Add the Governance group to V2 Admin Settings

[Paul Lizer (paullizer)]

Summary

Governance was missing from V2 Admin Settings. The only trace of it was a fallback switch labelled "Mcp destination governance", so feature policies, delegated item policies, MCP destination policies, and inbound MCP source policies could only be managed on the classic page. This brings the whole surface into V2 as a Governance group that follows the V2 admin design language, built on the existing /api/admin/governance/* APIs.

• Five Governance sections, declared in admin_settings_fields.py: feature toggles (paired personal/group, with a short "how a request is checked" explainer); feature policies edited in place with a live Enforced / Not enforced / Waiting badge; a searchable, paged delegated item policy list (create, edit or move, duplicate, inverse, delete); MCP action destination governance with a pattern builder and a deployment-restrictions readout; and inbound MCP source governance. People and group pickers search the directory plus group and public workspaces, and accept pasted IDs.
• Governance is reachable from the settings it qualifies: Agents & Actions permissions show their governance switch with a Review link, the Inbound MCP card shows the source-policy count and can create one, and each AI connection gets Manage access for its endpoint policies.
• Backend: the review endpoint accepts several entity types (an unknown type is a 400 instead of returning everything) and one exact item_id; two new admin-only routes, GET /api/admin/governance/principal-groups and GET /api/admin/governance/mcp-destination-catalog; V2 settings saves now write the same governance_feature_toggles_updated audit entry the classic save does; a status readout reports environment-forced MCP restrictions as booleans and a count only.
• Fix: fields gated on a runtime flag were filtered out a second time inside SettingsSection, which left the V2 Inbound MCP card empty even with ENABLE_MCP_UI=true.

Worth a careful look:

• Fail-closed switches (deliberate difference from classic). The classic save clears a governance switch whose feature is off. V2 keeps the value and shows "Prerequisite missing" and, for example, "Waiting for Allow Group Agents"; enforcement starts when the feature is turned on, so toggling a feature off and on cannot silently drop governance. Runtime enforcement is unchanged, and the classic page is intentionally untouched.
• Patterns that can never match are refused. The server stores any string, and the classic examples suggest transport:streamable-http, which never matches because transports are stored as streamable_http. The builder offers ids and transports from what the server loads, and rejects hyphenated transports, host patterns with a port or path, and a * before the end of a URL path.
• Policies save immediately through the governance API, not through the page Save bar. Unsaved inline policy edits lock category navigation and search (the same lock the Call agent manager uses), and a policy dialog stays open while its save is in flight. Both came out of an independent review and have regression tests.
• Allow all is sent with empty allow lists, because _normalize_policy_state reads a non-empty list as restricted.

Screenshots (synthetic fixture data)

Governance Feature Toggles, with a switch kept while its feature is off:

Feature Policies with enforcement state:

MCP Action Destination Governance:

Destination policy editor building a group-scoped transport pattern:

Related governance on Workspace Agent Permissions:

Manage access on an AI connection:

Linked issue

N/A (no associated issue)

Release Notes & Latest Features

• New Feature
• Bug Fix
• UI Enhancement
• Breaking Change
• Internal only

Is this visible to end users?

• Yes
• No

Is this admin-facing (Admin Settings, governance, deployment, config)?

• Yes
• No

Should this become a Latest Feature card?

• Yes
• No
• Already added

Screenshot needed for the card?

• Yes
• No
• Attached

Version bump

• application/single_app/config.py VERSION third segment bumped, or not needed because this is docs-only (0.261.259 -> 0.261.260)
• deployers/version.txt bumped, or not needed because deployers/ was not changed

Testing / validation

• npm run typecheck and npm run build in application/v2_ui: passed
• python functional_tests/test_v2_admin_governance_parity.py: 15/15 passed (includes the 29 checks in test_v2_admin_governance_logic.ts)
• python functional_tests/test_v2_admin_section_shell.py: 5/5 passed (31 TypeScript checks, including the runtime-flag regression)
• python -m pytest ui_tests/test_v2_admin_governance_settings.py: 9 passed
• python -m pytest ui_tests/test_v2_admin_*.py ui_tests/test_agent_delegation_v2.py ui_tests/test_admin_orchestration_planner_model.py ui_tests/test_chat_content_checks.py ui_tests/test_model_endpoint_capacity_editor.py ui_tests/test_v2_content_screening.py: 139 passed, 1 failed (test_admin_policy_is_discoverable_and_editable_before_citations_are_enabled, which fails the same way on the base commit 1b8d4c4)
• python functional_tests/test_v2_admin_*.py (each file): 27/29 passed; test_v2_admin_security_parity.py (chat_content_settings_present) and test_v2_admin_settings_schema.py (m365_trusted_download_hosts default) fail identically on the base
• python functional_tests/route_tests/test_route_blueprint_policy_inventory.py (12/12), test_route_unauthenticated_policy_contract.py (7/7), test_route_policy_test_coverage.py (2/2): passed
• python functional_tests/test_governance_enforcement_logic.py, test_governance_activity_logging.py, test_mcp_destination_governance_and_preconfigurations.py, test_inbound_mcp_governance_and_tools.py, test_content_screening_admin_settings_fix.py, test_pytest_return_false_guard.py: passed
• python -m pytest functional_tests/test_v2_admin_governance_parity.py functional_tests/test_governance_enforcement_logic.py functional_tests/test_mcp_destination_governance_and_preconfigurations.py functional_tests/test_v2_admin_inbound_mcp_parity.py (one process): 38 passed
• python scripts/build_docs_inventory.py (inventory unchanged), python functional_tests/test_docs_app_surface_coverage.py (7/7), python functional_tests/test_docs_site_quality.py (6/6): passed
• Pre-existing failures, unchanged by this PR: test_governance_route_and_wiring_coverage.py 3/4 (marker missing from the untouched route_frontend_workspace.py), test_ai_connection_embedding_defaults_api.py (classic save KeyVaultSecretStorageError NameError), test_logging_tag_standardization.py (existing Workflow* tags)
• Manual: Playwright visual pass at 1440px light and 390px dark (no overflow, no console errors); impeccable design detector reported no findings; an independent code review, whose two findings are fixed above

Documentation

• Release notes updated, or not needed (not yet: the v0.261.260 entry is still to add)
• Feature documentation updated, or not needed (docs/explanation/features/V2_ADMIN_GOVERNANCE_SETTINGS.md; docs/admin/governance.md rewritten; tie-in notes in docs/admin/agents-actions.md and docs/admin/ai-models.md)
• Fix documentation updated, or not needed (docs/explanation/fixes/V2_ADMIN_RUNTIME_FLAG_FIELD_VISIBILITY_FIX.md)

Security checklist

• New Flask routes include @swagger_route(security=get_auth_security()) (both new routes also require @login_required and @admin_required)
• Settings sent to non-admin frontends use sanitize_settings_for_user() (no non-admin settings payloads changed)
• Browser JavaScript is served from local SimpleChat static assets only; no CDN-hosted JS
• No secrets, keys, connection strings, or local-only artifacts are included (the environment readout returns booleans and a count, never the configured patterns)