Problem

Creating a user via Setup → Users → Create User (POST /api/v1/auth/admin/create-user) built a login-capable account but never wrote a sys_member row. The invite (organization/invite-member) and add-member (organization/add-member) flows do create membership. So the two "add a teammate" paths diverged: in a single-org deployment the admin-created user did not "belong to" the Default Organization and was missing from the Members list — while invited users showed up. That inconsistency is what this PR resolves.

Note: in single-org mode membership does not gate data access (no OrgScopingPlugin, the tenant_isolation RLS wildcard is stripped — RBAC permission sets gate access), so this was a UX/consistency wart, not a security hole. But the missing membership was still confusing and out of step with the invite flow.

Change

1. Tactical fix — runAdminCreateUser binds membership

runAdminCreateUser now binds the newly-created user to the organization when it is unambiguous — i.e. exactly one sys_organization exists (single-org, including plugin-auth's default-org bootstrap, ADR-0081 D1):

• Single-org (1 org): insert sys_member { organization_id, user_id, role: 'member' }.
• Multi-org (≥2 orgs): no-op — the active-org-aware invite / add-member endpoints own membership there, and guessing an org from this header-less server call would risk the wrong one.
• Zero orgs (org capability off): no-op.

The bind is:

• idempotent — checks for an existing (organization_id, user_id) membership before inserting, so retries/races don't trip the unique index;
• best-effort — a failure logs a warning and never fails account creation;
• observable — the outcome is surfaced in the response data and the audit metadata (organizationId, membershipCreated).

This mirrors the existing pattern in ensure-default-organization.ts, which already inserts sys_member directly under a system context.

2. Strategic proposal — ADR-0093 (Proposed)

The bug is one instance of two structural gaps, so this PR also carries the ADR that proposes closing them for good:

docs/adr/0093-tenancy-mode-and-membership-lifecycle.md — Tenancy mode as a first-class capability, and a single owner for the user→membership lifecycle. Key decisions (Proposed, for review):

• D1/D2 — the "every new user gets a membership" invariant gets ONE owner: a membership reconciler composed into better-auth's user.create.after hook (covers create-user, import-users, signup, SSO JIT — present and future paths), with an explicit membershipPolicy: 'auto' | 'invite-only'. Host hooks chain first and win; the reconciler yields to existing memberships.
• D3/D4 — a tenancy kernel service (mode, isolationActive, requested, degraded, defaultOrgId()) becomes the single source of truth, retiring the four ambient signals (env flag, org-scoping probe, org row-counting — including this PR's own heuristic — and frontend feature flags re-derivation).
• D5 — the silent degraded state (OS_MULTI_ORG_ENABLED=true but @objectstack/organizations missing → tenant RLS stripped with only a console warning) fails fast at boot, with an explicit OS_ALLOW_DEGRADED_TENANCY=1 escape hatch that brands the deployment degraded end-to-end.
• D6 — bounded idempotent backfill for pre-existing member-less users (single-org only).
• D8 — explicit non-goals: single-org membership still does not gate data access; RLS strip semantics and the dual feature flags keep their meaning.

Phase 2 of the ADR retires this PR's endpoint-level bind in favor of the reconciler; the two are safely double-covered in the interim (both idempotent, both yield-to-existing).

Files

• packages/plugins/plugin-auth/src/admin-user-endpoints.ts — add bindUserToSoleOrganization (+ findRows / genMemberId helpers), an optional find on the AdminUserDataEngine surface, and wire the bind into runAdminCreateUser.
• packages/plugins/plugin-auth/src/admin-user-endpoints.test.ts — cover single-org bind, multi-org no-op, idempotency, bind-failure isolation, and the no-find no-op.
• docs/adr/0093-tenancy-mode-and-membership-lifecycle.md — the strategic proposal above (Status: Proposed).

Testing

• vitest run src/admin-user-endpoints.test.ts → 29 passed (24 existing + 5 new).
• The endpoint wiring (auth-plugin.ts) already passes the full ObjectQL engine through getDataEngine(), which exposes find, so no route change was needed.

Out of scope / follow-up

• The bulk import-users flow (admin-import-users.ts) creates users the same way; per ADR-0093 Phase 2 it gets covered by the reconciler rather than another endpoint-level patch.
• ADR-0093 Phases 1–3 (tenancy service, fail-fast, reconciler, backfill) land as separate PRs once the ADR is accepted.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XbLd2KKeVs39EztDYMQ1YU