — invisible, because nobody clicked and the two stored
feedbacks predate the gate (D-270).
- Endpoint catalog: `AllowedDuringOffboarding` on exactly
POST tenants/{tenantId}/offboard/feedback; the middleware ignores the
offboarding tombstone for that route only. An operator suspension still
denies; policy tier, cross-tenant check and the function's active-marker
check are untouched. Middleware tests for all four cases plus a catalog
ratchet pinning the flag to the one route.
- Farewell mail: IOffboardFarewellEmailSender.SendAsync returns whether the
provider accepted; the offboarding worker records FarewellEmailSent /
FarewellEmailSkipped / FarewellEmailFailed like the welcome mail — the
send's own success log is Information and never reaches Application
Insights. Missing Email:ApiKey logs Warning. Handler tests per outcome;
OpsAlertRules catalog and the generated manifests (Web + MCP vocabulary)
carry the three types.
- Web: auth/me 403 TenantSuspended no longer shows alert() and forces a
logout. AuthContext keeps the user signed in with claims-only info and
ProtectedRoute renders a suspended page: for the offboarding tombstone
"Your organization is being offboarded" with the farewell feedback form
(a reload during the 6-minute preparation window keeps the form), for any
other suspension the reason text plus contact links; both with sign-out.
The feedback widget moved from OffboardingSection into
components/OffboardingFeedbackForm.tsx.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Every tool description and both role instructions now fit the 2048-char host cap
(Claude Code truncated the Global Admin instructions and two descriptions silently).
Pagination, projection and scope mechanics are said once in the instructions and
the shared pageSize/continuation/tenantId/days texts; role markers left tools only
that role can see; get_resource lists its sections on the argument. The SDK $schema
URL is stripped from tools/list.
Measured (compact tools/list): Global Admin 111.4 KB -> 89.1 KB, Reader 94.7 -> 74.3,
tenant user 64.5 -> 51.3, delegated 75.1 -> 55.6 KB. catalog-budget.test.ts pins the
new ceilings, the variant counts and the empty over-cap list.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
list_tenants named a real customer domain as its example; the filterTenantList test
carried three real tenants. Both now use synthetic names and GUIDs.
catalog-budget.test.ts pins the per-role wire catalog as a ratchet (size, repeated
argument texts, no foreign domains or GUIDs, the 2048-char host truncation cap with
the four known over-cap texts listed). scripts/audit-catalog.ts dumps and sizes the
catalog per role for before/after measurements.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The XML summary added on TenantManagerItem.Name in ef280b7b rides into
shared-manifests.json as the field's doc; CI run 35701895623 caught the drift.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
GET delegations/managers no longer exposes the name of an operator-created
Tenant Group to the customer; both operator shapes (groups and direct grants)
carry the neutral label. Assignees, reader count and source stay visible.
Web badge reads "Provisioned by operators". D-267.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
network_state_change and network_connectivity_check join the non-activity
list next to power_state_change. A flapping dock or access point kept the
idle-stopped periodic collectors running for the whole episode and reset
the stall clock of the enrollment it was stalling. The events are emitted
unchanged; the device-info network snapshots stay activity.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Oliver Kieselbach